Ben Goertzel: Artificial General Intelligence
AI 与机器学习心理与人性技术与编程生物与进化音乐与艺术
📋 章节目录
暂无章节信息
🔑 关键词
agihumanneurallearningintelligenceopencogdoingdonlogicwholedatabrainputsophiasuperdeepsystemsinterestinggeneralgot
💬 精彩语录
暂无语录
🎙️ 完整对话（5325 条）
Lex Fridman (00:00.000)
The following is a conversation with Ben Goertzel,
以下是与 Ben Goertzel 的对话，
Lex Fridman (00:03.000)
one of the most interesting minds
最有趣的想法之一
Lex Fridman (00:04.560)
in the artificial intelligence community.
在人工智能界。
Lex Fridman (00:06.680)
He's the founder of SingularityNet,
他是SingularityNet的创始人，
Lex Fridman (00:08.920)
designer of OpenCog AI Framework,
OpenCog人工智能框架的设计者，
Ben Goertzel (00:11.520)
formerly a director of research
曾任研究总监
Lex Fridman (00:13.220)
at the Machine Intelligence Research Institute,
在机器智能研究所，
Lex Fridman (00:15.720)
and chief scientist of Hanson Robotics,
兼汉森机器人公司首席科学家，
Lex Fridman (00:18.440)
the company that created the Sophia robot.
创造索菲亚机器人的公司。
Ben Goertzel (00:21.000)
He has been a central figure in the AGI community
他一直是 AGI 社区的核心人物
Lex Fridman (00:23.680)
for many years, including in his organizing
多年来，包括他的组织
Lex Fridman (00:26.960)
and contributing to the conference
并对会议做出贡献
Lex Fridman (00:28.720)
on artificial general intelligence,
关于通用人工智能，
Ben Goertzel (00:30.920)
the 2020 version of which is actually happening this week,
2020 年版本实际上将于本周发生，
Lex Fridman (00:34.440)
Wednesday, Thursday, and Friday.
周三、周四和周五。
Ben Goertzel (00:36.480)
It's virtual and free.
它是虚拟的且免费的。
Lex Fridman (00:38.480)
I encourage you to check out the talks,
我鼓励你看看这些演讲，
Ben Goertzel (00:40.040)
including by Yosha Bach from episode 101 of this podcast.
包括本播客第 101 集中 Yosha Bach 的作品。
Lex Fridman (00:45.160)
Quick summary of the ads.
广告的快速摘要。
Ben Goertzel (00:46.600)
Two sponsors, The Jordan Harbinger Show and Masterclass.
两个赞助商：乔丹先驱秀和大师班。
Lex Fridman (00:51.040)
Please consider supporting this podcast
Ben Goertzel (00:52.800)
by going to jordanharbinger.com slash lex
Lex Fridman (00:56.500)
and signing up at masterclass.com slash lex.
Ben Goertzel (01:00.380)
Click the links, buy all the stuff.
Lex Fridman (01:02.840)
It's the best way to support this podcast
Lex Fridman (01:04.640)
and the journey I'm on in my research and startup.
Lex Fridman (01:08.840)
This is the Artificial Intelligence Podcast.
Ben Goertzel (01:11.480)
If you enjoy it, subscribe on YouTube,
Lex Fridman (01:13.680)
review it with five stars on Apple Podcast,
Ben Goertzel (01:15.940)
support it on Patreon, or connect with me on Twitter
Lex Fridman (01:18.920)
at lexfriedman, spelled without the E, just F R I D M A N.
Ben Goertzel (01:23.920)
As usual, I'll do a few minutes of ads now
Lex Fridman (01:25.980)
and never any ads in the middle
Ben Goertzel (01:27.340)
that can break the flow of the conversation.
Lex Fridman (01:29.980)
This episode is supported by The Jordan Harbinger Show.
Ben Goertzel (01:33.340)
Go to jordanharbinger.com slash lex.
Lex Fridman (01:35.900)
It's how he knows I sent you.
Ben Goertzel (01:37.740)
On that page, there's links to subscribe to it
Lex Fridman (01:40.140)
on Apple Podcast, Spotify, and everywhere else.
Ben Goertzel (01:43.300)
I've been binging on his podcast.
Lex Fridman (01:45.100)
Jordan is great.
Ben Goertzel (01:46.220)
He gets the best out of his guests,
Lex Fridman (01:47.900)
dives deep, calls them out when it's needed,
Lex Fridman (01:50.100)
and makes the whole thing fun to listen to.
Lex Fridman (01:52.260)
He's interviewed Kobe Bryant, Mark Cuban,
Ben Goertzel (01:55.340)
Neil deGrasse Tyson, Keira Kasparov, and many more.
Lex Fridman (01:59.060)
His conversation with Kobe is a reminder
Lex Fridman (02:01.540)
how much focus and hard work is required for greatness
Lex Fridman (02:06.060)
in sport, business, and life.
Ben Goertzel (02:09.540)
I highly recommend the episode if you want to be inspired.
Lex Fridman (02:12.420)
Again, go to jordanharbinger.com slash lex.
Ben Goertzel (02:15.940)
It's how Jordan knows I sent you.
Lex Fridman (02:18.900)
This show is sponsored by Master Class.
Ben Goertzel (02:21.300)
Sign up at masterclass.com slash lex
Lex Fridman (02:24.300)
to get a discount and to support this podcast.
Ben Goertzel (02:27.660)
When I first heard about Master Class,
Lex Fridman (02:29.220)
I thought it was too good to be true.
Ben Goertzel (02:31.060)
For 180 bucks a year, you get an all access pass
Lex Fridman (02:34.220)
to watch courses from to list some of my favorites.
Ben Goertzel (02:37.540)
Chris Hadfield on Space Exploration,
Lex Fridman (02:39.780)
Neil deGrasse Tyson on Scientific Thinking
Lex Fridman (02:41.780)
and Communication, Will Wright, creator of
Lex Fridman (02:44.980)
the greatest city building game ever, Sim City,
Lex Fridman (02:47.940)
and Sims on Space Exploration.
Lex Fridman (02:50.700)
Ben Sims on Game Design, Carlos Santana on Guitar,
Ben Goertzel (02:54.860)
Keira Kasparov, the greatest chess player ever on chess,
Lex Fridman (02:59.460)
Daniel Negrano on Poker, and many more.
Ben Goertzel (03:01.980)
Chris Hadfield explaining how rockets work
Lex Fridman (03:04.660)
and the experience of being launched into space alone
Ben Goertzel (03:07.300)
is worth the money.
Lex Fridman (03:08.700)
Once again, sign up at masterclass.com slash lex
Ben Goertzel (03:12.020)
to get a discount and to support this podcast.
Lex Fridman (03:15.820)
Now, here's my conversation with Ben Kurtzell.
Lex Fridman (03:20.780)
What books, authors, ideas had a lot of impact on you
Lex Fridman (03:25.100)
in your life in the early days?
Ben Goertzel (03:27.900)
You know, what got me into AI and science fiction
Lex Fridman (03:32.180)
and such in the first place wasn't a book,
Lex Fridman (03:34.580)
but the original Star Trek TV show,
Lex Fridman (03:37.020)
which my dad watched with me like in its first run.
Ben Goertzel (03:39.860)
It would have been 1968, 69 or something,
Lex Fridman (03:42.700)
and that was incredible because every show
Ben Goertzel (03:45.340)
they visited a different alien civilization
Lex Fridman (03:49.140)
with different culture and weird mechanisms.
Lex Fridman (03:51.300)
But that got me into science fiction,
Lex Fridman (03:55.020)
and there wasn't that much science fiction
Ben Goertzel (03:57.180)
to watch on TV at that stage,
Lex Fridman (03:58.660)
so that got me into reading the whole literature
Ben Goertzel (04:01.420)
of science fiction, you know,
Lex Fridman (04:03.340)
from the beginning of the previous century until that time.
Lex Fridman (04:07.500)
And I mean, there was so many science fiction writers
Lex Fridman (04:10.860)
who were inspirational to me.
Ben Goertzel (04:12.420)
I'd say if I had to pick two,
Lex Fridman (04:14.820)
it would have been Stanisław Lem, the Polish writer.
Ben Goertzel (04:18.820)
Yeah, Solaris, and then he had a bunch
Lex Fridman (04:22.020)
of more obscure writings on superhuman AIs
Ben Goertzel (04:25.780)
that were engineered.
Lex Fridman (04:26.620)
Solaris was sort of a superhuman,
Ben Goertzel (04:28.660)
naturally occurring intelligence.
Lex Fridman (04:31.540)
Then Philip K. Dick, who, you know,
Ben Goertzel (04:34.780)
ultimately my fandom for Philip K. Dick
Lex Fridman (04:37.340)
is one of the things that brought me together
Ben Goertzel (04:39.100)
with David Hansen, my collaborator on robotics projects.
Lex Fridman (04:43.740)
So, you know, Stanisław Lem was very much an intellectual,
Ben Goertzel (04:47.620)
right, so he had a very broad view of intelligence
Lex Fridman (04:51.020)
going beyond the human and into what I would call,
Ben Goertzel (04:54.420)
you know, open ended superintelligence.
Lex Fridman (04:56.900)
The Solaris superintelligent ocean was intelligent,
Ben Goertzel (05:01.900)
in some ways more generally intelligent than people,
Lex Fridman (05:04.420)
but in a complex and confusing way
Lex Fridman (05:07.340)
so that human beings could never quite connect to it,
Lex Fridman (05:10.180)
but it was still probably very, very smart.
Lex Fridman (05:13.260)
And then the Golem 4 supercomputer
Lex Fridman (05:16.620)
in one of Lem's books, this was engineered by people,
Lex Fridman (05:20.420)
but eventually it became very intelligent
Lex Fridman (05:24.420)
in a different direction than humans
Lex Fridman (05:26.020)
and decided that humans were kind of trivial,
Lex Fridman (05:29.260)
not that interesting.
Lex Fridman (05:30.260)
So it put some impenetrable shield around itself,
Lex Fridman (05:35.300)
shut itself off from humanity,
Lex Fridman (05:36.700)
and then issued some philosophical screed
Lex Fridman (05:40.060)
about the pathetic and hopeless nature of humanity
Lex Fridman (05:44.540)
and all human thought, and then disappeared.
Lex Fridman (05:48.380)
Now, Philip K. Dick, he was a bit different.
Lex Fridman (05:51.140)
He was human focused, right?
Lex Fridman (05:52.460)
His main thing was, you know, human compassion
Lex Fridman (05:55.860)
and the human heart and soul are going to be the constant
Lex Fridman (05:59.540)
that will keep us going through whatever aliens we discover
Ben Goertzel (06:03.620)
or telepathy machines or super AIs or whatever it might be.
Lex Fridman (06:08.620)
So he didn't believe in reality,
Ben Goertzel (06:10.660)
like the reality that we see may be a simulation
Lex Fridman (06:13.740)
or a dream or something else we can't even comprehend,
Lex Fridman (06:17.100)
but he believed in love and compassion
Lex Fridman (06:19.100)
as something persistent
Ben Goertzel (06:20.660)
through the various simulated realities.
Lex Fridman (06:22.460)
So those two science fiction writers had a huge impact on me.
Ben Goertzel (06:26.740)
Then a little older than that, I got into Dostoevsky
Lex Fridman (06:30.300)
and Friedrich Nietzsche and Rimbaud
Lex Fridman (06:33.620)
and a bunch of more literary type writing.
Lex Fridman (06:36.900)
Can we talk about some of those things?
Lex Fridman (06:38.620)
So on the Solaris side, Stanislaw Lem,
Lex Fridman (06:43.180)
this kind of idea of there being intelligences out there
Ben Goertzel (06:47.020)
that are different than our own,
Lex Fridman (06:49.540)
do you think there are intelligences maybe all around us
Lex Fridman (06:53.020)
that we're not able to even detect?
Lex Fridman (06:56.420)
So this kind of idea of,
Ben Goertzel (06:58.700)
maybe you can comment also on Stephen Wolfram
Lex Fridman (07:01.580)
thinking that there's computations all around us
Lex Fridman (07:04.340)
and we're just not smart enough to kind of detect
Lex Fridman (07:07.460)
their intelligence or appreciate their intelligence.
Ben Goertzel (07:10.380)
Yeah, so my friend Hugo de Gares,
Lex Fridman (07:13.540)
who I've been talking to about these things
Ben Goertzel (07:15.780)
for many decades, since the early 90s,
Lex Fridman (07:19.300)
he had an idea he called SIPI,
Ben Goertzel (07:21.740)
the Search for Intraparticulate Intelligence.
Lex Fridman (07:25.100)
So the concept there was as AIs get smarter
Lex Fridman (07:28.140)
and smarter and smarter,
Lex Fridman (07:30.820)
assuming the laws of physics as we know them now
Ben Goertzel (07:33.660)
are still what these super intelligences
Lex Fridman (07:37.420)
perceived to hold and are bound by,
Ben Goertzel (07:39.220)
as they get smarter and smarter,
Lex Fridman (07:40.420)
they're gonna shrink themselves littler and littler
Ben Goertzel (07:42.980)
because special relativity makes it
Lex Fridman (07:45.380)
so they can communicate
Ben Goertzel (07:47.220)
between two spatially distant points.
Lex Fridman (07:49.300)
So they're gonna get smaller and smaller,
Lex Fridman (07:50.780)
but then ultimately, what does that mean?
Lex Fridman (07:53.220)
The minds of the super, super, super intelligences,
Ben Goertzel (07:56.500)
they're gonna be packed into the interaction
Lex Fridman (07:59.020)
of elementary particles or quarks
Ben Goertzel (08:01.940)
or the partons inside quarks or whatever it is.
Lex Fridman (08:04.580)
So what we perceive as random fluctuations
Ben Goertzel (08:07.620)
on the quantum or sub quantum level
Lex Fridman (08:09.740)
may actually be the thoughts
Ben Goertzel (08:11.500)
of the micro, micro, micro miniaturized super intelligences
Lex Fridman (08:16.300)
because there's no way we can tell random
Ben Goertzel (08:19.140)
from structured but within algorithmic information
Lex Fridman (08:21.620)
more complex than our brains, right?
Ben Goertzel (08:23.100)
We can't tell the difference.
Lex Fridman (08:24.300)
So what we think is random could be the thought processes
Ben Goertzel (08:27.020)
of some really tiny super minds.
Lex Fridman (08:29.980)
And if so, there is not a damn thing we can do about it,
Ben Goertzel (08:34.020)
except try to upgrade our intelligences
Lex Fridman (08:37.180)
and expand our minds so that we can perceive
Ben Goertzel (08:40.060)
more of what's around us.
Lex Fridman (08:41.300)
But if those random fluctuations,
Ben Goertzel (08:43.980)
like even if we go to like quantum mechanics,
Lex Fridman (08:46.540)
if that's actually super intelligent systems,
Lex Fridman (08:51.220)
aren't we then part of the super of super intelligence?
Lex Fridman (08:54.620)
Aren't we just like a finger of the entirety
Lex Fridman (08:58.340)
of the body of the super intelligent system?
Lex Fridman (09:01.300)
It could be, I mean, a finger is a strange metaphor.
Ben Goertzel (09:05.940)
I mean, we...
Lex Fridman (09:08.060)
A finger is dumb is what I mean.
Lex Fridman (09:10.700)
But the finger is also useful
Lex Fridman (09:12.260)
and is controlled with intent by the brain
Lex Fridman (09:14.780)
whereas we may be much less than that, right?
Lex Fridman (09:16.700)
I mean, yeah, we may be just some random epiphenomenon
Ben Goertzel (09:21.340)
that they don't care about too much.
Lex Fridman (09:23.300)
Like think about the shape of the crowd emanating
Lex Fridman (09:26.380)
from a sports stadium or something, right?
Lex Fridman (09:28.260)
There's some emergent shape to the crowd, it's there.
Ben Goertzel (09:31.580)
You could take a picture of it, it's kind of cool.
Lex Fridman (09:33.700)
It's irrelevant to the main point of the sports event
Ben Goertzel (09:36.300)
or where the people are going
Lex Fridman (09:37.860)
or what's on the minds of the people
Lex Fridman (09:40.220)
making that shape in the crowd, right?
Lex Fridman (09:41.860)
So we may just be some semi arbitrary higher level pattern
Ben Goertzel (09:47.660)
popping out of a lower level
Lex Fridman (09:49.700)
hyper intelligent self organization.
Lex Fridman (09:52.260)
And I mean, so be it, right?
Lex Fridman (09:55.860)
I mean, that's one thing that...
Ben Goertzel (09:57.060)
Yeah, I mean, the older I've gotten,
Lex Fridman (09:59.500)
the more respect I've achieved for our fundamental ignorance.
Ben Goertzel (10:04.220)
I mean, mine and everybody else's.
Lex Fridman (10:06.260)
I mean, I look at my two dogs,
Ben Goertzel (10:08.820)
two beautiful little toy poodles
Lex Fridman (10:10.940)
and they watch me sitting at the computer typing.
Ben Goertzel (10:14.780)
They just think I'm sitting there wiggling my fingers
Lex Fridman (10:16.980)
to exercise them maybe or guarding the monitor on the desk
Ben Goertzel (10:19.980)
that they have no idea that I'm communicating
Lex Fridman (10:22.340)
with other people halfway around the world,
Ben Goertzel (10:24.420)
let alone creating complex algorithms
Lex Fridman (10:27.660)
running in RAM on some computer server
Lex Fridman (10:30.220)
in St. Petersburg or something, right?
Lex Fridman (10:32.540)
Although they're right there in the room with me.
Lex Fridman (10:35.100)
So what things are there right around us
Lex Fridman (10:37.780)
that we're just too stupid or close minded to comprehend?
Ben Goertzel (10:40.780)
Probably quite a lot.
Lex Fridman (10:42.140)
Your very poodle could also be communicating
Ben Goertzel (10:46.220)
across multiple dimensions with other beings
Lex Fridman (10:49.980)
and you're too unintelligent to understand
Ben Goertzel (10:53.180)
the kind of communication mechanism they're going through.
Lex Fridman (10:55.700)
There have been various TV shows and science fiction novels,
Ben Goertzel (10:59.820)
poisoning cats, dolphins, mice and whatnot
Lex Fridman (11:03.220)
are actually super intelligences here to observe that.
Ben Goertzel (11:07.220)
I would guess as one or the other quantum physics founders
Lex Fridman (11:12.580)
said, those theories are not crazy enough to be true.
Ben Goertzel (11:15.500)
The reality is probably crazier than that.
Lex Fridman (11:17.660)
Beautifully put.
Lex Fridman (11:18.500)
So on the human side, with Philip K. Dick
Lex Fridman (11:22.020)
and in general, where do you fall on this idea
Ben Goertzel (11:27.260)
that love and just the basic spirit of human nature
Lex Fridman (11:30.580)
persists throughout these multiple realities?
Ben Goertzel (11:34.980)
Are you on the side, like the thing that inspires you
Lex Fridman (11:38.420)
about artificial intelligence,
Ben Goertzel (11:40.980)
is it the human side of somehow persisting
Lex Fridman (11:46.740)
through all of the different systems we engineer
Ben Goertzel (11:49.820)
or is AI inspire you to create something
Lex Fridman (11:53.340)
that's greater than human, that's beyond human,
Lex Fridman (11:55.500)
that's almost nonhuman?
Lex Fridman (11:59.140)
I would say my motivation to create AGI
Ben Goertzel (12:02.820)
comes from both of those directions actually.
Lex Fridman (12:05.220)
So when I first became passionate about AGI
Ben Goertzel (12:08.620)
when I was, it would have been two or three years old
Lex Fridman (12:11.420)
after watching robots on Star Trek.
Ben Goertzel (12:14.700)
I mean, then it was really a combination
Lex Fridman (12:18.180)
of intellectual curiosity, like can a machine really think,
Lex Fridman (12:21.460)
how would you do that?
Lex Fridman (12:22.860)
And yeah, just ambition to create something much better
Ben Goertzel (12:27.180)
than all the clearly limited
Lex Fridman (12:28.660)
and fundamentally defective humans I saw around me.
Ben Goertzel (12:31.900)
Then as I got older and got more enmeshed
Lex Fridman (12:35.340)
in the human world and got married, had children,
Ben Goertzel (12:38.780)
saw my parents begin to age, I started to realize,
Lex Fridman (12:41.900)
well, not only will AGI let you go far beyond
Ben Goertzel (12:45.300)
the limitations of the human,
Lex Fridman (12:46.860)
but it could also stop us from dying and suffering
Lex Fridman (12:50.860)
and feeling pain and tormenting ourselves mentally.
Lex Fridman (12:54.980)
So you can see AGI has amazing capability
Ben Goertzel (12:58.060)
to do good for humans, as humans,
Lex Fridman (13:01.380)
alongside with its capability
Ben Goertzel (13:03.420)
to go far, far beyond the human level.
Lex Fridman (13:06.620)
So I mean, both aspects are there,
Ben Goertzel (13:09.980)
which makes it even more exciting and important.
Lex Fridman (13:13.220)
So you mentioned Dostoevsky and Nietzsche.
Lex Fridman (13:15.500)
Where did you pick up from those guys?
Lex Fridman (13:17.060)
I mean.
Ben Goertzel (13:18.980)
That would probably go beyond the scope
Lex Fridman (13:21.500)
of a brief interview, certainly.
Ben Goertzel (13:24.340)
I mean, both of those are amazing thinkers
Lex Fridman (13:26.780)
who one, will necessarily have
Lex Fridman (13:29.020)
a complex relationship with, right?
Lex Fridman (13:32.060)
So, I mean, Dostoevsky on the minus side,
Ben Goertzel (13:36.460)
he's kind of a religious fanatic
Lex Fridman (13:38.460)
and he sort of helped squash the Russian nihilist movement,
Ben Goertzel (13:42.020)
which was very interesting.
Lex Fridman (13:43.140)
Because what nihilism meant originally
Ben Goertzel (13:45.820)
in that period of the mid, late 1800s in Russia
Lex Fridman (13:48.660)
was not taking anything fully 100% for granted.
Ben Goertzel (13:52.180)
It was really more like what we'd call Bayesianism now,
Lex Fridman (13:54.420)
where you don't wanna adopt anything
Ben Goertzel (13:56.900)
as a dogmatic certitude and always leave your mind open.
Lex Fridman (14:01.060)
And how Dostoevsky parodied nihilism
Lex Fridman (14:04.420)
was a bit different, right?
Lex Fridman (14:06.660)
He parodied as people who believe absolutely nothing.
Lex Fridman (14:10.340)
So they must assign an equal probability weight
Lex Fridman (14:13.020)
to every proposition, which doesn't really work.
Lex Fridman (14:17.780)
So on the one hand, I didn't really agree with Dostoevsky
Lex Fridman (14:22.540)
on his sort of religious point of view.
Ben Goertzel (14:26.140)
On the other hand, if you look at his understanding
Lex Fridman (14:29.660)
of human nature and sort of the human mind
Lex Fridman (14:32.660)
and heart and soul, it's really unparalleled.
Lex Fridman (14:37.100)
He had an amazing view of how human beings construct a world
Ben Goertzel (14:42.100)
for themselves based on their own understanding
Lex Fridman (14:45.380)
and their own mental predisposition.
Lex Fridman (14:47.500)
And I think if you look in the brothers Karamazov
Lex Fridman (14:50.100)
in particular, the Russian literary theorist Mikhail Bakhtin
Ben Goertzel (14:56.140)
wrote about this as a polyphonic mode of fiction,
Lex Fridman (14:59.580)
which means it's not third person,
Lex Fridman (15:02.300)
but it's not first person from any one person really.
Lex Fridman (15:05.020)
There are many different characters in the novel
Lex Fridman (15:07.020)
and each of them is sort of telling part of the story
Lex Fridman (15:10.020)
from their own point of view.
Lex Fridman (15:11.580)
So the reality of the whole story is an intersection
Lex Fridman (15:15.900)
like synergetically of the many different characters
Ben Goertzel (15:19.020)
world views.
Lex Fridman (15:19.860)
And that really, it's a beautiful metaphor
Lex Fridman (15:23.220)
and even a reflection I think of how all of us
Lex Fridman (15:26.100)
socially create our reality.
Ben Goertzel (15:27.700)
Like each of us sees the world in a certain way.
Lex Fridman (15:31.060)
Each of us in a sense is making the world as we see it
Ben Goertzel (15:34.780)
based on our own minds and understanding,
Lex Fridman (15:37.620)
but it's polyphony like in music
Ben Goertzel (15:40.980)
where multiple instruments are coming together
Lex Fridman (15:43.300)
to create the sound.
Ben Goertzel (15:44.620)
The ultimate reality that's created
Lex Fridman (15:46.700)
comes out of each of our subjective understandings,
Ben Goertzel (15:50.220)
intersecting with each other.
Lex Fridman (15:51.340)
And that was one of the many beautiful things in Dostoevsky.
Lex Fridman (15:55.660)
So maybe a little bit to mention,
Lex Fridman (15:57.980)
you have a connection to Russia and the Soviet culture.
Ben Goertzel (16:02.260)
I mean, I'm not sure exactly what the nature
Lex Fridman (16:03.860)
of the connection is, but at least the spirit
Ben Goertzel (16:06.180)
of your thinking is in there.
Lex Fridman (16:07.380)
Well, my ancestry is three quarters Eastern European Jewish.
Lex Fridman (16:12.740)
So I mean, my three of my great grandparents
Lex Fridman (16:16.740)
emigrated to New York from Lithuania
Lex Fridman (16:20.340)
and sort of border regions of Poland,
Lex Fridman (16:23.060)
which are in and out of Poland
Ben Goertzel (16:24.980)
in around the time of World War I.
Lex Fridman (16:28.020)
And they were socialists and communists as well as Jews,
Ben Goertzel (16:33.700)
mostly Menshevik, not Bolshevik.
Lex Fridman (16:35.940)
And they sort of, they fled at just the right time
Ben Goertzel (16:39.260)
to the US for their own personal reasons.
Lex Fridman (16:41.260)
And then almost all, or maybe all of my extended family
Ben Goertzel (16:45.580)
that remained in Eastern Europe was killed
Lex Fridman (16:47.220)
either by Hitlands or Stalin's minions at some point.
Lex Fridman (16:50.380)
So the branch of the family that emigrated to the US
Lex Fridman (16:53.580)
was pretty much the only one.
Lex Fridman (16:56.740)
So how much of the spirit of the people
Lex Fridman (16:58.700)
is in your blood still?
Ben Goertzel (16:59.900)
Like, when you look in the mirror, do you see,
Lex Fridman (17:03.900)
what do you see?
Ben Goertzel (17:04.860)
Meat, I see a bag of meat that I want to transcend
Lex Fridman (17:08.460)
by uploading into some sort of superior reality.
Lex Fridman (17:12.180)
But very, I mean, yeah, very clearly,
Lex Fridman (17:18.340)
I mean, I'm not religious in a traditional sense,
Lex Fridman (17:22.260)
but clearly the Eastern European Jewish tradition
Lex Fridman (17:27.260)
was what I was raised in.
Ben Goertzel (17:28.780)
I mean, there was, my grandfather, Leo Zwell,
Lex Fridman (17:32.700)
was a physical chemist who worked with Linus Pauling
Lex Fridman (17:35.380)
and a bunch of the other early greats in quantum mechanics.
Lex Fridman (17:38.100)
I mean, he was into X ray diffraction.
Ben Goertzel (17:41.220)
He was on the material science side,
Lex Fridman (17:42.940)
an experimentalist rather than a theorist.
Ben Goertzel (17:45.420)
His sister was also a physicist.
Lex Fridman (17:47.700)
And my father's father, Victor Gertzel,
Ben Goertzel (17:51.100)
was a PhD in psychology who had the unenviable job
Lex Fridman (17:57.100)
of giving Soka therapy to the Japanese
Ben Goertzel (17:59.260)
in internment camps in the US in World War II,
Lex Fridman (18:03.100)
like to counsel them why they shouldn't kill themselves,
Ben Goertzel (18:05.820)
even though they'd had all their stuff taken away
Lex Fridman (18:08.420)
and been imprisoned for no good reason.
Ben Goertzel (18:10.300)
So, I mean, yeah, there's a lot of Eastern European
Lex Fridman (18:15.780)
Jewishness in my background.
Ben Goertzel (18:18.060)
One of my great uncles was, I guess,
Lex Fridman (18:20.180)
conductor of San Francisco Orchestra.
Lex Fridman (18:22.420)
So there's a lot of Mickey Salkind,
Lex Fridman (18:25.620)
bunch of music in there also.
Lex Fridman (18:27.660)
And clearly this culture was all about learning
Lex Fridman (18:31.540)
and understanding the world,
Lex Fridman (18:34.860)
and also not quite taking yourself too seriously
Lex Fridman (18:38.820)
while you do it, right?
Ben Goertzel (18:39.900)
There's a lot of Yiddish humor in there.
Lex Fridman (18:42.060)
So I do appreciate that culture,
Ben Goertzel (18:45.220)
although the whole idea that like the Jews
Lex Fridman (18:47.580)
are the chosen people of God
Ben Goertzel (18:49.020)
never resonated with me too much.
Lex Fridman (18:51.740)
The graph of the Gertzel family,
Ben Goertzel (18:55.100)
I mean, just the people I've encountered
Lex Fridman (18:56.940)
just doing some research and just knowing your work
Ben Goertzel (18:59.540)
through the decades, it's kind of fascinating.
Lex Fridman (19:03.580)
Just the number of PhDs.
Ben Goertzel (19:06.380)
Yeah, yeah, I mean, my dad is a sociology professor
Lex Fridman (19:10.740)
who recently retired from Rutgers University,
Lex Fridman (19:15.060)
but clearly that gave me a head start in life.
Lex Fridman (19:18.540)
I mean, my grandfather gave me
Ben Goertzel (19:20.260)
all those quantum mechanics books
Lex Fridman (19:21.620)
when I was like seven or eight years old.
Ben Goertzel (19:24.220)
I remember going through them,
Lex Fridman (19:26.060)
and it was all the old quantum mechanics
Ben Goertzel (19:28.020)
like Rutherford Adams and stuff.
Lex Fridman (19:30.420)
So I got to the part of wave functions,
Ben Goertzel (19:32.860)
which I didn't understand, although I was very bright kid.
Lex Fridman (19:36.140)
And I realized he didn't quite understand it either,
Lex Fridman (19:38.660)
but at least like he pointed me to some professor
Lex Fridman (19:41.980)
he knew at UPenn nearby who understood these things, right?
Lex Fridman (19:45.340)
So that's an unusual opportunity for a kid to have, right?
Lex Fridman (19:49.620)
My dad, he was programming Fortran
Ben Goertzel (19:52.380)
when I was 10 or 11 years old
Lex Fridman (19:53.900)
on like HP 3000 mainframes at Rutgers University.
Lex Fridman (19:57.660)
So I got to do linear regression in Fortran
Lex Fridman (1:00:00.160)
Automated.
Lex Fridman (1:00:01.000)
Learning, right?
Lex Fridman (1:00:01.840)
It should be learning from experience.
Lex Fridman (1:00:03.800)
And the AI field then was not interested
Lex Fridman (1:00:06.120)
in learning from experience.
Ben Goertzel (1:00:08.320)
I mean, some researchers certainly were.
Lex Fridman (1:00:11.020)
I mean, I remember in mid eighties,
Ben Goertzel (1:00:13.960)
I discovered a book by John Andreas,
Lex Fridman (1:00:17.160)
which was, it was about a reinforcement learning system
Ben Goertzel (1:00:21.920)
called PURRDASHPUSS, which was an acronym
Lex Fridman (1:00:27.080)
that I can't even remember what it was for,
Lex Fridman (1:00:28.640)
but purpose anyway.
Lex Fridman (1:00:30.400)
But he, I mean, that was a system
Ben Goertzel (1:00:32.000)
that was supposed to be an AGI
Lex Fridman (1:00:34.360)
and basically by some sort of fancy
Ben Goertzel (1:00:38.120)
like Markov decision process learning,
Lex Fridman (1:00:41.000)
it was supposed to learn everything
Ben Goertzel (1:00:43.440)
just from the bits coming into it
Lex Fridman (1:00:44.880)
and learn to maximize its reward
Lex Fridman (1:00:46.720)
and become intelligent, right?
Lex Fridman (1:00:49.080)
So that was there in academia back then,
Lex Fridman (1:00:51.800)
but it was like isolated, scattered, weird people.
Lex Fridman (1:00:55.240)
But all these isolated, scattered, weird people
Ben Goertzel (1:00:57.440)
in that period, I mean, they laid the intellectual grounds
Lex Fridman (1:01:01.280)
for what happened later.
Lex Fridman (1:01:02.120)
So you look at John Andreas at University of Canterbury
Lex Fridman (1:01:05.300)
with his PURRDASHPUSS reinforcement learning Markov system.
Ben Goertzel (1:01:09.720)
He was the PhD supervisor for John Cleary in New Zealand.
Lex Fridman (1:01:14.080)
Now, John Cleary worked with me
Ben Goertzel (1:01:17.080)
when I was at Waikato University in 1993 in New Zealand.
Lex Fridman (1:01:21.680)
And he worked with Ian Whitten there
Lex Fridman (1:01:23.900)
and they launched WEKA,
Lex Fridman (1:01:25.940)
which was the first open source machine learning toolkit,
Ben Goertzel (1:01:29.840)
which was launched in, I guess, 93 or 94
Lex Fridman (1:01:33.520)
when I was at Waikato University.
Ben Goertzel (1:01:35.160)
Written in Java, unfortunately.
Lex Fridman (1:01:36.480)
Written in Java, which was a cool language back then.
Ben Goertzel (1:01:39.620)
I guess it's still, well, it's not cool anymore,
Lex Fridman (1:01:41.720)
but it's powerful.
Ben Goertzel (1:01:43.280)
I find, like most programmers now,
Lex Fridman (1:01:45.760)
I find Java unnecessarily bloated,
Lex Fridman (1:01:48.820)
but back then it was like Java or C++ basically.
Lex Fridman (1:01:52.020)
And Java was easier for students.
Ben Goertzel (1:01:55.760)
Amusingly, a lot of the work on WEKA
Lex Fridman (1:01:57.760)
when we were in New Zealand was funded by a US,
Ben Goertzel (1:02:01.200)
sorry, a New Zealand government grant
Lex Fridman (1:02:03.880)
to use machine learning
Ben Goertzel (1:02:05.440)
to predict the menstrual cycles of cows.
Lex Fridman (1:02:08.240)
So in the US, all the grant funding for AI
Ben Goertzel (1:02:10.440)
was about how to kill people or spy on people.
Lex Fridman (1:02:13.600)
In New Zealand, it's all about cows or kiwi fruits, right?
Ben Goertzel (1:02:16.400)
Yeah.
Lex Fridman (1:02:17.560)
So yeah, anyway, I mean, John Andreas
Ben Goertzel (1:02:20.560)
had his probability theory based reinforcement learning,
Lex Fridman (1:02:24.320)
proto AGI.
Ben Goertzel (1:02:25.780)
John Cleary was trying to do much more ambitious,
Lex Fridman (1:02:29.400)
probabilistic AGI systems.
Ben Goertzel (1:02:31.820)
Now, John Cleary helped do WEKA,
Lex Fridman (1:02:36.160)
which is the first open source machine learning toolkit.
Lex Fridman (1:02:39.360)
So the predecessor for TensorFlow and Torch
Lex Fridman (1:02:41.520)
and all these things.
Ben Goertzel (1:02:43.040)
Also, Shane Legg was at Waikato
Lex Fridman (1:02:46.800)
working with John Cleary and Ian Witten
Lex Fridman (1:02:50.240)
and this whole group.
Lex Fridman (1:02:51.500)
And then working with my own companies,
Ben Goertzel (1:02:55.800)
my company, WebMind, an AI company I had in the late 90s
Lex Fridman (1:02:59.840)
with a team there at Waikato University,
Ben Goertzel (1:03:02.320)
which is how Shane got his head full of AGI,
Lex Fridman (1:03:05.360)
which led him to go on
Lex Fridman (1:03:06.440)
and with Demis Hassabis found DeepMind.
Lex Fridman (1:03:08.660)
So what you can see through that lineage is,
Ben Goertzel (1:03:11.060)
you know, in the 80s and 70s,
Lex Fridman (1:03:12.580)
John Andreas was trying to build probabilistic
Ben Goertzel (1:03:14.800)
reinforcement learning AGI systems.
Lex Fridman (1:03:17.200)
The technology, the computers just weren't there to support
Ben Goertzel (1:03:19.680)
his ideas were very similar to what people are doing now.
Lex Fridman (1:03:23.920)
But, you know, although he's long since passed away
Lex Fridman (1:03:27.720)
and didn't become that famous outside of Canterbury,
Lex Fridman (1:03:30.940)
I mean, the lineage of ideas passed on from him
Ben Goertzel (1:03:33.720)
to his students, to their students,
Lex Fridman (1:03:35.140)
you can go trace directly from there to me
Lex Fridman (1:03:37.920)
and to DeepMind, right?
Lex Fridman (1:03:39.480)
So that there was a lot going on in AGI
Ben Goertzel (1:03:42.180)
that did ultimately lay the groundwork
Lex Fridman (1:03:46.460)
for what we have today, but there wasn't a community, right?
Lex Fridman (1:03:48.560)
And so when I started trying to pull together
Lex Fridman (1:03:53.520)
an AGI community, it was in the, I guess,
Ben Goertzel (1:03:56.920)
the early aughts when I was living in Washington, D.C.
Lex Fridman (1:04:00.400)
and making a living doing AI consulting
Ben Goertzel (1:04:03.440)
for various U.S. government agencies.
Lex Fridman (1:04:07.080)
And I organized the first AGI workshop in 2006.
Lex Fridman (1:04:13.200)
And I mean, it wasn't like it was literally
Lex Fridman (1:04:15.780)
in my basement or something.
Ben Goertzel (1:04:17.000)
I mean, it was in the conference room at the Marriott
Lex Fridman (1:04:19.320)
in Bethesda, it's not that edgy or underground,
Ben Goertzel (1:04:23.200)
unfortunately, but still.
Lex Fridman (1:04:25.000)
How many people attended?
Ben Goertzel (1:04:25.840)
About 60 or something.
Lex Fridman (1:04:27.600)
That's not bad.
Ben Goertzel (1:04:28.480)
I mean, D.C. has a lot of AI going on,
Lex Fridman (1:04:30.780)
probably until the last five or 10 years,
Ben Goertzel (1:04:34.200)
much more than Silicon Valley, although it's just quiet
Lex Fridman (1:04:37.800)
because of the nature of what happens in D.C.
Ben Goertzel (1:04:41.280)
Their business isn't driven by PR.
Lex Fridman (1:04:43.600)
Mostly when something starts to work really well,
Lex Fridman (1:04:46.140)
it's taken black and becomes even more quiet, right?
Lex Fridman (1:04:49.640)
But yeah, the thing is that really had the feeling
Ben Goertzel (1:04:52.880)
of a group of starry eyed mavericks huddled in a basement,
Lex Fridman (1:04:58.400)
like plotting how to overthrow the narrow AI establishment.
Lex Fridman (1:05:02.520)
And for the first time, in some cases,
Lex Fridman (1:05:05.760)
coming together with others who shared their passion
Ben Goertzel (1:05:08.680)
for AGI and the technical seriousness about working on it.
Lex Fridman (1:05:13.200)
And that's very, very different than what we have today.
Ben Goertzel (1:05:19.160)
I mean, now it's a little bit different.
Lex Fridman (1:05:22.320)
We have AGI conference every year
Lex Fridman (1:05:24.640)
and there's several hundred people rather than 50.
Lex Fridman (1:05:29.300)
Now it's more like this is the main gathering
Ben Goertzel (1:05:32.760)
of people who want to achieve AGI
Lex Fridman (1:05:35.020)
and think that large scale nonlinear regression
Ben Goertzel (1:05:39.220)
is not the golden path to AGI.
Lex Fridman (1:05:42.480)
So I mean it's...
Ben Goertzel (1:05:43.320)
AKA neural networks.
Lex Fridman (1:05:44.160)
Yeah, yeah, yeah.
Ben Goertzel (1:05:44.980)
Well, certain architectures for learning using neural networks.
Lex Fridman (1:05:51.840)
So yeah, the AGI conferences are sort of now
Ben Goertzel (1:05:54.440)
the main concentration of people not obsessed
Lex Fridman (1:05:57.960)
with deep neural nets and deep reinforcement learning,
Lex Fridman (1:06:00.880)
but still interested in AGI, not the only ones.
Lex Fridman (1:06:06.460)
I mean, there's other little conferences and groupings
Ben Goertzel (1:06:10.200)
interested in human level AI
Lex Fridman (1:06:13.280)
and cognitive architectures and so forth.
Lex Fridman (1:06:16.040)
But yeah, it's been a big shift.
Lex Fridman (1:06:17.880)
Like back then, you couldn't really...
Ben Goertzel (1:06:21.960)
It'll be very, very edgy then
Lex Fridman (1:06:23.540)
to give a university department seminar
Ben Goertzel (1:06:26.220)
that mentioned AGI or human level AI.
Lex Fridman (1:06:28.440)
It was more like you had to talk about
Ben Goertzel (1:06:30.640)
something more short term and immediately practical
Lex Fridman (1:06:34.360)
than in the bar after the seminar,
Ben Goertzel (1:06:36.600)
you could bullshit about AGI in the same breath
Lex Fridman (1:06:39.540)
as time travel or the simulation hypothesis or something.
Ben Goertzel (1:06:44.200)
Whereas now, AGI is not only in the academic seminar room,
Lex Fridman (1:06:48.360)
like you have Vladimir Putin knows what AGI is.
Lex Fridman (1:06:51.960)
And he's like, Russia needs to become the leader in AGI.
Lex Fridman (1:06:55.480)
So national leaders and CEOs of large corporations.
Ben Goertzel (1:07:01.080)
I mean, the CTO of Intel, Justin Ratner,
Lex Fridman (1:07:04.240)
this was years ago, Singularity Summit Conference,
Ben Goertzel (1:07:06.840)
2008 or something.
Lex Fridman (1:07:07.780)
He's like, we believe Ray Kurzweil,
Ben Goertzel (1:07:10.080)
the singularity will happen in 2045
Lex Fridman (1:07:12.000)
and it will have Intel inside.
Ben Goertzel (1:07:13.640)
So, I mean, it's gone from being something
Lex Fridman (1:07:18.840)
which is the pursuit of like crazed mavericks,
Ben Goertzel (1:07:21.700)
crackpots and science fiction fanatics
Lex Fridman (1:07:24.540)
to being a marketing term for large corporations
Lex Fridman (1:07:30.120)
and the national leaders,
Lex Fridman (1:07:31.480)
which is a astounding transition.
Lex Fridman (1:07:35.160)
But yeah, in the course of this transition,
Lex Fridman (1:07:40.160)
I think a bunch of sub communities have formed
Lex Fridman (1:07:42.260)
and the community around the AGI conference series
Lex Fridman (1:07:45.800)
is certainly one of them.
Ben Goertzel (1:07:47.640)
It hasn't grown as big as I might've liked it to.
Lex Fridman (1:07:51.940)
On the other hand, sometimes a modest size community
Ben Goertzel (1:07:56.320)
can be better for making intellectual progress also.
Lex Fridman (1:07:59.080)
Like you go to a society for neuroscience conference,
Ben Goertzel (1:08:02.160)
you have 35 or 40,000 neuroscientists.
Lex Fridman (1:08:05.400)
On the one hand, it's amazing.
Ben Goertzel (1:08:07.480)
On the other hand, you're not gonna talk to the leaders
Lex Fridman (1:08:10.920)
of the field there if you're an outsider.
Ben Goertzel (1:08:14.160)
Yeah, in the same sense, the AAAI,
Lex Fridman (1:08:17.920)
the artificial intelligence,
Ben Goertzel (1:08:20.160)
the main kind of generic artificial intelligence
Lex Fridman (1:08:23.640)
conference is too big.
Ben Goertzel (1:08:26.920)
It's too amorphous.
Lex Fridman (1:08:28.280)
Like it doesn't make sense.
Ben Goertzel (1:08:30.240)
Well, yeah, and NIPS has become a company advertising outlet
Lex Fridman (1:08:35.240)
in the whole of it.
Ben Goertzel (1:08:37.000)
So, I mean, to comment on the role of AGI
Lex Fridman (1:08:40.240)
in the research community, I'd still,
Ben Goertzel (1:08:42.680)
if you look at NeurIPS, if you look at CVPR,
Lex Fridman (1:08:45.200)
if you look at these iClear,
Ben Goertzel (1:08:49.240)
AGI is still seen as the outcast.
Lex Fridman (1:08:51.860)
I would say in these main machine learning,
Ben Goertzel (1:08:55.020)
in these main artificial intelligence conferences
Lex Fridman (1:08:59.040)
amongst the researchers,
Ben Goertzel (1:09:00.880)
I don't know if it's an accepted term yet.
Lex Fridman (1:09:03.880)
What I've seen bravely, you mentioned Shane Legg's
Ben Goertzel (1:09:08.280)
DeepMind and then OpenAI are the two places that are,
Lex Fridman (1:09:13.000)
I would say unapologetically so far,
Ben Goertzel (1:09:15.580)
I think it's actually changing unfortunately,
Lex Fridman (1:09:17.440)
but so far they've been pushing the idea
Ben Goertzel (1:09:19.640)
that the goal is to create an AGI.
Lex Fridman (1:09:22.760)
Well, they have billions of dollars behind them.
Ben Goertzel (1:09:24.360)
So, I mean, they're in the public mind
Lex Fridman (1:09:27.220)
that certainly carries some oomph, right?
Ben Goertzel (1:09:30.120)
I mean, I mean.
Lex Fridman (1:09:30.960)
But they also have really strong researchers, right?
Ben Goertzel (1:09:33.160)
They do, they're great teams.
Lex Fridman (1:09:34.260)
I mean, DeepMind in particular, yeah.
Lex Fridman (1:09:36.660)
And they have, I mean, DeepMind has Marcus Hutter
Lex Fridman (1:09:39.280)
walking around.
Ben Goertzel (1:09:40.120)
I mean, there's all these folks who basically
Lex Fridman (1:09:43.480)
their full time position involves dreaming
Ben Goertzel (1:09:46.400)
about creating AGI.
Lex Fridman (1:09:47.800)
I mean, Google Brain has a lot of amazing
Ben Goertzel (1:09:51.320)
AGI oriented people also.
Lex Fridman (1:09:53.240)
And I mean, so I'd say from a public marketing view,
Ben Goertzel (1:09:59.840)
DeepMind and OpenAI are the two large well funded
Lex Fridman (1:10:03.820)
organizations that have put the term and concept AGI
Ben Goertzel (1:10:08.360)
out there sort of as part of their public image.
Lex Fridman (1:10:12.720)
But I mean, they're certainly not,
Ben Goertzel (1:10:15.200)
there are other groups that are doing research
Lex Fridman (1:10:17.160)
that seems just as AGI is to me.
Ben Goertzel (1:10:20.660)
I mean, including a bunch of groups in Google's
Lex Fridman (1:10:23.320)
main Mountain View office.
Lex Fridman (1:10:26.000)
So yeah, it's true.
Lex Fridman (1:10:27.960)
AGI is somewhat away from the mainstream now.
Lex Fridman (1:10:33.880)
But if you compare it to where it was 15 years ago,
Lex Fridman (1:10:38.040)
there's been an amazing mainstreaming.
Ben Goertzel (1:10:41.960)
You could say the same thing about super longevity research,
Lex Fridman (1:10:45.520)
which is one of my application areas that I'm excited about.
Ben Goertzel (1:10:49.120)
I mean, I've been talking about this since the 90s,
Lex Fridman (1:10:52.880)
but working on this since 2001.
Lex Fridman (1:10:54.560)
And back then, really to say,
Lex Fridman (1:10:57.280)
you're trying to create therapies to allow people
Ben Goertzel (1:10:59.440)
to live hundreds of thousands of years,
Lex Fridman (1:11:02.360)
you were way, way, way, way out of the industry,
Ben Goertzel (1:11:05.520)
academic mainstream.
Lex Fridman (1:11:06.720)
But now, Google had Project Calico,
Ben Goertzel (1:11:11.540)
Craig Venter had Human Longevity Incorporated.
Lex Fridman (1:11:14.080)
And then once the suits come marching in, right?
Ben Goertzel (1:11:17.160)
I mean, once there's big money in it,
Lex Fridman (1:11:20.200)
then people are forced to take it seriously
Ben Goertzel (1:11:22.720)
because that's the way modern society works.
Lex Fridman (1:11:24.880)
So it's still not as mainstream as cancer research,
Ben Goertzel (1:11:28.400)
just as AGI is not as mainstream
Lex Fridman (1:11:31.060)
as automated driving or something.
Lex Fridman (1:11:32.960)
But the degree of mainstreaming that's happened
Lex Fridman (1:11:36.020)
in the last 10 to 15 years is astounding
Ben Goertzel (1:11:40.120)
to those of us who've been at it for a while.
Lex Fridman (1:11:42.080)
Yeah, but there's a marketing aspect to the term,
Lex Fridman (1:11:45.360)
but in terms of actual full force research
Lex Fridman (1:11:48.800)
that's going on under the header of AGI,
Ben Goertzel (1:11:51.280)
it's currently, I would say dominated,
Lex Fridman (1:11:54.280)
maybe you can disagree,
Ben Goertzel (1:11:55.960)
dominated by neural networks research,
Lex Fridman (1:11:57.740)
that the nonlinear regression, as you mentioned.
Ben Goertzel (1:12:02.740)
Like what's your sense with OpenCog, with your work,
Lex Fridman (1:12:06.520)
but in general, I was logic based systems
Lex Fridman (1:12:10.920)
and expert systems.
Lex Fridman (1:12:12.000)
For me, always seemed to capture a deep element
Ben Goertzel (1:12:18.440)
of intelligence that needs to be there.
Lex Fridman (1:12:21.400)
Like you said, it needs to learn,
Ben Goertzel (1:12:23.020)
it needs to be automated somehow,
Lex Fridman (1:12:24.900)
but that seems to be missing from a lot of research currently.
Lex Fridman (1:12:31.360)
So what's your sense?
Lex Fridman (1:12:34.360)
I guess one way to ask this question,
Ben Goertzel (1:12:36.280)
what's your sense of what kind of things
Lex Fridman (1:12:39.200)
will an AGI system need to have?
Ben Goertzel (1:12:43.480)
Yeah, that's a very interesting topic
Lex Fridman (1:12:45.960)
that I've thought about for a long time.
Lex Fridman (1:12:47.900)
And I think there are many, many different approaches
Lex Fridman (1:12:53.840)
that can work for getting to human level AI.
Lex Fridman (1:12:56.920)
So I don't think there's like one golden algorithm,
Lex Fridman (1:13:02.600)
or one golden design that can work.
Lex Fridman (1:13:05.840)
And I mean, flying machines is the much worn
Lex Fridman (1:13:10.720)
analogy here, right?
Ben Goertzel (1:13:11.680)
Like, I mean, you have airplanes, you have helicopters,
Lex Fridman (1:13:13.760)
you have balloons, you have stealth bombers
Ben Goertzel (1:13:17.160)
that don't look like regular airplanes.
Lex Fridman (1:13:18.760)
You've got all blimps.
Ben Goertzel (1:13:21.040)
Birds too.
Lex Fridman (1:13:21.880)
Birds, yeah, and bugs, right?
Ben Goertzel (1:13:24.280)
Yeah.
Lex Fridman (1:13:25.120)
And there are certainly many kinds of flying machines that.
Lex Fridman (1:13:29.920)
And there's a catapult that you can just launch.
Lex Fridman (1:13:32.360)
And there's bicycle powered like flying machines, right?
Ben Goertzel (1:13:36.160)
Nice, yeah.
Lex Fridman (1:13:37.000)
Yeah, so now these are all analyzable
Lex Fridman (1:13:40.920)
by a basic theory of aerodynamics, right?
Lex Fridman (1:13:43.800)
Now, so one issue with AGI is we don't yet have the analog
Ben Goertzel (1:13:48.920)
of the theory of aerodynamics.
Lex Fridman (1:13:50.800)
And that's what Marcus Hutter was trying to make
Ben Goertzel (1:13:54.640)
with the AXI and his general theory of general intelligence.
Lex Fridman (1:13:58.820)
But that theory in its most clearly articulated parts
Ben Goertzel (1:14:03.360)
really only works for either infinitely powerful machines
Lex Fridman (1:14:07.120)
or almost, or insanely impractically powerful machines.
Lex Fridman (1:14:11.840)
So I mean, if you were gonna take a theory based approach
Lex Fridman (1:14:14.880)
to AGI, what you would do is say, well, let's take
Ben Goertzel (1:14:20.040)
what's called say AXE TL, which is Hutter's AXE machine
Lex Fridman (1:14:25.040)
that can work on merely insanely much processing power
Ben Goertzel (1:14:29.000)
rather than infinitely much.
Lex Fridman (1:14:30.200)
What does TL stand for?
Ben Goertzel (1:14:32.240)
Time and length.
Lex Fridman (1:14:33.560)
Okay.
Lex Fridman (1:14:34.400)
So you're basically how it.
Lex Fridman (1:14:35.600)
Like constrained somehow.
Ben Goertzel (1:14:36.480)
Yeah, yeah, yeah.
Lex Fridman (1:14:37.320)
So how AXE works basically is each action
Ben Goertzel (1:14:42.420)
that it wants to take, before taking that action,
Lex Fridman (1:14:45.040)
it looks at all its history.
Lex Fridman (1:14:47.080)
And then it looks at all possible programs
Lex Fridman (1:14:49.880)
that it could use to make a decision.
Lex Fridman (1:14:51.760)
And it decides like which decision program
Lex Fridman (1:14:54.320)
would have let it make the best decisions
Ben Goertzel (1:14:56.120)
according to its reward function over its history.
Lex Fridman (1:14:58.400)
And it uses that decision program
Lex Fridman (1:15:00.000)
to make the next decision, right?
Lex Fridman (1:15:02.080)
It's not afraid of infinite resources.
Ben Goertzel (1:15:04.760)
It's searching through the space
Lex Fridman (1:15:06.360)
of all possible computer programs
Ben Goertzel (1:15:08.440)
in between each action and each next action.
Lex Fridman (1:15:10.720)
Now, AXE TL searches through all possible computer programs
Ben Goertzel (1:15:15.320)
that have runtime less than T and length less than L.
Lex Fridman (1:15:18.160)
So it's, which is still an impractically humongous space,
Lex Fridman (1:15:22.680)
right?
Lex Fridman (1:15:23.520)
So what you would like to do to make an AGI
Lex Fridman (1:15:27.960)
and what will probably be done 50 years from now
Lex Fridman (1:15:29.840)
to make an AGI is say, okay, well, we have some constraints.
Ben Goertzel (1:15:34.840)
We have these processing power constraints
Lex Fridman (1:15:37.480)
and we have the space and time constraints on the program.
Ben Goertzel (1:15:42.700)
We have energy utilization constraints
Lex Fridman (1:15:45.360)
and we have this particular class environments,
Ben Goertzel (1:15:48.160)
class of environments that we care about,
Lex Fridman (1:15:50.320)
which may be say, you know, manipulating physical objects
Ben Goertzel (1:15:54.400)
on the surface of the earth,
Lex Fridman (1:15:55.400)
communicating in human language.
Ben Goertzel (1:15:57.360)
I mean, whatever our particular, not annihilating humanity,
Lex Fridman (1:16:02.240)
whatever our particular requirements happen to be.
Ben Goertzel (1:16:05.440)
If you formalize those requirements
Lex Fridman (1:16:07.280)
in some formal specification language,
Ben Goertzel (1:16:10.300)
you should then be able to run
Lex Fridman (1:16:13.320)
automated program specializer on AXE TL,
Ben Goertzel (1:16:17.040)
specialize it to the computing resource constraints
Lex Fridman (1:16:21.400)
and the particular environment and goal.
Lex Fridman (1:16:23.600)
And then it will spit out like the specialized version
Lex Fridman (1:16:27.600)
of AXE TL to your resource restrictions
Lex Fridman (1:16:30.620)
and your environment, which will be your AGI, right?
Lex Fridman (1:16:32.700)
And that I think is how our super AGI
Lex Fridman (1:16:36.160)
will create new AGI systems, right?
Lex Fridman (1:16:38.560)
But that's a very rush.
Ben Goertzel (1:16:40.600)
It seems really inefficient.
Lex Fridman (1:16:41.600)
It's a very Russian approach by the way,
Ben Goertzel (1:16:43.160)
like the whole field of program specialization
Lex Fridman (1:16:45.240)
came out of Russia.
Lex Fridman (1:16:47.280)
Can you backtrack?
Lex Fridman (1:16:48.120)
So what is program specialization?
Lex Fridman (1:16:49.680)
So it's basically...
Lex Fridman (1:16:51.120)
Well, take sorting, for example.
Ben Goertzel (1:16:53.640)
You can have a generic program for sorting lists,
Lex Fridman (1:16:56.640)
but what if all your lists you care about
Lex Fridman (1:16:58.280)
are length 10,000 or less?
Lex Fridman (1:16:59.920)
Got it.
Ben Goertzel (1:17:00.760)
You can run an automated program specializer
Lex Fridman (1:17:02.560)
on your sorting algorithm,
Lex Fridman (1:17:04.080)
and it will come up with the algorithm
Lex Fridman (1:17:05.400)
that's optimal for sorting lists of length 1,000 or less,
Lex Fridman (1:17:08.400)
or 10,000 or less, right?
Lex Fridman (1:17:09.800)
That's kind of like, isn't that the kind of the process
Lex Fridman (1:17:12.200)
of evolution as a program specializer to the environment?
Lex Fridman (1:17:17.440)
So you're kind of evolving human beings,
Ben Goertzel (1:17:20.000)
or you're living creatures.
Lex Fridman (1:17:21.840)
Your Russian heritage is showing there.
Lex Fridman (1:17:24.320)
So with Alexander Vityaev and Peter Anokhin and so on,
Lex Fridman (1:17:28.480)
I mean, there's a long history
Lex Fridman (1:17:31.800)
of thinking about evolution that way also, right?
Lex Fridman (1:17:36.760)
So, well, my point is that what we're thinking of
Ben Goertzel (1:17:40.120)
as a human level general intelligence,
Lex Fridman (1:17:44.160)
if you start from narrow AIs,
Ben Goertzel (1:17:46.680)
like are being used in the commercial AI field now,
Lex Fridman (1:17:50.320)
then you're thinking,
Lex Fridman (1:17:51.440)
okay, how do we make it more and more general?
Lex Fridman (1:17:53.400)
On the other hand,
Ben Goertzel (1:17:54.400)
if you start from AICSI or Schmidhuber's Gödel machine,
Lex Fridman (1:17:58.080)
or these infinitely powerful,
Lex Fridman (1:18:01.120)
but practically infeasible AIs,
Lex Fridman (1:18:04.000)
then getting to a human level AGI
Ben Goertzel (1:18:06.440)
is a matter of specialization.
Lex Fridman (1:18:08.240)
It's like, how do you take these
Ben Goertzel (1:18:10.200)
maximally general learning processes
Lex Fridman (1:18:12.880)
and how do you specialize them
Lex Fridman (1:18:15.760)
so that they can operate
Lex Fridman (1:18:17.600)
within the resource constraints that you have,
Lex Fridman (1:18:20.520)
but will achieve the particular things that you care about?
Lex Fridman (1:18:24.360)
Because we humans are not maximally general intelligence.
Ben Goertzel (1:18:28.200)
If I ask you to run a maze in 750 dimensions,
Lex Fridman (1:18:31.400)
you'd probably be very slow.
Ben Goertzel (1:18:33.040)
Whereas at two dimensions,
Lex Fridman (1:18:34.600)
you're probably way better, right?
Ben Goertzel (1:18:37.080)
So, I mean, we're special because our hippocampus
Lex Fridman (1:18:40.800)
has a two dimensional map in it, right?
Lex Fridman (1:18:43.080)
And it does not have a 750 dimensional map in it.
Lex Fridman (1:18:46.000)
So, I mean, we're a peculiar mix
Lex Fridman (1:18:51.440)
of generality and specialization, right?
Lex Fridman (1:18:56.000)
We'll probably start quite general at birth.
Ben Goertzel (1:18:59.200)
Not obviously still narrow,
Lex Fridman (1:19:00.760)
but like more general than we are
Ben Goertzel (1:19:03.200)
at age 20 and 30 and 40 and 50 and 60.
Lex Fridman (1:19:07.520)
I don't think that, I think it's more complex than that
Ben Goertzel (1:19:10.240)
because I mean, in some sense,
Lex Fridman (1:19:13.800)
a young child is less biased
Lex Fridman (1:19:17.520)
and the brain has yet to sort of crystallize
Lex Fridman (1:19:20.000)
into appropriate structures
Ben Goertzel (1:19:22.360)
for processing aspects of the physical and social world.
Lex Fridman (1:19:25.360)
On the other hand,
Ben Goertzel (1:19:26.560)
the young child is very tied to their sensorium.
Lex Fridman (1:19:30.120)
Whereas we can deal with abstract mathematics,
Ben Goertzel (1:19:33.880)
like 750 dimensions and the young child cannot
Lex Fridman (1:19:37.600)
because they haven't grown what Piaget
Ben Goertzel (1:19:40.920)
called the formal capabilities.
Lex Fridman (1:19:44.000)
They haven't learned to abstract yet, right?
Lex Fridman (1:19:46.240)
And the ability to abstract
Lex Fridman (1:19:48.120)
gives you a different kind of generality
Ben Goertzel (1:19:49.720)
than what the baby has.
Lex Fridman (1:19:51.680)
So, there's both more specialization
Lex Fridman (1:19:55.400)
and more generalization that comes
Lex Fridman (1:19:57.240)
with the development process actually.
Ben Goertzel (1:19:59.760)
I mean, I guess just the trajectories
Lex Fridman (1:20:02.320)
of the specialization are most controllable
Ben Goertzel (1:20:06.320)
at the young age, I guess is one way to put it.
Lex Fridman (1:20:09.720)
Do you have kids?
Ben Goertzel (1:20:10.720)
No.
Lex Fridman (1:20:11.680)
They're not as controllable as you think.
Ben Goertzel (1:20:13.600)
So, you think it's interesting.
Lex Fridman (1:20:15.880)
I think, honestly, I think a human adult
Ben Goertzel (1:20:19.040)
is much more generally intelligent than a human baby.
Lex Fridman (1:20:23.240)
Babies are very stupid, you know what I mean?
Ben Goertzel (1:20:25.800)
I mean, they're cute, which is why we put up
Lex Fridman (1:20:29.480)
with their repetitiveness and stupidity.
Lex Fridman (1:20:33.080)
And they have what the Zen guys would call
Lex Fridman (1:20:35.040)
a beginner's mind, which is a beautiful thing,
Lex Fridman (1:20:38.200)
but that doesn't necessarily correlate
Lex Fridman (1:20:40.760)
with a high level of intelligence.
Ben Goertzel (1:20:43.320)
On the plot of cuteness and stupidity,
Lex Fridman (1:20:46.120)
there's a process that allows us to put up
Ben Goertzel (1:20:48.720)
with their stupidity as they become more intelligent.
Lex Fridman (1:20:50.880)
So, by the time you're an ugly old man like me,
Ben Goertzel (1:20:52.400)
you gotta get really, really smart to compensate.
Lex Fridman (1:20:54.720)
To compensate, okay, cool.
Lex Fridman (1:20:56.160)
But yeah, going back to your original question,
Lex Fridman (1:20:59.160)
so the way I look at human level AGI
Ben Goertzel (1:21:05.280)
is how do you specialize, you know,
Lex Fridman (1:21:08.640)
unrealistically inefficient, superhuman,
Ben Goertzel (1:21:12.160)
brute force learning processes
Lex Fridman (1:21:14.600)
to the specific goals that humans need to achieve
Lex Fridman (1:21:18.320)
and the specific resources that we have.
Lex Fridman (1:21:21.920)
And both of these, the goals and the resources
Lex Fridman (1:21:24.600)
and the environments, I mean, all this is important.
Lex Fridman (1:21:27.120)
And on the resources side, it's important
Ben Goertzel (1:21:31.320)
that the hardware resources we're bringing to bear
Lex Fridman (1:21:35.600)
are very different than the human brain.
Lex Fridman (1:21:38.240)
So the way I would want to implement AGI
Lex Fridman (1:21:42.680)
on a bunch of neurons in a vat
Ben Goertzel (1:21:45.960)
that I could rewire arbitrarily is quite different
Lex Fridman (1:21:48.880)
than the way I would want to create AGI
Ben Goertzel (1:21:51.760)
on say a modern server farm of CPUs and GPUs,
Lex Fridman (1:21:55.760)
which in turn may be quite different
Ben Goertzel (1:21:57.440)
than the way I would want to implement AGI
Lex Fridman (1:22:00.200)
on whatever quantum computer we'll have in 10 years,
Ben Goertzel (1:22:03.760)
supposing someone makes a robust quantum turing machine
Lex Fridman (1:22:06.680)
or something, right?
Lex Fridman (1:22:08.240)
So I think there's been coevolution
Lex Fridman (1:22:12.640)
of the patterns of organization in the human brain
Lex Fridman (1:22:16.960)
and the physiological particulars
Lex Fridman (1:22:19.960)
of the human brain over time.
Lex Fridman (1:22:23.240)
And when you look at neural networks,
Lex Fridman (1:22:25.240)
that is one powerful class of learning algorithms,
Lex Fridman (1:22:28.040)
but it's also a class of learning algorithms
Lex Fridman (1:22:30.040)
that evolve to exploit the particulars of the human brain
Ben Goertzel (1:22:33.400)
as a computational substrate.
Lex Fridman (1:22:36.320)
If you're looking at the computational substrate
Ben Goertzel (1:22:38.880)
of a modern server farm,
Lex Fridman (1:22:41.040)
you won't necessarily want the same algorithms
Ben Goertzel (1:22:43.200)
that you want on the human brain.
Lex Fridman (1:22:45.760)
And from the right level of abstraction,
Ben Goertzel (1:22:48.920)
you could look at maybe the best algorithms on the brain
Lex Fridman (1:22:51.760)
and the best algorithms on a modern computer network
Ben Goertzel (1:22:54.480)
as implementing the same abstract learning
Lex Fridman (1:22:56.480)
and representation processes,
Lex Fridman (1:22:59.080)
but finding that level of abstraction
Lex Fridman (1:23:01.680)
is its own AGI research project then, right?
Lex Fridman (1:23:04.960)
So that's about the hardware side
Lex Fridman (1:23:07.800)
and the software side, which follows from that.
Ben Goertzel (1:23:10.880)
Then regarding what are the requirements,
Lex Fridman (1:23:14.200)
I wrote the paper years ago
Ben Goertzel (1:23:16.440)
on what I called the embodied communication prior,
Lex Fridman (1:23:20.360)
which was quite similar in intent
Ben Goertzel (1:23:22.960)
to Yoshua Bengio's recent paper on the consciousness prior,
Lex Fridman (1:23:26.760)
except I didn't wanna wrap up consciousness in it
Ben Goertzel (1:23:30.440)
because to me, the qualia problem and subjective experience
Lex Fridman (1:23:34.240)
is a very interesting issue also,
Ben Goertzel (1:23:35.880)
which we can chat about,
Lex Fridman (1:23:37.880)
but I would rather keep that philosophical debate distinct
Ben Goertzel (1:23:43.200)
from the debate of what kind of biases
Lex Fridman (1:23:45.240)
do you wanna put in a general intelligence
Ben Goertzel (1:23:47.040)
to give it human like general intelligence.
Lex Fridman (1:23:49.800)
And I'm not sure Yoshua Bengio is really addressing
Ben Goertzel (1:23:53.320)
that kind of consciousness.
Lex Fridman (1:23:55.080)
He's just using the term.
Ben Goertzel (1:23:56.560)
I love Yoshua to pieces.
Lex Fridman (1:23:58.600)
Like he's by far my favorite of the lines of deep learning.
Ben Goertzel (1:24:02.960)
Yeah.
Lex Fridman (1:24:03.800)
He's such a good hearted guy.
Ben Goertzel (1:24:05.800)
He's a good human being.
Lex Fridman (1:24:07.000)
Yeah, for sure.
Ben Goertzel (1:24:07.840)
I am not sure he has plumbed to the depths
Lex Fridman (1:24:11.200)
of the philosophy of consciousness.
Ben Goertzel (1:24:13.520)
No, he's using it as a sexy term.
Lex Fridman (1:24:15.040)
Yeah, yeah, yeah.
Lex Fridman (1:24:15.880)
So what I called it was the embodied communication prior.
Lex Fridman (1:24:21.160)
Can you maybe explain it a little bit?
Ben Goertzel (1:24:22.520)
Yeah, yeah.
Lex Fridman (1:24:23.360)
What I meant was, what are we humans evolved for?
Lex Fridman (1:24:26.640)
You can say being human, but that's very abstract, right?
Lex Fridman (1:24:29.720)
I mean, our minds control individual bodies,
Ben Goertzel (1:24:32.960)
which are autonomous agents moving around in a world
Lex Fridman (1:24:36.920)
that's composed largely of solid objects, right?
Lex Fridman (1:24:41.280)
And we've also evolved to communicate via language
Lex Fridman (1:24:46.240)
with other solid object agents that are going around
Ben Goertzel (1:24:49.960)
doing things collectively with us
Lex Fridman (1:24:52.200)
in a world of solid objects.
Lex Fridman (1:24:54.400)
And these things are very obvious,
Lex Fridman (1:24:56.920)
but if you compare them to the scope
Ben Goertzel (1:24:58.400)
of all possible intelligences
Lex Fridman (1:25:01.400)
or even all possible intelligences
Ben Goertzel (1:25:03.120)
that are physically realizable,
Lex Fridman (1:25:05.400)
that actually constrains things a lot.
Lex Fridman (1:25:07.400)
So if you start to look at how would you realize
Lex Fridman (1:25:13.000)
some specialized or constrained version
Ben Goertzel (1:25:15.880)
of universal general intelligence
Lex Fridman (1:25:18.360)
in a system that has limited memory
Lex Fridman (1:25:21.160)
and limited speed of processing,
Lex Fridman (1:25:23.160)
but whose general intelligence will be biased
Ben Goertzel (1:25:26.200)
toward controlling a solid object agent,
Lex Fridman (1:25:28.840)
which is mobile in a solid object world
Ben Goertzel (1:25:31.360)
for manipulating solid objects
Lex Fridman (1:25:33.480)
and communicating via language with other similar agents
Lex Fridman (1:25:38.560)
in that same world, right?
Lex Fridman (1:25:39.920)
Then starting from that,
Ben Goertzel (1:25:41.560)
you're starting to get a requirements analysis
Lex Fridman (1:25:43.640)
for human level general intelligence.
Lex Fridman (1:25:48.120)
And then that leads you into cognitive science
Lex Fridman (1:25:50.920)
and you can look at, say, what are the different types
Lex Fridman (1:25:53.080)
of memory that the human mind and brain has?
Lex Fridman (1:25:56.960)
And this has matured over the last decades
Lex Fridman (1:26:00.840)
and I got into this a lot.
Lex Fridman (1:26:02.920)
So after getting my PhD in math,
Ben Goertzel (1:26:04.600)
I was an academic for eight years.
Lex Fridman (1:26:06.080)
I was in departments of mathematics,
Ben Goertzel (1:26:08.720)
computer science, and psychology.
Lex Fridman (1:26:11.320)
When I was in the psychology department
Ben Goertzel (1:26:12.760)
at the University of Western Australia,
Lex Fridman (1:26:14.240)
I was focused on cognitive science of memory and perception.
Ben Goertzel (1:26:18.720)
Actually, I was teaching neural nets and deep neural nets
Lex Fridman (1:26:21.280)
and it was multi layer perceptrons, right?
Lex Fridman (1:26:23.600)
Psychology?
Lex Fridman (1:26:24.640)
Yeah.
Ben Goertzel (1:26:25.800)
Cognitive science, it was cross disciplinary
Lex Fridman (1:26:27.880)
among engineering, math, psychology, philosophy,
Ben Goertzel (1:26:31.280)
linguistics, computer science.
Lex Fridman (1:26:33.280)
But yeah, we were teaching psychology students
Ben Goertzel (1:26:35.960)
to try to model the data from human cognition experiments
Lex Fridman (1:26:40.040)
using multi layer perceptrons,
Ben Goertzel (1:26:42.080)
which was the early version of a deep neural network.
Lex Fridman (1:26:45.040)
Very, very, yeah, recurrent back prop
Lex Fridman (1:26:47.880)
was very, very slow to train back then, right?
Lex Fridman (1:26:51.200)
So this is the study of these constraint systems
Ben Goertzel (1:26:53.920)
that are supposed to deal with physical objects.
Lex Fridman (1:26:55.640)
So if you look at cognitive psychology,
Ben Goertzel (1:27:01.480)
you can see there's multiple types of memory,
Lex Fridman (1:27:04.520)
which are to some extent represented
Ben Goertzel (1:27:06.560)
by different subsystems in the human brain.
Lex Fridman (1:27:08.480)
So we have episodic memory,
Ben Goertzel (1:27:10.360)
which takes into account our life history
Lex Fridman (1:27:13.520)
and everything that's happened to us.
Ben Goertzel (1:27:15.240)
We have declarative or semantic memory,
Lex Fridman (1:27:17.320)
which is like facts and beliefs abstracted
Ben Goertzel (1:27:20.080)
from the particular situations that they occurred in.
Lex Fridman (1:27:22.840)
There's sensory memory, which to some extent
Ben Goertzel (1:27:26.120)
is sense modality specific,
Lex Fridman (1:27:27.600)
and then to some extent is unified across sense modalities.
Ben Goertzel (1:27:33.360)
There's procedural memory, memory of how to do stuff,
Lex Fridman (1:27:36.120)
like how to swing the tennis racket, right?
Ben Goertzel (1:27:38.160)
Which is, there's motor memory,
Lex Fridman (1:27:39.920)
but it's also a little more abstract than motor memory.
Ben Goertzel (1:27:43.640)
It involves cerebellum and cortex working together.
Lex Fridman (1:27:47.520)
Then there's memory linkage with emotion
Ben Goertzel (1:27:51.600)
which has to do with linkages of cortex and limbic system.
Lex Fridman (1:27:55.920)
There's specifics of spatial and temporal modeling
Ben Goertzel (1:27:59.160)
connected with memory, which has to do with hippocampus
Lex Fridman (1:28:02.760)
and thalamus connecting to cortex.
Lex Fridman (1:28:05.360)
And the basal ganglia, which influences goals.
Lex Fridman (1:28:08.160)
So we have specific memory of what goals,
Ben Goertzel (1:28:10.960)
subgoals and sub subgoals we want to perceive
Lex Fridman (1:28:13.160)
in which context in the past.
Ben Goertzel (1:28:15.040)
Human brain has substantially different subsystems
Lex Fridman (1:28:18.240)
for these different types of memory
Lex Fridman (1:28:21.040)
and substantially differently tuned learning,
Lex Fridman (1:28:24.240)
like differently tuned modes of longterm potentiation
Ben Goertzel (1:28:27.280)
to do with the types of neurons and neurotransmitters
Lex Fridman (1:28:29.720)
in the different parts of the brain
Ben Goertzel (1:28:31.280)
corresponding to these different types of knowledge.
Lex Fridman (1:28:33.040)
And these different types of memory and learning
Ben Goertzel (1:28:35.880)
in the human brain, I mean, you can back these all
Lex Fridman (1:28:38.520)
into embodied communication for controlling agents
Ben Goertzel (1:28:41.920)
in worlds of solid objects.
Lex Fridman (1:28:44.680)
Now, so if you look at building an AGI system,
Ben Goertzel (1:28:47.720)
one way to do it, which starts more from cognitive science
Lex Fridman (1:28:50.440)
than neuroscience is to say,
Ben Goertzel (1:28:52.680)
okay, what are the types of memory
Lex Fridman (1:28:55.240)
that are necessary for this kind of world?
Ben Goertzel (1:28:57.360)
Yeah, yeah, necessary for this sort of intelligence.
Lex Fridman (1:29:00.720)
What types of learning work well
Lex Fridman (1:29:02.760)
with these different types of memory?
Lex Fridman (1:29:04.600)
And then how do you connect all these things together, right?
Lex Fridman (1:29:07.800)
And of course the human brain did it incrementally
Lex Fridman (1:29:10.800)
through evolution because each of the sub networks
Ben Goertzel (1:29:14.360)
of the brain, I mean, it's not really the lobes
Lex Fridman (1:29:16.680)
of the brain, it's the sub networks,
Ben Goertzel (1:29:18.240)
each of which is widely distributed,
Lex Fridman (1:29:20.800)
which of the, each of the sub networks of the brain
Ben Goertzel (1:29:23.680)
co evolves with the other sub networks of the brain,
Lex Fridman (1:29:27.160)
both in terms of its patterns of organization
Lex Fridman (1:29:29.480)
and the particulars of the neurophysiology.
Lex Fridman (1:29:31.840)
So they all grew up communicating
Lex Fridman (1:29:33.440)
and adapting to each other.
Lex Fridman (1:29:34.440)
It's not like they were separate black boxes
Lex Fridman (1:29:36.720)
that were then glommed together, right?
Lex Fridman (1:29:40.200)
Whereas as engineers, we would tend to say,
Ben Goertzel (1:29:43.320)
let's make the declarative memory box here
Lex Fridman (1:29:46.680)
and the procedural memory box here
Lex Fridman (1:29:48.440)
and the perception box here and wire them together.
Lex Fridman (1:29:51.400)
And when you can do that, it's interesting.
Lex Fridman (1:29:54.120)
I mean, that's how a car is built, right?
Lex Fridman (1:29:55.680)
But on the other hand, that's clearly not
Lex Fridman (1:29:58.560)
how biological systems are made.
Lex Fridman (1:30:01.400)
The parts co evolve so as to adapt and work together.
Ben Goertzel (1:30:05.360)
That's by the way, how every human engineered system
Lex Fridman (1:30:09.240)
that flies, that was, we were using that analogy
Ben Goertzel (1:30:11.640)
before it's built as well.
Lex Fridman (1:30:13.000)
So do you find this at all appealing?
Ben Goertzel (1:30:14.440)
Like there's been a lot of really exciting,
Lex Fridman (1:30:16.680)
which I find strange that it's ignored work
Ben Goertzel (1:30:20.160)
in cognitive architectures, for example,
Lex Fridman (1:30:21.880)
throughout the last few decades.
Lex Fridman (1:30:23.320)
Do you find that?
Lex Fridman (1:30:24.320)
Yeah, I mean, I had a lot to do with that community
Lex Fridman (1:30:27.960)
and you know, Paul Rosenbloom, who was one of the,
Lex Fridman (1:30:31.000)
and John Laird who built the SOAR architecture,
Ben Goertzel (1:30:33.480)
are friends of mine.
Lex Fridman (1:30:34.640)
And I learned SOAR quite well
Lex Fridman (1:30:37.160)
and ACTAR and these different cognitive architectures.
Lex Fridman (1:30:39.440)
And how I was looking at the AI world about 10 years ago
Ben Goertzel (1:30:44.520)
before this whole commercial deep learning explosion was,
Lex Fridman (1:30:47.840)
on the one hand, you had these cognitive architecture guys
Ben Goertzel (1:30:51.560)
who were working closely with psychologists
Lex Fridman (1:30:53.480)
and cognitive scientists who had thought a lot
Ben Goertzel (1:30:55.760)
about how the different parts of a human like mind
Lex Fridman (1:30:58.840)
should work together.
Ben Goertzel (1:31:00.400)
On the other hand, you had these learning theory guys
Lex Fridman (1:31:03.600)
who didn't care at all about the architecture,
Lex Fridman (1:31:06.040)
but we're just thinking about like,
Lex Fridman (1:31:07.360)
how do you recognize patterns in large amounts of data?
Lex Fridman (1:31:10.280)
And in some sense, what you needed to do
Lex Fridman (1:31:14.560)
was to get the learning that the learning theory guys
Ben Goertzel (1:31:18.440)
were doing and put it together with the architecture
Lex Fridman (1:31:21.440)
that the cognitive architecture guys were doing.
Lex Fridman (1:31:24.240)
And then you would have what you needed.
Lex Fridman (1:31:25.960)
Now, you can't, unfortunately, when you look at the details,
Ben Goertzel (1:31:31.600)
you can't just do that without totally rebuilding
Lex Fridman (1:31:34.960)
what is happening on both the cognitive architecture
Lex Fridman (1:31:37.840)
and the learning side.
Lex Fridman (1:31:38.760)
So, I mean, they tried to do that in SOAR,
Lex Fridman (1:31:41.760)
but what they ultimately did is like,
Lex Fridman (1:31:43.960)
take a deep neural net or something for perception
Lex Fridman (1:31:46.560)
and you include it as one of the black boxes.
Lex Fridman (1:31:50.800)
It becomes one of the boxes.
Ben Goertzel (1:31:51.960)
The learning mechanism becomes one of the boxes
Lex Fridman (1:31:53.800)
as opposed to fundamental part of the system.
Ben Goertzel (1:31:57.440)
You could look at some of the stuff DeepMind has done,
Lex Fridman (1:32:00.400)
like the differential neural computer or something
Ben Goertzel (1:32:03.240)
that sort of has a neural net for deep learning perception.
Lex Fridman (1:32:07.080)
It has another neural net, which is like a memory matrix
Ben Goertzel (1:32:10.640)
that stores, say, the map of the London subway or something.
Lex Fridman (1:32:13.080)
So probably Demis Tsabas was thinking about this
Ben Goertzel (1:32:16.440)
like part of cortex and part of hippocampus
Lex Fridman (1:32:18.520)
because hippocampus has a spatial map.
Lex Fridman (1:32:20.440)
And when he was a neuroscientist,
Lex Fridman (1:32:21.720)
he was doing a bunch on cortex hippocampus interconnection.
Lex Fridman (1:32:24.600)
So there, the DNC would be an example of folks
Lex Fridman (1:32:27.320)
from the deep neural net world trying to take a step
Ben Goertzel (1:32:30.160)
in the cognitive architecture direction
Lex Fridman (1:32:32.200)
by having two neural modules that correspond roughly
Ben Goertzel (1:32:35.000)
to two different parts of the human brain
Lex Fridman (1:32:36.720)
that deal with different kinds of memory and learning.
Lex Fridman (1:32:38.920)
But on the other hand, it's super, super, super crude
Lex Fridman (1:32:42.000)
from the cognitive architecture view, right?
Ben Goertzel (1:32:44.360)
Just as what John Laird and Soar did with neural nets
Lex Fridman (1:32:48.080)
was super, super crude from a learning point of view
Ben Goertzel (1:32:51.200)
because the learning was like off to the side,
Lex Fridman (1:32:53.360)
not affecting the core representations, right?
Ben Goertzel (1:32:55.880)
I mean, you weren't learning the representation.
Lex Fridman (1:32:57.880)
You were learning the data that feeds into the...
Ben Goertzel (1:33:00.080)
You were learning abstractions of perceptual data
Lex Fridman (1:33:02.600)
to feed into the representation that was not learned, right?
Lex Fridman (1:33:06.560)
So yeah, this was clear to me a while ago.
Lex Fridman (1:33:11.000)
And one of my hopes with the AGI community
Ben Goertzel (1:33:14.240)
was to sort of bring people
Lex Fridman (1:33:15.960)
from those two directions together.
Ben Goertzel (1:33:19.320)
That didn't happen much in terms of...
Lex Fridman (1:33:21.920)
Not yet.
Lex Fridman (1:33:22.760)
And what I was gonna say is it didn't happen
Lex Fridman (1:33:24.520)
in terms of bringing like the lions
Ben Goertzel (1:33:26.360)
of cognitive architecture together
Lex Fridman (1:33:28.560)
with the lions of deep learning.
Ben Goertzel (1:33:30.480)
It did work in the sense that a bunch of younger researchers
Lex Fridman (1:33:33.760)
have had their heads filled with both of those ideas.
Ben Goertzel (1:33:35.760)
This comes back to a saying my dad,
Lex Fridman (1:33:38.840)
who was a university professor, often quoted to me,
Ben Goertzel (1:33:41.360)
which was, science advances one funeral at a time,
Lex Fridman (1:33:45.840)
which I'm trying to avoid.
Ben Goertzel (1:33:47.840)
Like I'm 53 years old and I'm trying to invent
Lex Fridman (1:33:51.320)
amazing, weird ass new things
Ben Goertzel (1:33:53.480)
that nobody ever thought about,
Lex Fridman (1:33:56.160)
which we'll talk about in a few minutes.
Lex Fridman (1:33:59.240)
But there is that aspect, right?
Lex Fridman (1:34:02.280)
Like the people who've been at AI a long time
Lex Fridman (1:34:05.680)
and have made their career developing one aspect,
Lex Fridman (1:34:08.760)
like a cognitive architecture or a deep learning approach,
Ben Goertzel (1:34:12.880)
it can be hard once you're old
Lex Fridman (1:34:14.760)
and have made your career doing one thing,
Ben Goertzel (1:34:17.280)
it can be hard to mentally shift gears.
Lex Fridman (1:34:19.640)
I mean, I try quite hard to remain flexible minded.
Ben Goertzel (1:34:23.640)
Have you been successful somewhat in changing,
Lex Fridman (1:34:26.480)
maybe, have you changed your mind on some aspects
Lex Fridman (1:34:29.640)
of what it takes to build an AGI, like technical things?
Lex Fridman (1:34:32.920)
The hard part is that the world doesn't want you to.
Lex Fridman (1:34:36.040)
The world or your own brain?
Lex Fridman (1:34:37.360)
The world, well, that one point
Ben Goertzel (1:34:39.560)
is that your brain doesn't want to.
Lex Fridman (1:34:41.040)
The other part is that the world doesn't want you to.
Ben Goertzel (1:34:43.480)
Like the people who have followed your ideas
Lex Fridman (1:34:46.520)
get mad at you if you change your mind.
Lex Fridman (1:34:49.280)
And the media wants to pigeonhole you as an avatar
Lex Fridman (1:34:54.560)
of a certain idea.
Lex Fridman (1:34:57.160)
But yeah, I've changed my mind on a bunch of things.
Lex Fridman (1:35:01.480)
I mean, when I started my career,
Ben Goertzel (1:35:03.800)
I really thought quantum computing
Lex Fridman (1:35:05.240)
would be necessary for AGI.
Lex Fridman (1:35:07.920)
And I doubt it's necessary now,
Lex Fridman (1:35:10.800)
although I think it will be a super major enhancement.
Lex Fridman (1:35:14.680)
But I mean, I'm now in the middle of embarking
Lex Fridman (1:35:19.360)
on the complete rethink and rewrite from scratch
Ben Goertzel (1:35:23.400)
of our OpenCog AGI system together with Alexey Potapov
Lex Fridman (1:35:28.480)
and his team in St. Petersburg,
Ben Goertzel (1:35:29.840)
who's working with me in SingularityNet.
Lex Fridman (1:35:31.600)
So now we're trying to like go back to basics,
Ben Goertzel (1:35:35.680)
take everything we learned from working
Lex Fridman (1:35:37.800)
with the current OpenCog system,
Ben Goertzel (1:35:39.600)
take everything everybody else has learned
Lex Fridman (1:35:41.880)
from working with their proto AGI systems
Lex Fridman (1:35:45.680)
and design the best framework for the next stage.
Lex Fridman (1:35:50.000)
And I do think there's a lot to be learned
Ben Goertzel (1:35:53.320)
from the recent successes with deep neural nets
Lex Fridman (1:35:56.800)
and deep reinforcement systems.
Ben Goertzel (1:35:59.000)
I mean, people made these essentially trivial systems
Lex Fridman (1:36:02.680)
work much better than I thought they would.
Lex Fridman (1:36:04.840)
And there's a lot to be learned from that.
Lex Fridman (1:36:07.080)
And I wanna incorporate that knowledge appropriately
Ben Goertzel (1:36:10.720)
in our OpenCog 2.0 system.
Lex Fridman (1:36:13.520)
On the other hand, I also think current deep neural net
Ben Goertzel (1:36:18.520)
architectures as such will never get you anywhere near AGI.
Lex Fridman (1:36:22.240)
So I think you wanna avoid the pathology
Ben Goertzel (1:36:25.080)
of throwing the baby out with the bathwater
Lex Fridman (1:36:28.360)
and like saying, well, these things are garbage
Ben Goertzel (1:36:30.880)
because foolish journalists overblow them
Lex Fridman (1:36:33.840)
as being the path to AGI
Lex Fridman (1:36:37.040)
and a few researchers overblow them as well.
Lex Fridman (1:36:41.600)
There's a lot of interesting stuff to be learned there
Ben Goertzel (1:36:45.440)
even though those are not the golden path.
Lex Fridman (1:36:48.000)
So maybe this is a good chance to step back.
Ben Goertzel (1:36:50.160)
You mentioned OpenCog 2.0, but...
Lex Fridman (1:36:52.920)
Go back to OpenCog 0.0, which exists now.
Ben Goertzel (1:36:56.040)
Alpha, yeah.
Lex Fridman (1:36:58.440)
Yeah, maybe talk through the history of OpenCog
Lex Fridman (1:37:01.920)
and your thinking about these ideas.
Lex Fridman (1:37:03.920)
I would say OpenCog 2.0 is a term we're throwing around
Ben Goertzel (1:37:10.120)
sort of tongue in cheek because the existing OpenCog system
Lex Fridman (1:37:14.560)
that we're working on now is not remotely close
Lex Fridman (1:37:17.200)
to what we'd consider a 1.0, right?
Lex Fridman (1:37:20.000)
I mean, it's an early...
Ben Goertzel (1:37:23.360)
It's been around, what, 13 years or something,
Lex Fridman (1:37:27.400)
but it's still an early stage research system, right?
Lex Fridman (1:37:29.800)
And actually, we are going back to the beginning
Lex Fridman (1:37:37.360)
in terms of theory and implementation
Ben Goertzel (1:37:40.680)
because we feel like that's the right thing to do,
Lex Fridman (1:37:42.840)
but I'm sure what we end up with is gonna have
Ben Goertzel (1:37:45.560)
a huge amount in common with the current system.
Lex Fridman (1:37:48.560)
I mean, we all still like the general approach.
Lex Fridman (1:37:51.640)
So first of all, what is OpenCog?
Lex Fridman (1:37:54.400)
Sure, OpenCog is an open source software project
Ben Goertzel (1:37:59.800)
that I launched together with several others in 2008
Lex Fridman (1:38:04.400)
and probably the first code written toward that
Ben Goertzel (1:38:08.280)
was written in 2001 or two or something
Lex Fridman (1:38:11.160)
that was developed as a proprietary code base
Ben Goertzel (1:38:15.320)
within my AI company, Novamente LLC.
Lex Fridman (1:38:18.280)
Then we decided to open source it in 2008,
Ben Goertzel (1:38:22.000)
cleaned up the code throughout some things
Lex Fridman (1:38:23.840)
and added some new things and...
Lex Fridman (1:38:26.920)
What language is it written in?
Lex Fridman (1:38:28.080)
It's C++.
Ben Goertzel (1:38:29.440)
Primarily, there's a bunch of scheme as well,
Lex Fridman (1:38:31.400)
but most of it's C++.
Lex Fridman (1:38:33.040)
And it's separate from something we'll also talk about,
Lex Fridman (1:38:36.520)
the SingularityNet.
Lex Fridman (1:38:37.480)
So it was born as a non networked thing.
Lex Fridman (1:38:41.360)
Correct, correct.
Ben Goertzel (1:38:42.400)
Well, there are many levels of networks involved here.
Lex Fridman (1:38:47.000)
No connectivity to the internet, or no, at birth.
Ben Goertzel (1:38:52.000)
Yeah, I mean, SingularityNet is a separate project
Lex Fridman (1:38:57.240)
and a separate body of code.
Lex Fridman (1:38:59.440)
And you can use SingularityNet as part of the infrastructure
Lex Fridman (1:39:02.600)
for a distributed OpenCog system,
Lex Fridman (1:39:04.480)
but there are different layers.
Lex Fridman (1:39:07.520)
Yeah, got it.
Lex Fridman (1:39:08.360)
So OpenCog on the one hand as a software framework
Lex Fridman (1:39:14.840)
could be used to implement a variety
Ben Goertzel (1:39:17.000)
of different AI architectures and algorithms,
Lex Fridman (1:39:21.840)
but in practice, there's been a group of developers
Ben Goertzel (1:39:26.440)
which I've been leading together with Linus Vepstas,
Lex Fridman (1:39:29.440)
Neil Geisweiler, and a few others,
Ben Goertzel (1:39:31.680)
which have been using the OpenCog platform
Lex Fridman (1:39:35.080)
and infrastructure to implement certain ideas
Ben Goertzel (1:39:39.440)
about how to make an AGI.
Lex Fridman (1:39:41.280)
So there's been a little bit of ambiguity
Ben Goertzel (1:39:43.480)
about OpenCog, the software platform
Lex Fridman (1:39:46.120)
versus OpenCog, the AGI design,
Ben Goertzel (1:39:49.360)
because in theory, you could use that software to do,
Lex Fridman (1:39:52.160)
you could use it to make a neural net.
Ben Goertzel (1:39:53.440)
You could use it to make a lot of different AGI.
Lex Fridman (1:39:55.880)
What kind of stuff does the software platform provide,
Lex Fridman (1:39:58.640)
like in terms of utilities, tools, like what?
Lex Fridman (1:40:00.760)
Yeah, let me first tell about OpenCog
Ben Goertzel (1:40:03.840)
as a software platform,
Lex Fridman (1:40:05.520)
and then I'll tell you the specific AGI R&D
Ben Goertzel (1:40:08.680)
we've been building on top of it.
Lex Fridman (1:40:12.240)
So the core component of OpenCog as a software platform
Ben Goertzel (1:40:16.200)
is what we call the atom space,
Lex Fridman (1:40:17.920)
which is a weighted labeled hypergraph.
Ben Goertzel (1:40:21.240)
ATOM, atom space.
Lex Fridman (1:40:22.880)
Atom space, yeah, yeah, not atom, like Adam and Eve,
Ben Goertzel (1:40:25.880)
although that would be cool too.
Lex Fridman (1:40:28.080)
Yeah, so you have a hypergraph, which is like,
Lex Fridman (1:40:32.120)
so a graph in this sense is a bunch of nodes
Lex Fridman (1:40:35.360)
with links between them.
Ben Goertzel (1:40:37.120)
A hypergraph is like a graph,
Lex Fridman (1:40:40.960)
but links can go between more than two nodes.
Lex Fridman (1:40:43.960)
So you have a link between three nodes.
Lex Fridman (1:40:45.520)
And in fact, OpenCog's atom space
Ben Goertzel (1:40:49.560)
would properly be called a metagraph
Lex Fridman (1:40:51.760)
because you can have links pointing to links,
Lex Fridman (1:40:54.080)
or you could have links pointing to whole subgraphs, right?
Lex Fridman (1:40:56.840)
So it's an extended hypergraph or a metagraph.
Lex Fridman (1:41:00.920)
Is metagraph a technical term?
Lex Fridman (1:41:02.280)
It is now a technical term.
Ben Goertzel (1:41:03.640)
Interesting.
Lex Fridman (1:41:04.480)
But I don't think it was yet a technical term
Ben Goertzel (1:41:06.360)
when we started calling this a generalized hypergraph.
Lex Fridman (1:41:10.080)
But in any case, it's a weighted labeled
Ben Goertzel (1:41:13.400)
generalized hypergraph or weighted labeled metagraph.
Lex Fridman (1:41:16.920)
The weights and labels mean that the nodes and links
Ben Goertzel (1:41:19.200)
can have numbers and symbols attached to them.
Lex Fridman (1:41:22.360)
So they can have types on them.
Ben Goertzel (1:41:24.920)
They can have numbers on them that represent,
Lex Fridman (1:41:27.440)
say, a truth value or an importance value
Ben Goertzel (1:41:30.120)
for a certain purpose.
Lex Fridman (1:41:32.000)
And of course, like with all things,
Ben Goertzel (1:41:33.240)
you can reduce that to a hypergraph,
Lex Fridman (1:41:35.080)
and then the hypergraph can be reduced to a graph.
Ben Goertzel (1:41:35.920)
You can reduce hypergraph to a graph,
Lex Fridman (1:41:37.680)
and you could reduce a graph to an adjacency matrix.
Ben Goertzel (1:41:39.880)
So, I mean, there's always multiple representations.
Lex Fridman (1:41:42.720)
But there's a layer of representation
Ben Goertzel (1:41:44.000)
that seems to work well here.
Lex Fridman (1:41:45.120)
Got it.
Ben Goertzel (1:41:45.960)
Right, right, right.
Lex Fridman (1:41:46.800)
And so similarly, you could have a link to a whole graph
Ben Goertzel (1:41:52.080)
because a whole graph could represent,
Lex Fridman (1:41:53.440)
say, a body of information.
Lex Fridman (1:41:54.920)
And I could say, I reject this body of information.
Lex Fridman (1:41:58.640)
Then one way to do that is make that link
Ben Goertzel (1:42:00.320)
go to that whole subgraph representing
Lex Fridman (1:42:02.000)
the body of information, right?
Ben Goertzel (1:42:04.040)
I mean, there are many alternate representations,
Lex Fridman (1:42:07.200)
but that's, anyway, what we have in OpenCOG,
Ben Goertzel (1:42:10.720)
we have an atom space, which is this weighted, labeled,
Lex Fridman (1:42:13.160)
generalized hypergraph.
Ben Goertzel (1:42:15.080)
Knowledge store, it lives in RAM.
Lex Fridman (1:42:17.840)
There's also a way to back it up to disk.
Ben Goertzel (1:42:20.120)
There are ways to spread it among
Lex Fridman (1:42:22.320)
multiple different machines.
Ben Goertzel (1:42:24.120)
Then there are various utilities for dealing with that.
Lex Fridman (1:42:27.960)
So there's a pattern matcher,
Ben Goertzel (1:42:29.800)
which lets you specify a sort of abstract pattern
Lex Fridman (1:42:33.880)
and then search through a whole atom space
Ben Goertzel (1:42:36.200)
with labeled hypergraph to see what subhypergraphs
Lex Fridman (1:42:39.800)
may match that pattern, for an example.
Lex Fridman (1:42:42.880)
So that's, then there's something called
Lex Fridman (1:42:45.920)
the COG server in OpenCOG,
Ben Goertzel (1:42:48.760)
which lets you run a bunch of different agents
Lex Fridman (1:42:52.560)
or processes in a scheduler.
Lex Fridman (1:42:55.880)
And each of these agents, basically it reads stuff
Lex Fridman (1:42:59.160)
from the atom space and it writes stuff to the atom space.
Lex Fridman (1:43:01.880)
So this is sort of the basic operational model.
Lex Fridman (1:43:05.640)
That's the software framework.
Lex Fridman (1:43:07.760)
And of course that's, there's a lot there
Lex Fridman (1:43:10.360)
just from a scalable software engineering standpoint.
Lex Fridman (1:43:13.200)
So you could use this, I don't know if you've,
Lex Fridman (1:43:15.080)
have you looked into the Stephen Wolfram's physics project
Lex Fridman (1:43:18.000)
recently with the hypergraphs and stuff?
Lex Fridman (1:43:20.160)
Could you theoretically use like the software framework
Ben Goertzel (1:43:22.840)
to play with it? You certainly could,
Lex Fridman (1:43:23.800)
although Wolfram would rather die
Ben Goertzel (1:43:26.160)
than use anything but Mathematica for his work.
Lex Fridman (1:43:29.080)
Well that's, yeah, but there's a big community of people
Ben Goertzel (1:43:32.120)
who are, you know, would love integration.
Lex Fridman (1:43:36.080)
Like you said, the young minds love the idea
Ben Goertzel (1:43:38.400)
of integrating, of connecting things.
Lex Fridman (1:43:40.440)
Yeah, that's right.
Lex Fridman (1:43:41.280)
And I would add on that note,
Lex Fridman (1:43:42.840)
the idea of using hypergraph type models in physics
Ben Goertzel (1:43:46.600)
is not very new.
Lex Fridman (1:43:47.680)
Like if you look at...
Ben Goertzel (1:43:49.120)
The Russians did it first.
Lex Fridman (1:43:50.360)
Well, I'm sure they did.
Lex Fridman (1:43:52.200)
And a guy named Ben Dribis, who's a mathematician,
Lex Fridman (1:43:55.880)
a professor in Louisiana or somewhere,
Ben Goertzel (1:43:58.200)
had a beautiful book on quantum sets and hypergraphs
Lex Fridman (1:44:01.960)
and algebraic topology for discrete models of physics.
Lex Fridman (1:44:05.520)
And carried it much farther than Wolfram has,
Lex Fridman (1:44:09.080)
but he's not rich and famous,
Lex Fridman (1:44:10.920)
so it didn't get in the headlines.
Lex Fridman (1:44:13.280)
But yeah, Wolfram aside, yeah,
Ben Goertzel (1:44:15.280)
certainly that's a good way to put it.
Lex Fridman (1:44:17.120)
The whole OpenCog framework,
Ben Goertzel (1:44:19.280)
you could use it to model biological networks
Lex Fridman (1:44:22.200)
and simulate biology processes.
Ben Goertzel (1:44:24.200)
You could use it to model physics
Lex Fridman (1:44:26.480)
on discrete graph models of physics.
Lex Fridman (1:44:30.160)
So you could use it to do, say, biologically realistic
Lex Fridman (1:44:36.840)
neural networks, for example.
Lex Fridman (1:44:39.280)
And that's a framework.
Lex Fridman (1:44:42.360)
What do agents and processes do?
Lex Fridman (1:44:44.240)
Do they grow the graph?
Lex Fridman (1:44:45.880)
What kind of computations, just to get a sense,
Lex Fridman (1:44:48.200)
are they supposed to do?
Lex Fridman (1:44:49.040)
So in theory, they could do anything they want to do.
Ben Goertzel (1:44:51.200)
They're just C++ processes.
Lex Fridman (1:44:53.320)
On the other hand, the computation framework
Ben Goertzel (1:44:56.880)
is sort of designed for agents
Lex Fridman (1:44:59.160)
where most of their processing time
Ben Goertzel (1:45:02.000)
is taken up with reads and writes to the atom space.
Lex Fridman (1:45:05.400)
And so that's a very different processing model
Ben Goertzel (1:45:09.000)
than, say, the matrix multiplication based model
Lex Fridman (1:45:12.440)
as underlies most deep learning systems, right?
Lex Fridman (1:45:15.080)
So you could create an agent
Lex Fridman (1:45:19.560)
that just factored numbers for a billion years.
Ben Goertzel (1:45:22.720)
It would run within the OpenCog platform,
Lex Fridman (1:45:25.000)
but it would be pointless, right?
Ben Goertzel (1:45:26.600)
I mean, the point of doing OpenCog
Lex Fridman (1:45:28.880)
is because you want to make agents
Ben Goertzel (1:45:30.520)
that are cooperating via reading and writing
Lex Fridman (1:45:33.160)
into this weighted labeled hypergraph, right?
Lex Fridman (1:45:36.400)
And that has both cognitive architecture importance
Lex Fridman (1:45:41.560)
because then this hypergraph is being used
Ben Goertzel (1:45:43.400)
as a sort of shared memory
Lex Fridman (1:45:46.040)
among different cognitive processes,
Lex Fridman (1:45:48.240)
but it also has software and hardware
Lex Fridman (1:45:51.000)
implementation implications
Ben Goertzel (1:45:52.840)
because current GPU architectures
Lex Fridman (1:45:54.840)
are not so useful for OpenCog,
Lex Fridman (1:45:57.120)
whereas a graph chip would be incredibly useful, right?
Lex Fridman (1:46:01.200)
And I think Graphcore has those now,
Lex Fridman (1:46:03.640)
but they're not ideally suited for this.
Lex Fridman (1:46:05.240)
But I think in the next, let's say, three to five years,
Ben Goertzel (1:46:10.640)
we're gonna see new chips
Lex Fridman (1:46:12.000)
where like a graph is put on the chip
Lex Fridman (1:46:14.680)
and the back and forth between multiple processes
Lex Fridman (1:46:19.320)
acting SIMD and MIMD on that graph is gonna be fast.
Lex Fridman (1:46:23.600)
And then that may do for OpenCog type architectures
Lex Fridman (1:46:26.480)
what GPUs did for deep neural architecture.
Ben Goertzel (1:46:29.840)
It's a small tangent.
Lex Fridman (1:46:31.320)
Can you comment on thoughts about neuromorphic computing?
Lex Fridman (1:46:34.600)
So like hardware implementations
Lex Fridman (1:46:36.400)
of all these different kind of, are you interested?
Lex Fridman (1:46:39.360)
Are you excited by that possibility?
Lex Fridman (1:46:41.000)
I'm excited by graph processors
Ben Goertzel (1:46:42.680)
because I think they can massively speed up OpenCog,
Lex Fridman (1:46:46.440)
which is a class of architectures that I'm working on.
Ben Goertzel (1:46:50.680)
I think if, you know, in principle, neuromorphic computing
Lex Fridman (1:46:57.240)
should be amazing.
Ben Goertzel (1:46:58.760)
I haven't yet been fully sold
Lex Fridman (1:47:00.480)
on any of the systems that are out.
Lex Fridman (1:47:03.320)
They're like, memristors should be amazing too, right?
Lex Fridman (1:47:06.400)
So a lot of these things have obvious potential,
Lex Fridman (1:47:09.400)
but I haven't yet put my hands on a system
Lex Fridman (1:47:11.360)
that seemed to manifest that.
Ben Goertzel (1:47:13.280)
Mark's system should be amazing,
Lex Fridman (1:47:14.880)
but the current systems have not been great.
Ben Goertzel (1:47:17.880)
Yeah, I mean, look, for example,
Lex Fridman (1:47:19.640)
if you wanted to make a biologically realistic
Ben Goertzel (1:47:23.960)
hardware neural network,
Lex Fridman (1:47:25.680)
like making a circuit in hardware
Ben Goertzel (1:47:31.520)
that emulated like the Hodgkin–Huxley equation
Lex Fridman (1:47:34.360)
or the Izhekevich equation,
Ben Goertzel (1:47:35.640)
like differential equations
Lex Fridman (1:47:38.240)
for a biologically realistic neuron
Lex Fridman (1:47:40.680)
and putting that in hardware on the chip,
Lex Fridman (1:47:43.800)
that would seem that it would make more feasible
Ben Goertzel (1:47:46.360)
to make a large scale, truly biologically realistic
Lex Fridman (1:47:50.320)
neural network.
Ben Goertzel (1:47:51.160)
Now, what's been done so far is not like that.
Lex Fridman (1:47:54.320)
So I guess personally, as a researcher,
Ben Goertzel (1:47:57.120)
I mean, I've done a bunch of work in computational neuroscience
Lex Fridman (1:48:02.480)
where I did some work with IARPA in DC,
Ben Goertzel (1:48:05.600)
Intelligence Advanced Research Project Agency.
Lex Fridman (1:48:08.240)
We were looking at how do you make
Ben Goertzel (1:48:10.880)
a biologically realistic simulation
Lex Fridman (1:48:13.000)
of seven different parts of the brain
Ben Goertzel (1:48:15.720)
cooperating with each other,
Lex Fridman (1:48:17.080)
using like realistic nonlinear dynamical models of neurons,
Lex Fridman (1:48:20.440)
and how do you get that to simulate
Lex Fridman (1:48:21.920)
what's going on in the mind of a geo intelligence analyst
Lex Fridman (1:48:24.800)
while they're trying to find terrorists on a map, right?
Lex Fridman (1:48:27.160)
So if you want to do something like that,
Ben Goertzel (1:48:29.880)
having neuromorphic hardware that really let you simulate
Lex Fridman (1:48:34.080)
like a realistic model of the neuron would be amazing.
Lex Fridman (1:48:38.720)
But that's sort of with my computational neuroscience
Lex Fridman (1:48:42.280)
hat on, right?
Ben Goertzel (1:48:43.120)
With an AGI hat on, I'm just more interested
Lex Fridman (1:48:47.160)
in these hypergraph knowledge representation
Ben Goertzel (1:48:50.200)
based architectures, which would benefit more
Lex Fridman (1:48:54.480)
from various types of graph processors
Ben Goertzel (1:48:57.720)
because the main processing bottleneck
Lex Fridman (1:49:00.480)
is reading writing to RAM.
Ben Goertzel (1:49:02.000)
It's reading writing to the graph in RAM.
Lex Fridman (1:49:03.960)
The main processing bottleneck for this kind of
Ben Goertzel (1:49:06.120)
proto AGI architecture is not multiplying matrices.
Lex Fridman (1:49:09.840)
And for that reason, GPUs, which are really good
Ben Goertzel (1:49:13.280)
at multiplying matrices, don't apply as well.
Lex Fridman (1:49:17.520)
There are frameworks like Gunrock and others
Ben Goertzel (1:49:20.240)
that try to boil down graph processing
Lex Fridman (1:49:22.160)
to matrix operations, and they're cool,
Lex Fridman (1:49:24.640)
but you're still putting a square peg
Lex Fridman (1:49:26.160)
into a round hole in a certain way.
Ben Goertzel (1:49:28.800)
The same is true, I mean, current quantum machine learning,
Lex Fridman (1:49:32.760)
which is very cool.
Ben Goertzel (1:49:34.240)
It's also all about how to get matrix and vector operations
Lex Fridman (1:49:37.320)
in quantum mechanics, and I see why that's natural to do.
Ben Goertzel (1:49:41.280)
I mean, quantum mechanics is all unitary matrices
Lex Fridman (1:49:44.240)
and vectors, right?
Ben Goertzel (1:49:45.800)
On the other hand, you could also try
Lex Fridman (1:49:48.040)
to make graph centric quantum computers,
Ben Goertzel (1:49:50.760)
which I think is where things will go.
Lex Fridman (1:49:54.400)
And then we can have, then we can make,
Ben Goertzel (1:49:57.080)
like take the open cog implementation layer,
Lex Fridman (1:50:00.120)
implement it in a collapsed state inside a quantum computer.
Lex Fridman (1:50:04.000)
But that may be the singularity squared, right?
Lex Fridman (1:50:06.480)
I'm not sure we need that to get to human level.
Ben Goertzel (1:50:12.360)
That's already beyond the first singularity.
Lex Fridman (1:50:14.680)
But can we just go back to open cog?
Ben Goertzel (1:50:17.640)
Yeah, and the hypergraph and open cog.
Lex Fridman (1:50:20.040)
That's the software framework, right?
Lex Fridman (1:50:21.640)
So the next thing is our cognitive architecture
Lex Fridman (1:50:25.440)
tells us particular algorithms to put there.
Ben Goertzel (1:50:27.960)
Got it.
Lex Fridman (1:50:28.800)
Can we backtrack on the kind of, is this graph designed,
Ben Goertzel (1:50:33.720)
is it in general supposed to be sparse
Lex Fridman (1:50:37.680)
and the operations constantly grow and change the graph?
Ben Goertzel (1:50:40.640)
Yeah, the graph is sparse.
Lex Fridman (1:50:42.320)
But is it constantly adding links and so on?
Ben Goertzel (1:50:45.040)
It is a self modifying hypergraph.
Lex Fridman (1:50:47.200)
So it's not, so the write and read operations
Ben Goertzel (1:50:49.800)
you're referring to, this isn't just a fixed graph
Lex Fridman (1:50:53.040)
to which you change the way, it's a constantly growing graph.
Ben Goertzel (1:50:55.840)
Yeah, that's true.
Lex Fridman (1:50:58.000)
So it is different model than,
Ben Goertzel (1:51:03.000)
say current deep neural nets
Lex Fridman (1:51:04.680)
and have a fixed neural architecture
Lex Fridman (1:51:06.840)
and you're updating the weights.
Lex Fridman (1:51:08.600)
Although there have been like cascade correlational
Ben Goertzel (1:51:10.880)
neural net architectures that grow new nodes and links,
Lex Fridman (1:51:13.920)
but the most common neural architectures now
Ben Goertzel (1:51:16.640)
have a fixed neural architecture,
Lex Fridman (1:51:17.960)
you're updating the weights.
Lex Fridman (1:51:19.080)
And then open cog, you can update the weights
Lex Fridman (1:51:22.520)
and that certainly happens a lot,
Lex Fridman (1:51:24.760)
but adding new nodes, adding new links,
Lex Fridman (1:51:28.200)
removing nodes and links is an equally critical part
Ben Goertzel (1:51:30.720)
of the system's operations.
Lex Fridman (1:51:32.160)
Got it.
Lex Fridman (1:51:33.000)
So now when you start to add these cognitive algorithms
Lex Fridman (1:51:37.040)
on top of this open cog architecture,
Lex Fridman (1:51:39.840)
what does that look like?
Lex Fridman (1:51:41.280)
Yeah, so within this framework then,
Ben Goertzel (1:51:44.800)
creating a cognitive architecture is basically two things.
Lex Fridman (1:51:48.040)
It's choosing what type system you wanna put
Ben Goertzel (1:51:52.080)
on the nodes and links in the hypergraph,
Lex Fridman (1:51:53.800)
what types of nodes and links you want.
Lex Fridman (1:51:56.120)
And then it's choosing what collection of agents,
Lex Fridman (1:52:01.000)
what collection of AI algorithms or processes
Ben Goertzel (1:52:04.640)
are gonna run to operate on this hypergraph.
Lex Fridman (1:52:08.040)
And of course those two decisions
Ben Goertzel (1:52:10.520)
are closely connected to each other.
Lex Fridman (1:52:13.920)
So in terms of the type system,
Ben Goertzel (1:52:17.480)
there are some links that are more neural net like,
Lex Fridman (1:52:19.920)
they're just like have weights to get updated
Ben Goertzel (1:52:22.360)
by heavy and learning and activation spreads along them.
Lex Fridman (1:52:26.000)
There are other links that are more logic like
Lex Fridman (1:52:29.080)
and nodes that are more logic like.
Lex Fridman (1:52:30.520)
So you could have a variable node
Lex Fridman (1:52:32.240)
and you can have a node representing a universal
Lex Fridman (1:52:34.240)
or existential quantifier as in predicate logic
Ben Goertzel (1:52:37.680)
or term logic.
Lex Fridman (1:52:39.160)
So you can have logic like nodes and links,
Ben Goertzel (1:52:42.080)
or you can have neural like nodes and links.
Lex Fridman (1:52:44.440)
You can also have procedure like nodes and links
Ben Goertzel (1:52:47.400)
as in say a combinatorial logic or Lambda calculus
Lex Fridman (1:52:51.960)
representing programs.
Lex Fridman (1:52:53.680)
So you can have nodes and links representing
Lex Fridman (1:52:56.520)
many different types of semantics,
Ben Goertzel (1:52:58.640)
which means you could make a horrible ugly mess
Lex Fridman (1:53:00.840)
or you could make a system
Ben Goertzel (1:53:02.800)
where these different types of knowledge
Lex Fridman (1:53:04.280)
all interpenetrate and synergize
Lex Fridman (1:53:06.840)
with each other beautifully, right?
Lex Fridman (1:53:08.960)
So the hypergraph can contain programs.
Ben Goertzel (1:53:12.800)
Yeah, it can contain programs,
Lex Fridman (1:53:14.440)
although in the current version,
Ben Goertzel (1:53:17.960)
it is a very inefficient way
Lex Fridman (1:53:19.760)
to guide the execution of programs,
Ben Goertzel (1:53:21.960)
which is one thing that we are aiming to resolve
Lex Fridman (1:53:25.000)
with our rewrite of the system now.
Lex Fridman (1:53:27.520)
So what to you is the most beautiful aspect of OpenCog?
Lex Fridman (1:53:32.720)
Just to you personally,
Ben Goertzel (1:53:34.600)
some aspect that captivates your imagination
Lex Fridman (1:53:38.080)
from beauty or power?
Lex Fridman (1:53:42.000)
What fascinates me is finding a common representation
Lex Fridman (1:53:48.320)
that underlies abstract, declarative knowledge
Lex Fridman (1:53:53.320)
and sensory knowledge and movement knowledge
Lex Fridman (1:53:57.320)
and procedural knowledge and episodic knowledge,
Ben Goertzel (1:54:00.760)
finding the right level of representation
Lex Fridman (1:54:03.960)
where all these types of knowledge are stored
Ben Goertzel (1:54:06.560)
in a sort of universal and interconvertible
Lex Fridman (1:54:10.560)
yet practically manipulable way, right?
Lex Fridman (1:54:13.440)
So to me, that's the core,
Lex Fridman (1:54:16.840)
because once you've done that,
Ben Goertzel (1:54:18.640)
then the different learning algorithms
Lex Fridman (1:54:20.800)
can help each other out. Like what you want is,
Ben Goertzel (1:54:23.640)
if you have a logic engine
Lex Fridman (1:54:25.120)
that helps with declarative knowledge
Lex Fridman (1:54:26.840)
and you have a deep neural net
Lex Fridman (1:54:28.040)
that gathers perceptual knowledge,
Lex Fridman (1:54:29.960)
and you have, say, an evolutionary learning system
Lex Fridman (1:54:32.400)
that learns procedures,
Ben Goertzel (1:54:34.120)
you want these to not only interact
Lex Fridman (1:54:36.600)
on the level of sharing results
Lex Fridman (1:54:38.880)
and passing inputs and outputs to each other,
Lex Fridman (1:54:41.120)
you want the logic engine, when it gets stuck,
Ben Goertzel (1:54:43.680)
to be able to share its intermediate state
Lex Fridman (1:54:46.240)
with the neural net and with the evolutionary system
Lex Fridman (1:54:49.360)
and with the evolutionary learning algorithm
Lex Fridman (1:54:52.240)
so that they can help each other out of bottlenecks
Lex Fridman (1:54:55.440)
and help each other solve combinatorial explosions
Lex Fridman (1:54:58.320)
by intervening inside each other's cognitive processes.
Lex Fridman (1:55:02.040)
But that can only be done
Lex Fridman (1:55:03.520)
if the intermediate state of a logic engine,
Ben Goertzel (1:55:05.960)
the evolutionary learning engine,
Lex Fridman (1:55:07.400)
and a deep neural net are represented in the same form.
Lex Fridman (1:55:11.120)
And that's what we figured out how to do
Lex Fridman (1:55:13.120)
by putting the right type system
Ben Goertzel (1:55:14.800)
on top of this weighted labeled hypergraph.
Lex Fridman (1:55:17.040)
So is there, can you maybe elaborate
Ben Goertzel (1:55:19.680)
on what are the different characteristics
Lex Fridman (1:55:21.880)
of a type system that can coexist
Ben Goertzel (1:55:26.520)
amongst all these different kinds of knowledge
Lex Fridman (1:55:28.760)
that needs to be represented?
Lex Fridman (1:55:30.080)
And is, I mean, like, is it hierarchical?
Lex Fridman (1:55:34.280)
Just any kind of insights you can give
Lex Fridman (1:55:36.720)
on that kind of type system?
Lex Fridman (1:55:37.840)
Yeah, yeah, so this gets very nitty gritty
Lex Fridman (1:55:41.680)
and mathematical, of course,
Lex Fridman (1:55:44.000)
but one key part is switching
Ben Goertzel (1:55:47.200)
from predicate logic to term logic.
Lex Fridman (1:55:50.440)
What is predicate logic?
Lex Fridman (1:55:51.640)
What is term logic?
Lex Fridman (1:55:53.200)
So term logic was invented by Aristotle,
Ben Goertzel (1:55:56.080)
or at least that's the oldest recollection we have of it.
Lex Fridman (1:56:01.320)
But term logic breaks down basic logic
Ben Goertzel (1:56:05.280)
into basically simple links between nodes,
Lex Fridman (1:56:07.480)
like an inheritance link between node A and node B.
Lex Fridman (1:56:12.480)
So in term logic, the basic deduction operation
Lex Fridman (1:56:16.280)
is A implies B, B implies C, therefore A implies C.
Ben Goertzel (1:56:21.080)
Whereas in predicate logic,
Lex Fridman (1:56:22.600)
the basic operation is modus ponens,
Ben Goertzel (1:56:24.520)
like A implies B, therefore B.
Lex Fridman (1:56:27.680)
So it's a slightly different way of breaking down logic,
Lex Fridman (1:56:31.440)
but by breaking down logic into term logic,
Lex Fridman (1:56:35.320)
you get a nice way of breaking logic down
Ben Goertzel (1:56:37.440)
into nodes and links.
Lex Fridman (1:56:40.120)
So your concepts can become nodes,
Ben Goertzel (1:56:42.960)
the logical relations become links.
Lex Fridman (1:56:45.200)
And so then inference is like,
Lex Fridman (1:56:46.640)
so if this link is A implies B,
Lex Fridman (1:56:48.720)
this link is B implies C,
Ben Goertzel (1:56:50.840)
then deduction builds a link A implies C.
Lex Fridman (1:56:53.360)
And your probabilistic algorithm
Ben Goertzel (1:56:54.920)
can assign a certain weight there.
Lex Fridman (1:56:57.440)
Now, you may also have like a Hebbian neural link
Ben Goertzel (1:57:00.040)
from A to C, which is the degree to which thinking,
Lex Fridman (1:57:03.600)
the degree to which A being the focus of attention
Lex Fridman (1:57:06.640)
should make B the focus of attention, right?
Lex Fridman (1:57:09.080)
So you could have then a neural link
Lex Fridman (1:57:10.880)
and you could have a symbolic,
Lex Fridman (1:57:13.720)
like logical inheritance link in your term logic.
Lex Fridman (1:57:17.000)
And they have separate meaning,
Lex Fridman (1:57:19.520)
but they could be used to guide each other as well.
Ben Goertzel (1:57:22.960)
Like if there's a large amount of neural weight
Lex Fridman (1:57:26.720)
on the link between A and B,
Ben Goertzel (1:57:28.400)
that may direct your logic engine to think about,
Lex Fridman (1:57:30.440)
well, what is the relation?
Lex Fridman (1:57:31.320)
Are they similar?
Lex Fridman (1:57:32.160)
Is there an inheritance relation?
Lex Fridman (1:57:33.880)
Are they similar in some context?
Lex Fridman (1:57:37.400)
On the other hand, if there's a logical relation
Ben Goertzel (1:57:39.920)
between A and B, that may direct your neural component
Lex Fridman (1:57:43.360)
to think, well, when I'm thinking about A,
Lex Fridman (1:57:45.520)
should I be directing some attention to B also?
Lex Fridman (1:57:48.240)
Because there's a logical relation.
Lex Fridman (1:57:50.160)
So in terms of logic,
Lex Fridman (1:57:53.000)
there's a lot of thought that went into
Lex Fridman (1:57:54.320)
how do you break down logic relations,
Lex Fridman (1:57:58.280)
including basic sort of propositional logic relations
Ben Goertzel (1:58:02.320)
as Aristotelian term logic deals with,
Lex Fridman (1:58:04.160)
and then quantifier logic relations also.
Lex Fridman (1:58:07.080)
How do you break those down elegantly into a hypergraph?
Lex Fridman (1:58:10.920)
Because you, I mean, you can boil logic expression
Ben Goertzel (1:58:13.480)
into a graph in many different ways.
Lex Fridman (1:58:14.840)
Many of them are very ugly, right?
Ben Goertzel (1:58:16.680)
We tried to find elegant ways
Lex Fridman (1:58:19.200)
of sort of hierarchically breaking down
Ben Goertzel (1:58:22.600)
complex logic expression into nodes and links.
Lex Fridman (1:58:26.880)
So that if you have say different nodes representing,
Ben Goertzel (1:58:31.400)
Ben, AI, Lex, interview or whatever,
Lex Fridman (1:58:34.200)
the logic relations between those things
Ben Goertzel (1:58:36.800)
are compact in the node and link representation.
Lex Fridman (1:58:40.480)
So that when you have a neural net acting
Ben Goertzel (1:58:42.080)
on the same nodes and links,
Lex Fridman (1:58:43.960)
the neural net and the logic engine
Ben Goertzel (1:58:45.760)
can sort of interoperate with each other.
Lex Fridman (1:58:48.240)
And also interpretable by humans.
Lex Fridman (1:58:49.920)
Is that an important?
Lex Fridman (1:58:51.400)
That's tough.
Ben Goertzel (1:58:52.240)
Yeah, in simple cases, it's interpretable by humans.
Lex Fridman (1:58:54.600)
But honestly, I would say logic systems
Ben Goertzel (1:58:59.600)
I would say logic systems give more potential
Lex Fridman (1:59:05.440)
for transparency and comprehensibility
Ben Goertzel (1:59:09.800)
than neural net systems,
Lex Fridman (1:59:11.640)
but you still have to work at it.
Ben Goertzel (1:59:12.840)
Because I mean, if I show you a predicate logic proposition
Lex Fridman (1:59:16.680)
with like 500 nested universal and existential quantifiers
Lex Fridman (1:59:20.080)
and 217 variables, that's no more comprehensible
Lex Fridman (1:59:23.680)
than the weight metrics of a neural network, right?
Lex Fridman (1:59:26.560)
So I'd say the logic expressions
Lex Fridman (1:59:28.560)
that AI learns from its experience
Ben Goertzel (1:59:30.920)
are mostly totally opaque to human beings
Lex Fridman (1:59:33.440)
and maybe even harder to understand than neural net.
Ben Goertzel (1:59:36.200)
Because I mean, when you have multiple
Lex Fridman (1:59:37.440)
nested quantifier bindings,
Ben Goertzel (1:59:38.960)
it's a very high level of abstraction.
Lex Fridman (1:59:41.520)
There is a difference though,
Ben Goertzel (1:59:42.680)
in that within logic, it's a little more straightforward
Lex Fridman (1:59:46.880)
to pose the problem of like normalize this
Lex Fridman (1:59:49.120)
and boil this down to a certain form.
Lex Fridman (1:59:51.080)
I mean, you can do that in neural nets too.
Ben Goertzel (1:59:52.720)
Like you can distill a neural net to a simpler form,
Lex Fridman (1:59:55.680)
but that's more often done to make a neural net
Ben Goertzel (1:59:57.280)
that'll run on an embedded device or something.
Lex Fridman (1:59:59.720)
It's harder to distill a net to a comprehensible form
Lex Fridman (20:00.900)
on punch cards when I was in middle school, right?
Lex Fridman (20:04.220)
Because he was doing, I guess, analysis of demographic
Lex Fridman (20:07.460)
and sociology data.
Lex Fridman (20:09.580)
So yes, certainly that gave me a head start
Lex Fridman (20:14.780)
and a push towards science beyond what would have been
Lex Fridman (20:17.220)
the case with many, many different situations.
Lex Fridman (20:19.700)
When did you first fall in love with AI?
Lex Fridman (20:22.220)
Is it the programming side of Fortran?
Ben Goertzel (20:24.700)
Is it maybe the sociology psychology
Lex Fridman (20:27.260)
that you picked up from your dad?
Lex Fridman (20:28.300)
Or is it the quantum mechanics?
Lex Fridman (20:29.140)
I fell in love with AI when I was probably three years old
Ben Goertzel (20:30.660)
when I saw a robot on Star Trek.
Lex Fridman (20:32.580)
It was turning around in a circle going,
Ben Goertzel (20:34.620)
error, error, error, error,
Lex Fridman (20:36.660)
because Spock and Kirk had tricked it
Ben Goertzel (20:39.540)
into a mechanical breakdown by presenting it
Lex Fridman (20:41.300)
with a logical paradox.
Lex Fridman (20:42.900)
And I was just like, well, this makes no sense.
Lex Fridman (20:45.660)
This AI is very, very smart.
Ben Goertzel (20:47.540)
It's been traveling all around the universe,
Lex Fridman (20:49.620)
but these people could trick it
Ben Goertzel (20:50.980)
with a simple logical paradox.
Lex Fridman (20:52.660)
Like why, if the human brain can get beyond that paradox,
Lex Fridman (20:57.020)
why can't this AI?
Lex Fridman (20:59.460)
So I felt the screenwriters of Star Trek
Ben Goertzel (21:03.140)
had misunderstood the nature of intelligence.
Lex Fridman (21:06.060)
And I complained to my dad about it,
Lex Fridman (21:07.580)
and he wasn't gonna say anything one way or the other.
Lex Fridman (21:12.220)
But before I was born, when my dad was at Antioch College
Ben Goertzel (21:18.460)
in the middle of the US,
Lex Fridman (21:20.860)
he led a protest movement called SLAM,
Ben Goertzel (21:25.860)
Student League Against Mortality.
Lex Fridman (21:27.460)
They were protesting against death,
Ben Goertzel (21:28.980)
wandering across the campus.
Lex Fridman (21:31.500)
So he was into some futuristic things even back then,
Lex Fridman (21:35.900)
but whether AI could confront logical paradoxes or not,
Lex Fridman (21:40.220)
he didn't know.
Lex Fridman (21:41.220)
But when I, 10 years after that or something,
Lex Fridman (21:44.780)
I discovered Douglas Hofstadter's book,
Ben Goertzel (21:46.980)
Gordalesh or Bach, and that was sort of to the same point of AI
Lex Fridman (21:51.100)
and paradox and logic, right?
Ben Goertzel (21:52.620)
Because he was over and over
Lex Fridman (21:54.460)
with Gordal's incompleteness theorem,
Lex Fridman (21:56.180)
and can an AI really fully model itself reflexively
Lex Fridman (22:00.500)
or does that lead you into some paradox?
Ben Goertzel (22:02.820)
Can the human mind truly model itself reflexively
Lex Fridman (22:05.260)
or does that lead you into some paradox?
Lex Fridman (22:07.500)
So I think that book, Gordalesh or Bach,
Lex Fridman (22:10.660)
which I think I read when it first came out,
Ben Goertzel (22:13.460)
I would have been 12 years old or something.
Lex Fridman (22:14.980)
I remember it was like 16 hour day.
Ben Goertzel (22:17.100)
I read it cover to cover and then reread it.
Lex Fridman (22:19.780)
I reread it after that,
Ben Goertzel (22:21.260)
because there was a lot of weird things
Lex Fridman (22:22.380)
with little formal systems in there
Ben Goertzel (22:24.380)
that were hard for me at the time.
Lex Fridman (22:25.660)
But that was the first book I read
Ben Goertzel (22:27.980)
that gave me a feeling for AI as like a practical academic
Lex Fridman (22:34.420)
or engineering discipline that people were working in.
Ben Goertzel (22:37.380)
Because before I read Gordalesh or Bach,
Lex Fridman (22:40.060)
I was into AI from the point of view of a science fiction fan.
Lex Fridman (22:43.980)
And I had the idea, well, it may be a long time
Lex Fridman (22:47.460)
before we can achieve immortality in superhuman AGI.
Lex Fridman (22:50.420)
So I should figure out how to build a spacecraft
Lex Fridman (22:54.380)
traveling close to the speed of light, go far away,
Ben Goertzel (22:57.060)
then come back to the earth in a million years
Lex Fridman (22:58.780)
when technology is more advanced
Lex Fridman (23:00.220)
and we can build these things.
Lex Fridman (23:01.700)
Reading Gordalesh or Bach,
Ben Goertzel (23:03.580)
while it didn't all ring true to me, a lot of it did,
Lex Fridman (23:06.580)
but I could see like there are smart people right now
Ben Goertzel (23:09.860)
at various universities around me
Lex Fridman (23:11.580)
who are actually trying to work on building
Lex Fridman (23:15.420)
what I would now call AGI,
Lex Fridman (23:16.980)
although Hofstadter didn't call it that.
Lex Fridman (23:19.020)
So really it was when I read that book,
Lex Fridman (23:21.100)
which would have been probably middle school,
Ben Goertzel (23:23.540)
that then I started to think,
Lex Fridman (23:24.820)
well, this is something that I could practically work on.
Ben Goertzel (23:29.020)
Yeah, as opposed to flying away and waiting it out,
Lex Fridman (23:31.660)
you can actually be one of the people
Ben Goertzel (23:33.500)
that actually builds the system.
Lex Fridman (23:34.580)
Yeah, exactly.
Lex Fridman (23:35.420)
And if you think about, I mean,
Lex Fridman (23:36.740)
I was interested in what we'd now call nanotechnology
Lex Fridman (23:40.700)
and in the human immortality and time travel,
Lex Fridman (23:44.820)
all the same cool things as every other,
Ben Goertzel (23:46.940)
like science fiction loving kid.
Lex Fridman (23:49.260)
But AI seemed like if Hofstadter was right,
Ben Goertzel (23:52.700)
you just figure out the right program,
Lex Fridman (23:54.180)
sit there and type it.
Ben Goertzel (23:55.060)
Like you don't need to spin stars into weird configurations
Lex Fridman (23:59.620)
or get government approval to cut people up
Lex Fridman (24:02.620)
and fiddle with their DNA or something, right?
Lex Fridman (24:05.020)
It's just programming.
Lex Fridman (24:06.180)
And then of course that can achieve anything else.
Lex Fridman (24:10.700)
There's another book from back then,
Ben Goertzel (24:12.220)
which was by Gerald Feinbaum,
Lex Fridman (24:17.060)
who was a physicist at Princeton.
Lex Fridman (24:21.580)
And that was the Prometheus Project.
Lex Fridman (24:24.580)
And this book was written in the late 1960s,
Ben Goertzel (24:26.700)
though I encountered it in the mid 70s.
Lex Fridman (24:28.780)
But what this book said is in the next few decades,
Ben Goertzel (24:30.940)
humanity is gonna create superhuman thinking machines,
Lex Fridman (24:34.500)
molecular nanotechnology and human immortality.
Lex Fridman (24:37.460)
And then the challenge we'll have is what to do with it.
Lex Fridman (24:41.140)
Do we use it to expand human consciousness
Lex Fridman (24:43.020)
in a positive direction?
Lex Fridman (24:44.500)
Or do we use it just to further vapid consumerism?
Lex Fridman (24:49.860)
And what he proposed was that the UN
Lex Fridman (24:51.820)
should do a survey on this.
Lex Fridman (24:53.460)
And the UN should send people out to every little village
Lex Fridman (24:56.460)
in remotest Africa or South America
Lex Fridman (24:58.940)
and explain to everyone what technology
Lex Fridman (25:01.300)
was gonna bring the next few decades
Lex Fridman (25:03.020)
and the choice that we had about how to use it.
Lex Fridman (25:05.020)
And let everyone on the whole planet vote
Ben Goertzel (25:07.780)
about whether we should develop super AI nanotechnology
Lex Fridman (25:11.740)
and immortality for expanded consciousness
Ben Goertzel (25:15.900)
or for rampant consumerism.
Lex Fridman (25:18.220)
And needless to say, that didn't quite happen.
Lex Fridman (25:22.060)
And I think this guy died in the mid 80s,
Lex Fridman (25:24.180)
so we didn't even see his ideas start
Ben Goertzel (25:25.900)
to become more mainstream.
Lex Fridman (25:28.220)
But it's interesting, many of the themes I'm engaged with now
Ben Goertzel (25:31.620)
from AGI and immortality,
Lex Fridman (25:33.340)
even to trying to democratize technology
Ben Goertzel (25:36.140)
as I've been pushing forward with Singularity,
Lex Fridman (25:38.100)
my work in the blockchain world,
Ben Goertzel (25:40.020)
many of these themes were there in Feinbaum's book
Lex Fridman (25:43.620)
in the late 60s even.
Lex Fridman (25:47.940)
And of course, Valentin Turchin, a Russian writer
Lex Fridman (25:52.220)
and a great Russian physicist who I got to know
Ben Goertzel (25:55.860)
when we both lived in New York in the late 90s
Lex Fridman (25:59.060)
and early aughts.
Ben Goertzel (25:59.900)
I mean, he had a book in the late 60s in Russia,
Lex Fridman (26:03.380)
which was the phenomenon of science,
Ben Goertzel (26:05.780)
which laid out all these same things as well.
Lex Fridman (26:10.220)
And Val died in, I don't remember,
Ben Goertzel (26:12.740)
2004 or five or something of Parkinson'sism.
Lex Fridman (26:15.420)
So yeah, it's easy for people to lose track now
Ben Goertzel (26:20.780)
of the fact that the futurist and Singularitarian
Lex Fridman (26:25.940)
advanced technology ideas that are now almost mainstream
Ben Goertzel (26:29.740)
are on TV all the time.
Lex Fridman (26:30.900)
I mean, these are not that new, right?
Ben Goertzel (26:34.100)
They're sort of new in the history of the human species,
Lex Fridman (26:37.100)
but I mean, these were all around in fairly mature form
Ben Goertzel (26:41.100)
in the middle of the last century,
Lex Fridman (26:43.660)
were written about quite articulately
Ben Goertzel (26:45.500)
by fairly mainstream people
Lex Fridman (26:47.340)
who were professors at top universities.
Ben Goertzel (26:50.140)
It's just until the enabling technologies
Lex Fridman (26:52.940)
got to a certain point, then you couldn't make it real.
Lex Fridman (26:57.940)
And even in the 70s, I was sort of seeing that
Lex Fridman (27:02.820)
and living through it, right?
Ben Goertzel (27:04.740)
From Star Trek to Douglas Hofstadter,
Lex Fridman (27:07.900)
things were getting very, very practical
Ben Goertzel (27:09.660)
from the late 60s to the late 70s.
Lex Fridman (27:11.980)
And the first computer I bought,
Ben Goertzel (27:15.020)
you could only program with hexadecimal machine code
Lex Fridman (27:17.580)
and you had to solder it together.
Lex Fridman (27:19.380)
And then like a few years later, there's punch cards.
Lex Fridman (27:23.420)
And a few years later, you could get like Atari 400
Lex Fridman (27:27.220)
and Commodore VIC 20, and you could type on the keyboard
Lex Fridman (27:30.300)
and program in higher level languages
Ben Goertzel (27:32.820)
alongside the assembly language.
Lex Fridman (27:34.660)
So these ideas have been building up a while.
Lex Fridman (27:38.700)
And I guess my generation got to feel them build up,
Lex Fridman (27:42.980)
which is different than people coming into the field now
Ben Goertzel (27:46.380)
for whom these things have just been part of the ambience
Lex Fridman (27:50.300)
of culture for their whole career
Ben Goertzel (27:52.180)
or even their whole life.
Lex Fridman (27:54.140)
Well, it's fascinating to think about there being all
Ben Goertzel (27:57.260)
of these ideas kind of swimming, almost with the noise
Lex Fridman (28:01.540)
all around the world, all the different generations,
Lex Fridman (28:04.380)
and then some kind of nonlinear thing happens
Lex Fridman (28:07.900)
where they percolate up
Lex Fridman (28:09.380)
and capture the imagination of the mainstream.
Lex Fridman (28:12.420)
And that seems to be what's happening with AI now.
Ben Goertzel (28:14.780)
I mean, Nietzsche, who you mentioned had the idea
Lex Fridman (28:16.580)
of the Superman, right?
Lex Fridman (28:18.260)
But he didn't understand enough about technology
Lex Fridman (28:21.580)
to think you could physically engineer a Superman
Ben Goertzel (28:24.860)
by piecing together molecules in a certain way.
Lex Fridman (28:28.180)
He was a bit vague about how the Superman would appear,
Lex Fridman (28:33.620)
but he was quite deep at thinking
Lex Fridman (28:35.820)
about what the state of consciousness
Lex Fridman (28:37.780)
and the mode of cognition of a Superman would be.
Lex Fridman (28:42.420)
He was a very astute analyst of how the human mind
Ben Goertzel (28:47.820)
constructs the illusion of a self,
Lex Fridman (28:49.420)
how it constructs the illusion of free will,
Lex Fridman (28:52.140)
how it constructs values like good and evil
Lex Fridman (28:56.660)
out of its own desire to maintain
Lex Fridman (28:59.780)
and advance its own organism.
Lex Fridman (29:01.420)
He understood a lot about how human minds work.
Ben Goertzel (29:04.020)
Then he understood a lot
Lex Fridman (29:05.660)
about how post human minds would work.
Ben Goertzel (29:07.620)
I mean, the Superman was supposed to be a mind
Lex Fridman (29:10.260)
that would basically have complete root access
Ben Goertzel (29:13.300)
to its own brain and consciousness
Lex Fridman (29:16.060)
and be able to architect its own value system
Lex Fridman (29:19.620)
and inspect and fine tune all of its own biases.
Lex Fridman (29:24.300)
So that's a lot of powerful thinking there,
Ben Goertzel (29:27.340)
which then fed in and sort of seeded
Lex Fridman (29:29.340)
all of postmodern continental philosophy
Lex Fridman (29:32.180)
and all sorts of things have been very valuable
Lex Fridman (29:35.540)
in development of culture and indirectly even of technology.
Lex Fridman (29:39.740)
But of course, without the technology there,
Lex Fridman (29:42.140)
it was all some quite abstract thinking.
Lex Fridman (29:44.860)
So now we're at a time in history
Lex Fridman (29:46.940)
when a lot of these ideas can be made real,
Lex Fridman (29:51.740)
which is amazing and scary, right?
Lex Fridman (29:54.300)
It's kind of interesting to think,
Lex Fridman (29:56.020)
what do you think Nietzsche would do
Lex Fridman (29:57.180)
if he was born a century later or transported through time?
Ben Goertzel (2:00:03.440)
than it is to simplify a logic expression
Lex Fridman (2:00:05.640)
to a comprehensible form, but it doesn't come for free.
Ben Goertzel (2:00:08.600)
Like what's in the AI's mind is incomprehensible
Lex Fridman (2:00:13.040)
to a human unless you do some special work
Ben Goertzel (2:00:15.720)
to make it comprehensible.
Lex Fridman (2:00:16.880)
So on the procedural side, there's some different
Lex Fridman (2:00:20.400)
and sort of interesting voodoo there.
Lex Fridman (2:00:23.000)
I mean, if you're familiar in computer science,
Ben Goertzel (2:00:25.800)
there's something called the Curry Howard correspondence,
Lex Fridman (2:00:27.800)
which is a one to one mapping between proofs and programs.
Lex Fridman (2:00:30.920)
So every program can be mapped into a proof.
Lex Fridman (2:00:33.520)
Every proof can be mapped into a program.
Ben Goertzel (2:00:35.960)
You can model this using category theory
Lex Fridman (2:00:37.800)
and a bunch of nice math,
Lex Fridman (2:00:40.960)
but we wanna make that practical, right?
Lex Fridman (2:00:43.280)
So that if you have an executable program
Ben Goertzel (2:00:46.520)
that like moves the robot's arm or figures out
Lex Fridman (2:00:49.960)
in what order to say things in a dialogue,
Ben Goertzel (2:00:51.840)
that's a procedure represented in OpenCog's hypergraph.
Lex Fridman (2:00:55.840)
But if you wanna reason on how to improve that procedure,
Ben Goertzel (2:01:00.120)
you need to map that procedure into logic
Lex Fridman (2:01:03.080)
using Curry Howard isomorphism.
Lex Fridman (2:01:05.520)
So then the logic engine can reason
Lex Fridman (2:01:09.320)
about how to improve that procedure
Lex Fridman (2:01:11.120)
and then map that back into the procedural representation
Lex Fridman (2:01:14.080)
that is efficient for execution.
Lex Fridman (2:01:16.160)
So again, that comes down to not just
Lex Fridman (2:01:18.800)
can you make your procedure into a bunch of nodes and links?
Ben Goertzel (2:01:21.440)
Cause I mean, that can be done trivially.
Lex Fridman (2:01:23.280)
A C++ compiler has nodes and links inside it.
Lex Fridman (2:01:26.440)
Can you boil down your procedure
Lex Fridman (2:01:27.960)
into a bunch of nodes and links
Ben Goertzel (2:01:29.840)
in a way that's like hierarchically decomposed
Lex Fridman (2:01:32.560)
and simple enough?
Ben Goertzel (2:01:33.680)
It can reason about.
Lex Fridman (2:01:34.520)
Yeah, yeah, that given the resource constraints at hand,
Ben Goertzel (2:01:37.040)
you can map it back and forth to your term logic,
Lex Fridman (2:01:40.920)
like fast enough
Lex Fridman (2:01:42.080)
and without having a bloated logic expression, right?
Lex Fridman (2:01:45.200)
So there's just a lot of,
Ben Goertzel (2:01:48.320)
there's a lot of nitty gritty particulars there,
Lex Fridman (2:01:50.360)
but by the same token, if you ask a chip designer,
Lex Fridman (2:01:54.520)
like how do you make the Intel I7 chip so good?
Lex Fridman (2:01:58.560)
There's a long list of technical answers there,
Lex Fridman (2:02:02.560)
which will take a while to go through, right?
Lex Fridman (2:02:04.800)
And this has been decades of work.
Ben Goertzel (2:02:06.640)
I mean, the first AI system of this nature I tried to build
Lex Fridman (2:02:10.880)
was called WebMind in the mid 1990s.
Lex Fridman (2:02:13.440)
And we had a big graph,
Lex Fridman (2:02:15.600)
a big graph operating in RAM implemented with Java 1.1,
Ben Goertzel (2:02:18.880)
which was a terrible, terrible implementation idea.
Lex Fridman (2:02:21.800)
And then each node had its own processing.
Lex Fridman (2:02:25.960)
So like that there,
Lex Fridman (2:02:27.440)
the core loop looped through all nodes in the network
Lex Fridman (2:02:29.560)
and let each node enact what its little thing was doing.
Lex Fridman (2:02:32.920)
And we had logic and neural nets in there,
Lex Fridman (2:02:35.880)
but an evolutionary learning,
Lex Fridman (2:02:38.400)
but we hadn't done enough of the math
Ben Goertzel (2:02:40.760)
to get them to operate together very cleanly.
Lex Fridman (2:02:43.400)
So it was really, it was quite a horrible mess.
Lex Fridman (2:02:46.240)
So as well as shifting an implementation
Lex Fridman (2:02:49.400)
where the graph is its own object
Lex Fridman (2:02:51.840)
and the agents are separately scheduled,
Lex Fridman (2:02:54.720)
we've also done a lot of work
Lex Fridman (2:02:56.800)
on how do you represent programs?
Lex Fridman (2:02:58.400)
How do you represent procedures?
Ben Goertzel (2:03:00.800)
You know, how do you represent genotypes for evolution
Lex Fridman (2:03:03.640)
in a way that the interoperability
Ben Goertzel (2:03:06.640)
between the different types of learning
Lex Fridman (2:03:09.000)
associated with these different types of knowledge
Lex Fridman (2:03:11.720)
actually works?
Lex Fridman (2:03:13.040)
And that's been quite difficult.
Ben Goertzel (2:03:14.960)
It's taken decades and it's totally off to the side
Lex Fridman (2:03:18.600)
of what the commercial mainstream of the AI field is doing,
Ben Goertzel (2:03:23.080)
which isn't thinking about representation at all really.
Lex Fridman (2:03:27.640)
Although you could see like in the DNC,
Ben Goertzel (2:03:30.800)
they had to think a little bit about
Lex Fridman (2:03:32.320)
how do you make representation of a map
Ben Goertzel (2:03:33.880)
in this memory matrix work together
Lex Fridman (2:03:36.680)
with the representation needed
Ben Goertzel (2:03:38.160)
for say visual pattern recognition
Lex Fridman (2:03:40.240)
in the hierarchical neural network.
Lex Fridman (2:03:42.120)
But I would say we have taken that direction
Lex Fridman (2:03:45.120)
of taking the types of knowledge you need
Ben Goertzel (2:03:47.920)
for different types of learning,
Lex Fridman (2:03:49.120)
like declarative, procedural, attentional,
Lex Fridman (2:03:52.040)
and how do you make these types of knowledge represent
Lex Fridman (2:03:55.520)
in a way that allows cross learning
Ben Goertzel (2:03:58.160)
across these different types of memory.
Lex Fridman (2:04:00.200)
We've been prototyping and experimenting with this
Ben Goertzel (2:04:03.920)
within OpenCog and before that WebMind
Lex Fridman (2:04:07.560)
since the mid 1990s.
Ben Goertzel (2:04:10.640)
Now, disappointingly to all of us,
Lex Fridman (2:04:13.840)
this has not yet been cashed out in an AGI system, right?
Ben Goertzel (2:04:18.400)
I mean, we've used this system
Lex Fridman (2:04:20.640)
within our consulting business.
Lex Fridman (2:04:22.440)
So we've built natural language processing
Lex Fridman (2:04:24.320)
and robot control and financial analysis.
Ben Goertzel (2:04:27.760)
We've built a bunch of sort of vertical market specific
Lex Fridman (2:04:31.160)
proprietary AI projects.
Ben Goertzel (2:04:33.600)
They use OpenCog on the backend,
Lex Fridman (2:04:36.720)
but we haven't, that's not the AGI goal, right?
Ben Goertzel (2:04:39.560)
It's interesting, but it's not the AGI goal.
Lex Fridman (2:04:42.680)
So now what we're looking at with our rebuild of the system.
Ben Goertzel (2:04:48.520)
2.0.
Lex Fridman (2:04:49.360)
Yeah, we're also calling it True AGI.
Lex Fridman (2:04:51.400)
So we're not quite sure what the name is yet.
Lex Fridman (2:04:54.800)
We made a website for trueagi.io,
Lex Fridman (2:04:57.480)
but we haven't put anything on there yet.
Lex Fridman (2:04:59.840)
We may come up with an even better name.
Ben Goertzel (2:05:02.160)
It's kind of like the real AI starting point
Lex Fridman (2:05:04.960)
for your AGI book.
Ben Goertzel (2:05:05.800)
Yeah, but I like True better
Lex Fridman (2:05:06.920)
because True has like, you can be true hearted, right?
Ben Goertzel (2:05:09.760)
You can be true to your girlfriend.
Lex Fridman (2:05:11.040)
So True has a number and it also has logic in it, right?
Ben Goertzel (2:05:15.720)
Because logic is a key part of the system.
Lex Fridman (2:05:18.280)
So yeah, with the True AGI system,
Ben Goertzel (2:05:22.400)
we're sticking with the same basic architecture,
Lex Fridman (2:05:25.400)
but we're trying to build on what we've learned.
Lex Fridman (2:05:29.640)
And one thing we've learned is that,
Lex Fridman (2:05:32.360)
we need type checking among dependent types
Ben Goertzel (2:05:36.920)
to be much faster
Lex Fridman (2:05:38.040)
and among probabilistic dependent types to be much faster.
Lex Fridman (2:05:41.120)
So as it is now,
Lex Fridman (2:05:43.600)
you can have complex types on the nodes and links.
Lex Fridman (2:05:47.120)
But if you wanna put,
Lex Fridman (2:05:48.360)
like if you want types to be first class citizens,
Lex Fridman (2:05:51.280)
so that you can have the types can be variables
Lex Fridman (2:05:53.800)
and then you do type checking
Ben Goertzel (2:05:55.680)
among complex higher order types.
Lex Fridman (2:05:58.040)
You can do that in the system now, but it's very slow.
Ben Goertzel (2:06:00.960)
This is stuff like it's done
Lex Fridman (2:06:02.560)
in cutting edge program languages like Agda or something,
Ben Goertzel (2:06:05.360)
these obscure research languages.
Lex Fridman (2:06:07.400)
On the other hand,
Ben Goertzel (2:06:08.600)
we've been doing a lot tying together deep neural nets
Lex Fridman (2:06:11.240)
with symbolic learning.
Lex Fridman (2:06:12.360)
So we did a project for Cisco, for example,
Lex Fridman (2:06:15.200)
which was on, this was street scene analysis,
Lex Fridman (2:06:17.440)
but they had deep neural models
Lex Fridman (2:06:18.600)
for a bunch of cameras watching street scenes,
Lex Fridman (2:06:21.000)
but they trained a different model for each camera
Lex Fridman (2:06:23.400)
because they couldn't get the transfer learning
Ben Goertzel (2:06:24.840)
to work between camera A and camera B.
Lex Fridman (2:06:27.040)
So we took what came out of all the deep neural models
Ben Goertzel (2:06:29.040)
for the different cameras,
Lex Fridman (2:06:30.400)
we fed it into an open called symbolic representation.
Ben Goertzel (2:06:33.440)
Then we did some pattern mining and some reasoning
Lex Fridman (2:06:36.280)
on what came out of all the different cameras
Ben Goertzel (2:06:38.120)
within the symbolic graph.
Lex Fridman (2:06:39.480)
And that worked well for that application.
Ben Goertzel (2:06:42.040)
I mean, Hugo Latapie from Cisco gave a talk touching on that
Lex Fridman (2:06:45.880)
at last year's AGI conference, it was in Shenzhen.
Ben Goertzel (2:06:48.760)
On the other hand, we learned from there,
Lex Fridman (2:06:51.000)
it was kind of clunky to get the deep neural models
Ben Goertzel (2:06:53.280)
to work well with the symbolic system
Lex Fridman (2:06:55.640)
because we were using torch.
Lex Fridman (2:06:58.560)
And torch keeps a sort of state computation graph,
Lex Fridman (2:07:03.560)
but you needed like real time access
Ben Goertzel (2:07:05.280)
to that computation graph within our hypergraph.
Lex Fridman (2:07:07.640)
And we certainly did it,
Ben Goertzel (2:07:10.640)
Alexey Polopov who leads our St. Petersburg team
Lex Fridman (2:07:13.080)
wrote a great paper on cognitive modules in OpenCog
Ben Goertzel (2:07:16.480)
explaining sort of how do you deal
Lex Fridman (2:07:17.720)
with the torch compute graph inside OpenCog.
Lex Fridman (2:07:19.960)
But in the end we realized like,
Lex Fridman (2:07:22.840)
that just hadn't been one of our design thoughts
Lex Fridman (2:07:25.400)
when we built OpenCog, right?
Lex Fridman (2:07:27.240)
So between wanting really fast dependent type checking
Lex Fridman (2:07:30.680)
and wanting much more efficient interoperation
Lex Fridman (2:07:33.640)
between the computation graphs
Ben Goertzel (2:07:35.160)
of deep neural net frameworks and OpenCog's hypergraph
Lex Fridman (2:07:37.720)
and adding on top of that,
Ben Goertzel (2:07:40.000)
wanting to more effectively run an OpenCog hypergraph
Lex Fridman (2:07:42.480)
distributed across RAM in 10,000 machines,
Ben Goertzel (2:07:45.200)
which is we're doing dozens of machines now,
Lex Fridman (2:07:47.280)
but it's just not, we didn't architect it
Ben Goertzel (2:07:50.720)
with that sort of modern scalability in mind.
Lex Fridman (2:07:53.080)
So these performance requirements are what have driven us
Ben Goertzel (2:07:56.280)
to want to rearchitect the base,
Lex Fridman (2:08:00.520)
but the core AGI paradigm doesn't really change.
Ben Goertzel (2:08:05.320)
Like the mathematics is the same.
Lex Fridman (2:08:07.760)
It's just, we can't scale to the level that we want
Ben Goertzel (2:08:11.440)
in terms of distributed processing
Lex Fridman (2:08:13.880)
or speed of various kinds of processing
Ben Goertzel (2:08:16.280)
with the current infrastructure
Lex Fridman (2:08:19.160)
that was built in the phase 2001 to 2008,
Ben Goertzel (2:08:22.880)
which is hardly shocking.
Lex Fridman (2:08:26.120)
Well, I mean, the three things you mentioned
Ben Goertzel (2:08:27.880)
are really interesting.
Lex Fridman (2:08:28.720)
So what do you think about in terms of interoperability
Lex Fridman (2:08:32.320)
communicating with computational graph of neural networks?
Lex Fridman (2:08:36.320)
What do you think about the representations
Lex Fridman (2:08:38.480)
that neural networks form?
Lex Fridman (2:08:40.680)
They're bad, but there's many ways
Ben Goertzel (2:08:42.920)
that you could deal with that.
Lex Fridman (2:08:44.360)
So I've been wrestling with this a lot
Ben Goertzel (2:08:46.880)
in some work on supervised grammar induction,
Lex Fridman (2:08:49.920)
and I have a simple paper on that.
Ben Goertzel (2:08:52.120)
They'll give it the next AGI conference,
Lex Fridman (2:08:55.400)
online portion of which is next week, actually.
Lex Fridman (2:08:58.200)
What is grammar induction?
Lex Fridman (2:09:00.400)
So this isn't AGI either,
Lex Fridman (2:09:02.560)
but it's sort of on the verge
Lex Fridman (2:09:05.200)
between narrow AI and AGI or something.
Ben Goertzel (2:09:08.280)
Unsupervised grammar induction is the problem.
Lex Fridman (2:09:11.320)
Throw your AI system, a huge body of text,
Lex Fridman (2:09:15.400)
and have it learn the grammar of the language
Lex Fridman (2:09:18.160)
that produced that text.
Lex Fridman (2:09:20.280)
So you're not giving it labeled examples.
Lex Fridman (2:09:22.600)
So you're not giving it like a thousand sentences
Ben Goertzel (2:09:24.440)
where the parses were marked up by graduate students.
Lex Fridman (2:09:27.120)
So it's just got to infer the grammar from the text.
Lex Fridman (2:09:30.280)
It's like the Rosetta Stone, but worse, right?
Lex Fridman (2:09:33.440)
Because you only have the one language,
Lex Fridman (2:09:35.320)
and you have to figure out what is the grammar.
Lex Fridman (2:09:37.160)
So that's not really AGI because,
Lex Fridman (2:09:41.440)
I mean, the way a human learns language is not that, right?
Lex Fridman (2:09:44.360)
I mean, we learn from language that's used in context.
Lex Fridman (2:09:47.720)
So it's a social embodied thing.
Lex Fridman (2:09:49.320)
We see how a given sentence is grounded in observation.
Ben Goertzel (2:09:53.520)
There's an interactive element, I guess.
Lex Fridman (2:09:55.200)
Yeah, yeah, yeah.
Ben Goertzel (2:09:56.520)
On the other hand, so I'm more interested in that.
Lex Fridman (2:10:00.360)
I'm more interested in making an AGI system learn language
Ben Goertzel (2:10:02.960)
from its social and embodied experience.
Lex Fridman (2:10:05.560)
On the other hand, that's also more of a pain to do,
Lex Fridman (2:10:08.240)
and that would lead us into Hanson Robotics
Lex Fridman (2:10:10.640)
and their robotics work I've known much.
Ben Goertzel (2:10:12.080)
We'll talk about it in a few minutes.
Lex Fridman (2:10:14.600)
But just as an intellectual exercise,
Ben Goertzel (2:10:17.120)
as a learning exercise,
Lex Fridman (2:10:18.840)
trying to learn grammar from a corpus
Lex Fridman (2:10:22.480)
is very, very interesting, right?
Lex Fridman (2:10:24.560)
And that's been a field in AI for a long time.
Ben Goertzel (2:10:27.520)
No one can do it very well.
Lex Fridman (2:10:29.200)
So we've been looking at transformer neural networks
Lex Fridman (2:10:32.080)
and tree transformers, which are amazing.
Lex Fridman (2:10:35.760)
These came out of Google Brain, actually.
Lex Fridman (2:10:39.080)
And actually on that team was Lucas Kaiser,
Lex Fridman (2:10:41.920)
who used to work for me in the one,
Ben Goertzel (2:10:44.080)
the period 2005 through eight or something.
Lex Fridman (2:10:46.960)
So it's been fun to see my former
Ben Goertzel (2:10:50.200)
sort of AGI employees disperse and do
Lex Fridman (2:10:52.760)
all these amazing things.
Ben Goertzel (2:10:54.080)
Way too many sucked into Google, actually.
Lex Fridman (2:10:56.080)
Well, yeah, anyway.
Ben Goertzel (2:10:57.640)
We'll talk about that too.
Lex Fridman (2:10:58.960)
Lucas Kaiser and a bunch of these guys,
Ben Goertzel (2:11:00.640)
they create transformer networks,
Lex Fridman (2:11:03.200)
that classic paper like attention is all you need
Lex Fridman (2:11:05.480)
and all these things following on from that.
Lex Fridman (2:11:08.160)
So we're looking at transformer networks.
Lex Fridman (2:11:10.160)
And like, these are able to,
Lex Fridman (2:11:13.520)
I mean, this is what underlies GPT2 and GPT3 and so on,
Ben Goertzel (2:11:16.480)
which are very, very cool
Lex Fridman (2:11:18.120)
and have absolutely no cognitive understanding
Ben Goertzel (2:11:20.320)
of any of the texts they're looking at.
Lex Fridman (2:11:21.680)
Like they're very intelligent idiots, right?
Lex Fridman (2:11:24.960)
So sorry to take, but this small, I'll bring this back,
Lex Fridman (2:11:28.080)
but do you think GPT3 understands language?
Ben Goertzel (2:11:31.760)
No, no, it understands nothing.
Lex Fridman (2:11:34.080)
It's a complete idiot.
Lex Fridman (2:11:35.320)
But it's a brilliant idiot.
Lex Fridman (2:11:36.720)
You don't think GPT20 will understand language?
Ben Goertzel (2:11:40.520)
No, no, no.
Lex Fridman (2:11:42.240)
So size is not gonna buy you understanding.
Lex Fridman (2:11:45.160)
And any more than a faster car is gonna get you to Mars.
Lex Fridman (2:11:48.840)
It's a completely different kind of thing.
Ben Goertzel (2:11:50.920)
I mean, these networks are very cool.
Lex Fridman (2:11:54.280)
And as an entrepreneur,
Ben Goertzel (2:11:55.520)
I can see many highly valuable uses for them.
Lex Fridman (2:11:57.760)
And as an artist, I love them, right?
Lex Fridman (2:12:01.080)
So I mean, we're using our own neural model,
Lex Fridman (2:12:05.240)
which is along those lines
Ben Goertzel (2:12:06.560)
to control the Philip K. Dick robot now.
Lex Fridman (2:12:09.000)
And it's amazing to like train a neural model
Ben Goertzel (2:12:12.200)
on the robot Philip K. Dick
Lex Fridman (2:12:14.000)
and see it come up with like crazed,
Ben Goertzel (2:12:15.840)
stoned philosopher pronouncements,
Lex Fridman (2:12:18.400)
very much like what Philip K. Dick might've said, right?
Ben Goertzel (2:12:21.440)
Like these models are super cool.
Lex Fridman (2:12:24.840)
And I'm working with Hanson Robotics now
Ben Goertzel (2:12:27.720)
on using a similar, but more sophisticated one for Sophia,
Lex Fridman (2:12:30.600)
which we haven't launched yet.
Lex Fridman (2:12:34.080)
But so I think it's cool.
Lex Fridman (2:12:36.080)
But no, these are recognizing a large number
Ben Goertzel (2:12:39.480)
of shallow patterns.
Lex Fridman (2:12:42.200)
They're not forming an abstract representation.
Lex Fridman (2:12:44.840)
And that's the point I was coming to
Lex Fridman (2:12:47.120)
when we're looking at grammar induction,
Ben Goertzel (2:12:50.680)
we tried to mine patterns out of the structure
Lex Fridman (2:12:53.520)
of the transformer network.
Lex Fridman (2:12:55.880)
And you can, but the patterns aren't what you want.
Lex Fridman (2:12:59.600)
They're nasty.
Lex Fridman (2:13:00.600)
So I mean, if you do supervised learning,
Lex Fridman (2:13:03.200)
if you look at sentences where you know
Ben Goertzel (2:13:04.560)
the correct parts of a sentence,
Lex Fridman (2:13:06.520)
you can learn a matrix that maps
Ben Goertzel (2:13:09.120)
between the internal representation of the transformer
Lex Fridman (2:13:12.240)
and the parse of the sentence.
Lex Fridman (2:13:14.120)
And so then you can actually train something
Lex Fridman (2:13:16.120)
that will output the sentence parse
Ben Goertzel (2:13:18.440)
from the transformer network's internal state.
Lex Fridman (2:13:20.680)
And we did this, I think Christopher Manning,
Ben Goertzel (2:13:24.720)
some others have not done this also.
Lex Fridman (2:13:28.080)
But I mean, what you get is that the representation
Ben Goertzel (2:13:30.600)
is hardly ugly and is scattered all over the network
Lex Fridman (2:13:33.200)
and doesn't look like the rules of grammar
Lex Fridman (2:13:34.920)
that you know are the right rules of grammar, right?
Lex Fridman (2:13:37.240)
It's kind of ugly.
Lex Fridman (2:13:38.240)
So what we're actually doing is we're using
Lex Fridman (2:13:41.440)
a symbolic grammar learning algorithm,
Lex Fridman (2:13:44.280)
but we're using the transformer neural network
Lex Fridman (2:13:46.760)
as a sentence probability oracle.
Lex Fridman (2:13:48.880)
So like if you have a rule of grammar
Lex Fridman (2:13:52.120)
and you aren't sure if it's a correct rule of grammar or not,
Ben Goertzel (2:13:54.800)
you can generate a bunch of sentences
Lex Fridman (2:13:56.440)
using that rule of grammar
Lex Fridman (2:13:58.040)
and a bunch of sentences violating that rule of grammar.
Lex Fridman (2:14:00.880)
And you can see the transformer model
Ben Goertzel (2:14:04.480)
doesn't think the sentences obeying the rule of grammar
Lex Fridman (2:14:06.720)
are more probable than the sentences
Ben Goertzel (2:14:08.280)
disobeying the rule of grammar.
Lex Fridman (2:14:10.080)
So in that way, you can use the neural model
Ben Goertzel (2:14:11.840)
as a sense probability oracle
Lex Fridman (2:14:13.840)
to guide a symbolic grammar learning process.
Lex Fridman (2:14:19.960)
And that seems to work better than trying to milk
Lex Fridman (2:14:24.000)
the grammar out of the neural network
Ben Goertzel (2:14:25.840)
that doesn't have it in there.
Lex Fridman (2:14:26.760)
So I think the thing is these neural nets
Ben Goertzel (2:14:29.480)
are not getting a semantically meaningful representation
Lex Fridman (2:14:32.880)
internally by and large.
Lex Fridman (2:14:35.360)
So one line of research is to try to get them to do that.
Lex Fridman (2:14:38.120)
And InfoGAN was trying to do that.
Lex Fridman (2:14:40.000)
So like if you look back like two years ago,
Lex Fridman (2:14:43.040)
there was all these papers on like at Edward,
Ben Goertzel (2:14:45.280)
this probabilistic programming neural net framework
Lex Fridman (2:14:47.400)
that Google had, which came out of InfoGAN.
Lex Fridman (2:14:49.640)
So the idea there was like you could train
Lex Fridman (2:14:53.720)
an InfoGAN neural net model,
Ben Goertzel (2:14:55.600)
which is a generative associative network
Lex Fridman (2:14:57.200)
to recognize and generate faces.
Lex Fridman (2:14:59.200)
And the model would automatically learn a variable
Lex Fridman (2:15:02.160)
for how long the nose is and automatically learn a variable
Ben Goertzel (2:15:04.400)
for how wide the eyes are
Lex Fridman (2:15:05.760)
or how big the lips are or something, right?
Lex Fridman (2:15:08.040)
So it automatically learned these variables,
Lex Fridman (2:15:11.040)
which have a semantic meaning.
Lex Fridman (2:15:12.480)
So that was a rare case where a neural net
Lex Fridman (2:15:15.320)
trained with a fairly standard GAN method
Ben Goertzel (2:15:18.080)
was able to actually learn the semantic representation.
Lex Fridman (2:15:20.880)
So for many years, many of us tried to take that
Ben Goertzel (2:15:23.240)
the next step and get a GAN type neural network
Lex Fridman (2:15:27.200)
that would have not just a list of semantic latent variables,
Lex Fridman (2:15:31.680)
but would have say a Bayes net of semantic latent variables
Lex Fridman (2:15:33.960)
with dependencies between them.
Ben Goertzel (2:15:35.440)
The whole programming framework Edward was made for that.
Lex Fridman (2:15:38.840)
I mean, no one got it to work, right?
Lex Fridman (2:15:40.720)
And it could be.
Lex Fridman (2:15:41.560)
Do you think it's possible?
Lex Fridman (2:15:42.960)
Yeah, do you think?
Lex Fridman (2:15:43.800)
I don't know.
Ben Goertzel (2:15:44.760)
It might be that back propagation just won't work for it
Lex Fridman (2:15:47.280)
because the gradients are too screwed up.
Ben Goertzel (2:15:49.720)
Maybe you could get it to work using CMAES
Lex Fridman (2:15:52.000)
or some like floating point evolutionary algorithm.
Ben Goertzel (2:15:54.840)
We tried, we didn't get it to work.
Lex Fridman (2:15:57.000)
Eventually we just paused that rather than gave it up.
Ben Goertzel (2:16:01.360)
We paused that and said, well, okay, let's try
Lex Fridman (2:16:04.000)
more innovative ways to learn implicit,
Ben Goertzel (2:16:08.640)
to learn what are the representations implicit
Lex Fridman (2:16:11.000)
in that network without trying to make it grow
Ben Goertzel (2:16:13.640)
inside that network.
Lex Fridman (2:16:14.720)
And I described how we're doing that in language.
Lex Fridman (2:16:19.720)
You can do similar things in vision, right?
Lex Fridman (2:16:21.440)
So what?
Ben Goertzel (2:16:22.280)
Use it as an oracle.
Lex Fridman (2:16:23.360)
Yeah, yeah, yeah.
Lex Fridman (2:16:24.200)
So you can, that's one way is that you use
Lex Fridman (2:16:26.240)
a structure learning algorithm, which is symbolic.
Lex Fridman (2:16:29.120)
And then you use the deep neural net as an oracle
Lex Fridman (2:16:32.480)
to guide the structure learning algorithm.
Ben Goertzel (2:16:34.240)
The other way to do it is like Infogam was trying to do
Lex Fridman (2:16:37.880)
and try to tweak the neural network
Ben Goertzel (2:16:40.040)
to have the symbolic representation inside it.
Lex Fridman (2:16:43.760)
I tend to think what the brain is doing
Ben Goertzel (2:16:46.440)
is more like using the deep neural net type thing
Lex Fridman (2:16:51.680)
as an oracle.
Ben Goertzel (2:16:52.520)
I think the visual cortex or the cerebellum
Lex Fridman (2:16:56.680)
are probably learning a non semantically meaningful
Ben Goertzel (2:17:00.280)
opaque tangled representation.
Lex Fridman (2:17:02.400)
And then when they interface with the more cognitive parts
Ben Goertzel (2:17:04.600)
of the cortex, the cortex is sort of using those
Lex Fridman (2:17:08.080)
as an oracle and learning the abstract representation.
Lex Fridman (2:17:10.720)
So if you do sports, say take for example,
Lex Fridman (2:17:13.200)
serving in tennis, right?
Ben Goertzel (2:17:15.240)
I mean, my tennis serve is okay, not great,
Lex Fridman (2:17:17.680)
but I learned it by trial and error, right?
Lex Fridman (2:17:19.760)
And I mean, I learned music by trial and error too.
Lex Fridman (2:17:22.120)
I just sit down and play, but then if you're an athlete,
Ben Goertzel (2:17:25.960)
which I'm not a good athlete,
Lex Fridman (2:17:27.080)
I mean, then you'll watch videos of yourself serving
Lex Fridman (2:17:30.360)
and your coach will help you think about what you're doing
Lex Fridman (2:17:32.760)
and you'll then form a declarative representation,
Lex Fridman (2:17:35.040)
but your cerebellum maybe didn't have
Lex Fridman (2:17:37.160)
a declarative representation.
Ben Goertzel (2:17:38.640)
Same way with music, like I will hear something in my head,
Lex Fridman (2:17:43.560)
I'll sit down and play the thing like I heard it.
Lex Fridman (2:17:46.960)
And then I will try to study what my fingers did
Lex Fridman (2:17:51.000)
to see like, what did you just play?
Lex Fridman (2:17:52.760)
Like how did you do that, right?
Lex Fridman (2:17:55.600)
Because if you're composing,
Ben Goertzel (2:17:57.720)
you may wanna see how you did it
Lex Fridman (2:17:59.720)
and then declaratively morph that in some way
Lex Fridman (2:18:02.680)
that your fingers wouldn't think of, right?
Lex Fridman (2:18:05.240)
But the physiological movement may come out of some opaque,
Lex Fridman (2:18:10.280)
like cerebellar reinforcement learned thing, right?
Lex Fridman (2:18:14.440)
And so that's, I think trying to milk the structure
Ben Goertzel (2:18:17.680)
of a neural net by treating it as an oracle,
Lex Fridman (2:18:19.320)
maybe more like how your declarative mind post processes
Lex Fridman (2:18:23.960)
what your visual or motor cortex.
Lex Fridman (2:18:27.760)
I mean, in vision, it's the same way,
Ben Goertzel (2:18:29.400)
like you can recognize beautiful art
Lex Fridman (2:18:34.800)
much better than you can say why
Ben Goertzel (2:18:36.760)
you think that piece of art is beautiful.
Lex Fridman (2:18:38.520)
But if you're trained as an art critic,
Ben Goertzel (2:18:40.520)
you do learn to say why.
Lex Fridman (2:18:41.680)
And some of it's bullshit, but some of it isn't, right?
Ben Goertzel (2:18:44.040)
Some of it is learning to map sensory knowledge
Lex Fridman (2:18:46.840)
into declarative and linguistic knowledge,
Ben Goertzel (2:18:51.120)
yet without necessarily making the sensory system itself
Lex Fridman (2:18:56.040)
use a transparent and an easily communicable representation.
Ben Goertzel (2:19:00.640)
Yeah, that's fascinating to think of neural networks
Lex Fridman (2:19:02.960)
as like dumb question answers that you can just milk
Ben Goertzel (2:19:08.200)
to build up a knowledge base.
Lex Fridman (2:19:10.920)
And then it can be multiple networks, I suppose,
Ben Goertzel (2:19:12.680)
from different.
Lex Fridman (2:19:13.600)
Yeah, yeah, so I think if a group like DeepMind or OpenAI
Ben Goertzel (2:19:18.160)
were to build AGI, and I think DeepMind is like
Lex Fridman (2:19:21.520)
a thousand times more likely from what I could tell,
Ben Goertzel (2:19:25.920)
because they've hired a lot of people with broad minds
Lex Fridman (2:19:30.040)
and many different approaches and angles on AGI,
Ben Goertzel (2:19:34.360)
whereas OpenAI is also awesome,
Lex Fridman (2:19:36.640)
but I see them as more of like a pure
Ben Goertzel (2:19:39.040)
deep reinforcement learning shop.
Lex Fridman (2:19:41.160)
Yeah, this time, I got you.
Lex Fridman (2:19:42.000)
So far. Yeah, there's a lot of,
Lex Fridman (2:19:43.880)
you're right, I mean, there's so much interdisciplinary
Ben Goertzel (2:19:48.600)
work at DeepMind, like neuroscience.
Lex Fridman (2:19:50.280)
And you put that together with Google Brain,
Ben Goertzel (2:19:52.240)
which granted they're not working that closely together now,
Lex Fridman (2:19:54.760)
but my oldest son Zarathustra is doing his PhD
Ben Goertzel (2:19:58.840)
in machine learning applied to automated theorem proving
Lex Fridman (2:20:01.640)
in Prague under Josef Urban.
Lex Fridman (2:20:03.840)
So the first paper, DeepMath, which applied deep neural nets
Lex Fridman (2:20:08.400)
to guide theorem proving was out of Google Brain.
Ben Goertzel (2:20:10.680)
I mean, by now, the automated theorem proving community
Lex Fridman (2:20:14.960)
is going way, way, way beyond anything Google was doing,
Lex Fridman (2:20:18.360)
but still, yeah, but anyway,
Lex Fridman (2:20:21.120)
if that community was gonna make an AGI,
Ben Goertzel (2:20:23.760)
probably one way they would do it was,
Lex Fridman (2:20:27.160)
take 25 different neural modules,
Ben Goertzel (2:20:30.680)
architected in different ways,
Lex Fridman (2:20:32.040)
maybe resembling different parts of the brain,
Ben Goertzel (2:20:33.800)
like a basal ganglia model, cerebellum model,
Lex Fridman (2:20:36.280)
a thalamus module, a few hippocampus models,
Ben Goertzel (2:20:40.440)
number of different models,
Lex Fridman (2:20:41.480)
representing parts of the cortex, right?
Ben Goertzel (2:20:43.680)
Take all of these and then wire them together
Lex Fridman (2:20:47.920)
to co train and learn them together like that.
Ben Goertzel (2:20:52.520)
That would be an approach to creating an AGI.
Lex Fridman (2:20:57.240)
One could implement something like that efficiently
Ben Goertzel (2:20:59.640)
on top of our true AGI, like OpenCog 2.0 system,
Lex Fridman (2:21:03.800)
once it exists, although obviously Google
Ben Goertzel (2:21:06.640)
has their own highly efficient implementation architecture.
Lex Fridman (2:21:10.240)
So I think that's a decent way to build AGI.
Ben Goertzel (2:21:13.280)
I was very interested in that in the mid 90s,
Lex Fridman (2:21:15.680)
but I mean, the knowledge about how the brain works
Ben Goertzel (2:21:19.440)
sort of pissed me off, like it wasn't there yet.
Lex Fridman (2:21:21.520)
Like, you know, in the hippocampus,
Ben Goertzel (2:21:23.080)
you have these concept neurons,
Lex Fridman (2:21:24.760)
like the so called grandmother neuron,
Ben Goertzel (2:21:26.720)
which everyone laughed at it, it's actually there.
Lex Fridman (2:21:28.520)
Like I have some Lex Friedman neurons
Ben Goertzel (2:21:31.080)
that fire differentially when I see you
Lex Fridman (2:21:33.280)
and not when I see any other person, right?
Lex Fridman (2:21:35.360)
So how do these Lex Friedman neurons,
Lex Fridman (2:21:38.880)
how do they coordinate with the distributed representation
Lex Fridman (2:21:41.400)
of Lex Friedman I have in my cortex, right?
Lex Fridman (2:21:44.520)
There's some back and forth between cortex and hippocampus
Ben Goertzel (2:21:47.680)
that lets these discrete symbolic representations
Lex Fridman (2:21:50.120)
in hippocampus correlate and cooperate
Ben Goertzel (2:21:53.200)
with the distributed representations in cortex.
Lex Fridman (2:21:55.680)
This probably has to do with how the brain
Lex Fridman (2:21:57.400)
does its version of abstraction and quantifier logic, right?
Lex Fridman (2:22:00.240)
Like you can have a single neuron in the hippocampus
Ben Goertzel (2:22:02.640)
that activates a whole distributed activation pattern
Lex Fridman (2:22:05.880)
in cortex, well, this may be how the brain does
Ben Goertzel (2:22:09.080)
like symbolization and abstraction
Lex Fridman (2:22:11.120)
as in functional programming or something,
Lex Fridman (2:22:14.280)
but we can't measure it.
Lex Fridman (2:22:15.360)
Like we don't have enough electrodes stuck
Ben Goertzel (2:22:17.560)
between the cortex and the hippocampus
Lex Fridman (2:22:20.960)
in any known experiment to measure it.
Lex Fridman (2:22:23.080)
So I got frustrated with that direction,
Lex Fridman (2:22:26.360)
not because it's impossible.
Ben Goertzel (2:22:27.560)
Because we just don't understand enough yet.
Lex Fridman (2:22:29.720)
Of course, it's a valid research direction.
Ben Goertzel (2:22:31.760)
You can try to understand more and more.
Lex Fridman (2:22:33.720)
And we are measuring more and more
Ben Goertzel (2:22:34.960)
about what happens in the brain now than ever before.
Lex Fridman (2:22:38.120)
So it's quite interesting.
Ben Goertzel (2:22:40.560)
On the other hand, I sort of got more
Lex Fridman (2:22:43.400)
of an engineering mindset about AGI.
Ben Goertzel (2:22:46.520)
I'm like, well, okay,
Lex Fridman (2:22:47.920)
we don't know how the brain works that well.
Ben Goertzel (2:22:50.200)
We don't know how birds fly that well yet either.
Lex Fridman (2:22:52.360)
We have no idea how a hummingbird flies
Ben Goertzel (2:22:54.080)
in terms of the aerodynamics of it.
Lex Fridman (2:22:56.280)
On the other hand, we know basic principles
Ben Goertzel (2:22:59.280)
of like flapping and pushing the air down.
Lex Fridman (2:23:01.760)
And we know the basic principles
Ben Goertzel (2:23:03.520)
of how the different parts of the brain work.
Lex Fridman (2:23:05.720)
So let's take those basic principles
Lex Fridman (2:23:07.480)
and engineer something that embodies those basic principles,
Lex Fridman (2:23:11.480)
but is well designed for the hardware
Ben Goertzel (2:23:14.040)
that we have on hand right now.
Lex Fridman (2:23:18.080)
So do you think we can create AGI
Lex Fridman (2:23:20.200)
before we understand how the brain works?
Lex Fridman (2:23:22.440)
I think that's probably what will happen.
Lex Fridman (2:23:25.120)
And maybe the AGI will help us do better brain imaging
Lex Fridman (2:23:28.560)
that will then let us build artificial humans,
Ben Goertzel (2:23:30.880)
which is very, very interesting to us
Lex Fridman (2:23:33.400)
because we are humans, right?
Ben Goertzel (2:23:34.960)
I mean, building artificial humans is super worthwhile.
Lex Fridman (2:23:38.840)
I just think it's probably not the shortest path to AGI.
Lex Fridman (2:23:42.760)
So it's fascinating idea that we would build AGI
Lex Fridman (2:23:45.680)
to help us understand ourselves.
Ben Goertzel (2:23:50.040)
A lot of people ask me if the young people
Lex Fridman (2:23:54.600)
interested in doing artificial intelligence,
Ben Goertzel (2:23:56.440)
they look at sort of doing graduate level, even undergrads,
Lex Fridman (2:24:01.440)
but graduate level research and they see
Ben Goertzel (2:24:04.520)
whether the artificial intelligence community stands now,
Lex Fridman (2:24:06.840)
it's not really AGI type research for the most part.
Lex Fridman (2:24:09.920)
So the natural question they ask is
Lex Fridman (2:24:12.080)
what advice would you give?
Ben Goertzel (2:24:13.640)
I mean, maybe I could ask if people were interested
Lex Fridman (2:24:17.320)
in working on OpenCog or in some kind of direct
Ben Goertzel (2:24:22.520)
or indirect connection to OpenCog or AGI research,
Lex Fridman (2:24:25.160)
what would you recommend?
Ben Goertzel (2:24:28.040)
OpenCog, first of all, is open source project.
Lex Fridman (2:24:30.960)
There's a Google group discussion list.
Ben Goertzel (2:24:35.360)
There's a GitHub repository.
Lex Fridman (2:24:36.760)
So if anyone's interested in lending a hand
Ben Goertzel (2:24:39.800)
with that aspect of AGI,
Lex Fridman (2:24:42.600)
introduce yourself on the OpenCog email list.
Lex Fridman (2:24:46.000)
And there's a Slack as well.
Lex Fridman (2:24:47.920)
I mean, we're certainly interested to have inputs
Ben Goertzel (2:24:53.080)
into our redesign process for a new version of OpenCog,
Lex Fridman (2:24:57.520)
but also we're doing a lot of very interesting research.
Ben Goertzel (2:25:01.160)
I mean, we're working on data analysis
Lex Fridman (2:25:04.080)
for COVID clinical trials.
Ben Goertzel (2:25:05.600)
We're working with Hanson Robotics.
Lex Fridman (2:25:06.960)
We're doing a lot of cool things
Ben Goertzel (2:25:08.000)
with the current version of OpenCog now.
Lex Fridman (2:25:10.720)
So there's certainly opportunity to jump into OpenCog
Ben Goertzel (2:25:14.720)
or various other open source AGI oriented projects.
Lex Fridman (2:25:18.760)
So would you say there's like masters
Lex Fridman (2:25:20.280)
and PhD theses in there?
Lex Fridman (2:25:22.080)
Plenty, yeah, plenty, of course.
Ben Goertzel (2:25:23.960)
I mean, the challenge is to find a supervisor
Lex Fridman (2:25:26.920)
who wants to foster that sort of research,
Lex Fridman (2:25:29.720)
but it's way easier than it was when I got my PhD, right?
Lex Fridman (2:25:32.840)
It's okay, great.
Ben Goertzel (2:25:33.680)
We talked about OpenCog, which is kind of one,
Lex Fridman (2:25:36.360)
the software framework,
Lex Fridman (2:25:38.000)
but also the actual attempt to build an AGI system.
Lex Fridman (2:25:44.160)
And then there is this exciting idea of SingularityNet.
Lex Fridman (2:25:48.600)
So maybe can you say first what is SingularityNet?
Lex Fridman (2:25:53.160)
Sure, sure.
Ben Goertzel (2:25:54.280)
SingularityNet is a platform
Lex Fridman (2:25:59.040)
for realizing a decentralized network
Ben Goertzel (2:26:05.880)
of artificial intelligences.
Lex Fridman (2:26:08.280)
So Marvin Minsky, the AI pioneer who I knew a little bit,
Ben Goertzel (2:26:14.440)
he had the idea of a society of minds,
Lex Fridman (2:26:16.560)
like you should achieve an AI
Ben Goertzel (2:26:18.360)
not by writing one algorithm or one program,
Lex Fridman (2:26:21.040)
but you should put a bunch of different AIs out there
Lex Fridman (2:26:24.000)
and the different AIs will interact with each other,
Lex Fridman (2:26:27.760)
each playing their own role.
Lex Fridman (2:26:29.480)
And then the totality of the society of AIs
Lex Fridman (2:26:32.560)
would be the thing
Ben Goertzel (2:26:34.240)
that displayed the human level intelligence.
Lex Fridman (2:26:36.560)
And I had, when he was alive,
Ben Goertzel (2:26:39.000)
I had many debates with Marvin about this idea.
Lex Fridman (2:26:43.000)
And I think he really thought the mind
Ben Goertzel (2:26:49.080)
was more like a society than I do.
Lex Fridman (2:26:51.200)
Like I think you could have a mind
Ben Goertzel (2:26:54.080)
that was as disorganized as a human society,
Lex Fridman (2:26:56.720)
but I think a human like mind
Ben Goertzel (2:26:57.880)
has a bit more central control than that actually.
Lex Fridman (2:27:00.080)
Like, I mean, we have this thalamus
Lex Fridman (2:27:02.840)
and the medulla and limbic system.
Lex Fridman (2:27:04.760)
We have a sort of top down control system
Ben Goertzel (2:27:07.960)
that guides much of what we do,
Lex Fridman (2:27:10.840)
more so than a society does.
Lex Fridman (2:27:12.760)
So I think he stretched that metaphor a little too far,
Lex Fridman (2:27:16.880)
but I also think there's something interesting there.
Lex Fridman (2:27:20.840)
And so in the 90s,
Lex Fridman (2:27:24.040)
when I started my first sort of nonacademic AI project,
Ben Goertzel (2:27:27.960)
WebMind, which was an AI startup in New York
Lex Fridman (2:27:30.960)
in the Silicon Alley area in the late 90s,
Lex Fridman (2:27:34.640)
what I was aiming to do there
Lex Fridman (2:27:36.280)
was make a distributed society of AIs,
Ben Goertzel (2:27:40.000)
the different parts of which would live
Lex Fridman (2:27:41.360)
on different computers all around the world.
Lex Fridman (2:27:43.640)
And each one would do its own thinking
Lex Fridman (2:27:45.240)
about the data local to it,
Lex Fridman (2:27:47.080)
but they would all share information with each other
Lex Fridman (2:27:48.960)
and outsource work with each other and cooperate.
Lex Fridman (2:27:51.320)
And the intelligence would be in the whole collective.
Lex Fridman (2:27:54.040)
And I organized a conference together with Francis Heiligen
Ben Goertzel (2:27:57.680)
at Free University of Brussels in 2001,
Lex Fridman (2:28:00.600)
which was the Global Brain Zero Conference.
Lex Fridman (2:28:02.920)
And we're planning the next version,
Lex Fridman (2:28:04.680)
the Global Brain One Conference
Ben Goertzel (2:28:06.920)
at the Free University of Brussels for next year, 2021.
Lex Fridman (2:28:10.120)
So 20 years after.
Lex Fridman (2:28:12.000)
And then maybe we can have the next one 10 years after that,
Lex Fridman (2:28:14.560)
like exponentially faster until the singularity comes, right?
Ben Goertzel (2:28:19.320)
The timing is right, yeah.
Lex Fridman (2:28:20.680)
Yeah, yeah, exactly.
Lex Fridman (2:28:22.160)
So yeah, the idea with the Global Brain
Lex Fridman (2:28:25.000)
was maybe the AI won't just be in a program
Ben Goertzel (2:28:28.120)
on one guy's computer,
Lex Fridman (2:28:29.560)
but the AI will be in the internet as a whole
Ben Goertzel (2:28:32.960)
with the cooperation of different AI modules
Lex Fridman (2:28:35.080)
living in different places.
Lex Fridman (2:28:37.040)
So one of the issues you face
Lex Fridman (2:28:39.280)
when architecting a system like that
Lex Fridman (2:28:41.160)
is, you know, how is the whole thing controlled?
Lex Fridman (2:28:44.760)
Do you have like a centralized control unit
Ben Goertzel (2:28:47.200)
that pulls the puppet strings
Lex Fridman (2:28:48.640)
of all the different modules there?
Ben Goertzel (2:28:50.720)
Or do you have a fundamentally decentralized network
Lex Fridman (2:28:55.480)
where the society of AIs is controlled
Ben Goertzel (2:28:59.320)
in some democratic and self organized way,
Lex Fridman (2:29:01.040)
but all the AIs in that society, right?
Lex Fridman (2:29:04.760)
And Francis and I had different view of many things,
Lex Fridman (2:29:08.680)
but we both wanted to make like a global society
Ben Goertzel (2:29:13.680)
of AI minds with a decentralized organizational mode.
Lex Fridman (2:29:19.840)
Now, the main difference was he wanted the individual AIs
Ben Goertzel (2:29:25.400)
to be all incredibly simple
Lex Fridman (2:29:27.440)
and all the intelligence to be on the collective level.
Ben Goertzel (2:29:30.360)
Whereas I thought that was cool,
Lex Fridman (2:29:32.960)
but I thought a more practical way to do it might be
Ben Goertzel (2:29:35.880)
if some of the agents in the society of minds
Lex Fridman (2:29:39.480)
were fairly generally intelligent on their own.
Lex Fridman (2:29:41.520)
So like you could have a bunch of open cogs out there
Lex Fridman (2:29:44.480)
and a bunch of simpler learning systems.
Lex Fridman (2:29:47.120)
And then these are all cooperating, coordinating together
Lex Fridman (2:29:49.840)
sort of like in the brain.
Ben Goertzel (2:29:51.760)
Okay, the brain as a whole is the general intelligence,
Lex Fridman (2:29:55.320)
but some parts of the cortex,
Ben Goertzel (2:29:56.640)
you could say have a fair bit of general intelligence
Lex Fridman (2:29:58.560)
on their own,
Ben Goertzel (2:29:59.720)
whereas say parts of the cerebellum or limbic system
Lex Fridman (2:30:02.120)
have very little general intelligence on their own.
Lex Fridman (2:30:04.520)
And they're contributing to general intelligence
Lex Fridman (2:30:07.240)
by way of their connectivity to other modules.
Lex Fridman (2:30:10.880)
Do you see instantiations of the same kind of,
Lex Fridman (2:30:13.680)
maybe different versions of open cog,
Lex Fridman (2:30:15.400)
but also just the same version of open cog
Lex Fridman (2:30:17.320)
and maybe many instantiations of it as being all parts of it?
Ben Goertzel (2:30:21.320)
That's what David and Hans and I want to do
Lex Fridman (2:30:23.040)
with many Sophia and other robots.
Ben Goertzel (2:30:25.320)
Each one has its own individual mind living on the server,
Lex Fridman (2:30:29.200)
but there's also a collective intelligence infusing them
Lex Fridman (2:30:32.080)
and a part of the mind living on the edge in each robot.
Lex Fridman (2:30:35.440)
So the thing is at that time,
Ben Goertzel (2:30:38.520)
as well as WebMind being implemented in Java 1.1
Lex Fridman (2:30:41.840)
as like a massive distributed system,
Ben Goertzel (2:30:46.920)
blockchain wasn't there yet.
Lex Fridman (2:30:48.160)
So had them do this decentralized control.
Ben Goertzel (2:30:51.880)
We sort of knew it.
Lex Fridman (2:30:52.880)
We knew about distributed systems.
Ben Goertzel (2:30:54.360)
We knew about encryption.
Lex Fridman (2:30:55.760)
So I mean, we had the key principles
Ben Goertzel (2:30:58.080)
of what underlies blockchain now,
Lex Fridman (2:31:00.080)
but I mean, we didn't put it together
Ben Goertzel (2:31:01.760)
in the way that it's been done now.
Lex Fridman (2:31:02.880)
So when Vitalik Buterin and colleagues
Ben Goertzel (2:31:05.360)
came out with Ethereum blockchain,
Lex Fridman (2:31:08.120)
many, many years later, like 2013 or something,
Ben Goertzel (2:31:11.840)
then I was like, well, this is interesting.
Lex Fridman (2:31:13.920)
Like this is solidity scripting language.
Ben Goertzel (2:31:17.000)
It's kind of dorky in a way.
Lex Fridman (2:31:18.520)
And I don't see why you need to turn complete language
Ben Goertzel (2:31:21.440)
for this purpose.
Lex Fridman (2:31:22.440)
But on the other hand,
Ben Goertzel (2:31:24.320)
this is like the first time I could sit down
Lex Fridman (2:31:27.160)
and start to like script infrastructure
Ben Goertzel (2:31:29.920)
for decentralized control of the AIs
Lex Fridman (2:31:32.440)
in this society of minds in a tractable way.
Ben Goertzel (2:31:35.240)
Like you can hack the Bitcoin code base,
Lex Fridman (2:31:37.200)
but it's really annoying.
Ben Goertzel (2:31:38.520)
Whereas solidity is Ethereum scripting language
Lex Fridman (2:31:41.720)
is just nicer and easier to use.
Ben Goertzel (2:31:44.440)
I'm very annoyed with it by this point.
Lex Fridman (2:31:45.880)
But like Java, I mean, these languages are amazing
Ben Goertzel (2:31:49.000)
when they first come out.
Lex Fridman (2:31:50.920)
So then I came up with the idea
Ben Goertzel (2:31:52.480)
that turned into SingularityNet.
Lex Fridman (2:31:53.840)
Okay, let's make a decentralized agent system
Ben Goertzel (2:31:58.200)
where a bunch of different AIs,
Lex Fridman (2:32:00.480)
wrapped up in say different Docker containers
Ben Goertzel (2:32:02.680)
or LXC containers,
Lex Fridman (2:32:04.320)
different AIs can each of them have their own identity
Ben Goertzel (2:32:07.440)
on the blockchain.
Lex Fridman (2:32:08.760)
And the coordination of this community of AIs
Lex Fridman (2:32:11.800)
has no central controller, no dictator, right?
Lex Fridman (2:32:14.680)
And there's no central repository of information.
Ben Goertzel (2:32:17.160)
The coordination of the society of minds
Lex Fridman (2:32:19.400)
is done entirely by the decentralized network
Lex Fridman (2:32:22.680)
in a decentralized way by the algorithms, right?
Lex Fridman (2:32:25.840)
Because the model of Bitcoin is in math we trust, right?
Lex Fridman (2:32:29.200)
And so that's what you need.
Lex Fridman (2:32:30.800)
You need the society of minds to trust only in math,
Ben Goertzel (2:32:33.880)
not trust only in one centralized server.
Lex Fridman (2:32:37.720)
So the AI systems themselves are outside of the blockchain,
Lex Fridman (2:32:40.640)
but then the communication between them.
Lex Fridman (2:32:41.800)
At the moment, yeah, yeah.
Ben Goertzel (2:32:43.960)
I would have loved to put the AI's operations on chain
Lex Fridman (2:32:46.880)
in some sense, but in Ethereum, it's just too slow.
Ben Goertzel (2:32:50.480)
You can't do it.
Lex Fridman (2:32:52.680)
Somehow it's the basic communication between AI systems.
Ben Goertzel (2:32:56.120)
That's the distribution.
Lex Fridman (2:32:58.360)
Basically an AI is just some software in singularity.
Ben Goertzel (2:33:02.520)
An AI is just some software process living in a container.
Lex Fridman (2:33:05.920)
And there's a proxy that lives in that container
Ben Goertzel (2:33:09.040)
along with the AI that handles the interaction
Lex Fridman (2:33:10.840)
with the rest of singularity net.
Lex Fridman (2:33:13.120)
And then when one AI wants to contribute
Lex Fridman (2:33:15.880)
with another one in the network,
Ben Goertzel (2:33:16.920)
they set up a number of channels.
Lex Fridman (2:33:18.600)
And the setup of those channels uses the Ethereum blockchain.
Ben Goertzel (2:33:22.600)
Once the channels are set up,
Lex Fridman (2:33:24.480)
then data flows along those channels
Ben Goertzel (2:33:26.160)
without having to be on the blockchain.
Lex Fridman (2:33:29.240)
All that goes on the blockchain is the fact
Ben Goertzel (2:33:31.080)
that some data went along that channel.
Lex Fridman (2:33:33.160)
So you can do...
Lex Fridman (2:33:34.240)
So there's not a shared knowledge.
Lex Fridman (2:33:38.720)
Well, the identity of each agent is on the blockchain,
Ben Goertzel (2:33:43.160)
on the Ethereum blockchain.
Lex Fridman (2:33:44.800)
If one agent rates the reputation of another agent,
Ben Goertzel (2:33:48.000)
that goes on the blockchain.
Lex Fridman (2:33:49.560)
And agents can publish what APIs they will fulfill
Ben Goertzel (2:33:52.880)
on the blockchain.
Lex Fridman (2:33:54.520)
But the actual data for AI and the results for AI
Ben Goertzel (2:33:58.040)
is not on the blockchain.
Lex Fridman (2:33:58.880)
Do you think it could be?
Lex Fridman (2:33:59.720)
Do you think it should be?
Lex Fridman (2:34:02.320)
In some cases, it should be.
Ben Goertzel (2:34:04.120)
In some cases, maybe it shouldn't be.
Lex Fridman (2:34:05.880)
But I mean, I think that...
Lex Fridman (2:34:09.320)
So I'll give you an example.
Lex Fridman (2:34:10.160)
Using Ethereum, you can't do it.
Ben Goertzel (2:34:11.640)
Using now, there's more modern and faster blockchains
Lex Fridman (2:34:16.640)
where you could start to do that in some cases.
Ben Goertzel (2:34:21.920)
Two years ago, that was less so.
Lex Fridman (2:34:23.360)
It's a very rapidly evolving ecosystem.
Lex Fridman (2:34:25.640)
So like one example, maybe you can comment on
Lex Fridman (2:34:28.920)
something I worked a lot on is autonomous vehicles.
Ben Goertzel (2:34:31.840)
You can see each individual vehicle as an AI system.
Lex Fridman (2:34:35.680)
And you can see vehicles from Tesla, for example,
Lex Fridman (2:34:39.600)
and then Ford and GM and all these as also like larger...
Lex Fridman (2:34:44.600)
I mean, they all are running the same kind of system
Ben Goertzel (2:34:47.000)
on each sets of vehicles.
Lex Fridman (2:34:49.280)
So it's individual AI systems and individual vehicles,
Lex Fridman (2:34:52.360)
but it's all different.
Lex Fridman (2:34:53.800)
The station is the same AI system within the same company.
Lex Fridman (2:34:57.520)
So you can envision a situation where all of those AI systems
Lex Fridman (2:35:02.360)
are put on SingularityNet, right?
Lex Fridman (2:35:05.400)
And how do you see that happening?
Lex Fridman (2:35:10.160)
And what would be the benefit?
Lex Fridman (2:35:11.520)
And could they share data?
Lex Fridman (2:35:13.000)
I guess one of the biggest things is that the power there's
Ben Goertzel (2:35:16.440)
in a decentralized control, but the benefit would have been,
Lex Fridman (2:35:20.440)
is really nice if they can somehow share the knowledge
Ben Goertzel (2:35:24.080)
in an open way if they choose to.
Lex Fridman (2:35:26.280)
Yeah, yeah, yeah, those are all quite good points.
Lex Fridman (2:35:29.920)
So I think the benefit from being on the decentralized network
Lex Fridman (2:35:37.760)
as we envision it is that we want the AIs in the network
Ben Goertzel (2:35:41.320)
to be outsourcing work to each other
Lex Fridman (2:35:43.800)
and making API calls to each other frequently.
Lex Fridman (2:35:47.440)
So the real benefit would be if that AI wanted to outsource
Lex Fridman (2:35:51.880)
some cognitive processing or data processing
Ben Goertzel (2:35:54.920)
or data pre processing, whatever,
Lex Fridman (2:35:56.720)
to some other AIs in the network,
Ben Goertzel (2:35:59.320)
which specialize in something different.
Lex Fridman (2:36:01.600)
And this really requires a different way of thinking
Lex Fridman (2:36:06.120)
about AI software development, right?
Lex Fridman (2:36:07.960)
So just like object oriented programming
Ben Goertzel (2:36:10.320)
was different than imperative programming.
Lex Fridman (2:36:12.720)
And now object oriented programmers all use these
Ben Goertzel (2:36:16.720)
frameworks to do things rather than just libraries even.
Lex Fridman (2:36:20.680)
You know, shifting to agent based programming
Ben Goertzel (2:36:23.120)
where AI agent is asking other like live real time
Lex Fridman (2:36:26.600)
evolving agents for feedback and what they're doing.
Ben Goertzel (2:36:29.960)
That's a different way of thinking.
Lex Fridman (2:36:31.480)
I mean, it's not a new one.
Ben Goertzel (2:36:32.960)
There was loads of papers on agent based programming
Lex Fridman (2:36:35.320)
in the 80s and onward.
Lex Fridman (2:36:37.120)
But if you're willing to shift to an agent based model
Lex Fridman (2:36:41.520)
of development, then you can put less and less in your AI
Lex Fridman (2:36:45.920)
and rely more and more on interactive calls
Lex Fridman (2:36:48.600)
to other AIs running in the network.
Lex Fridman (2:36:51.440)
And of course, that's not fully manifested yet
Lex Fridman (2:36:54.560)
because although we've rolled out a nice working version
Ben Goertzel (2:36:57.640)
of SingularityNet platform,
Lex Fridman (2:36:59.760)
there's only 50 to 100 AIs running in there now.
Ben Goertzel (2:37:03.760)
There's not tens of thousands of AIs.
Lex Fridman (2:37:05.880)
So we don't have the critical mass
Ben Goertzel (2:37:08.240)
for the whole society of mind to be doing
Lex Fridman (2:37:11.120)
what we want to do.
Ben Goertzel (2:37:11.960)
Yeah, the magic really happens
Lex Fridman (2:37:13.400)
when there's just a huge number of agents.
Ben Goertzel (2:37:15.320)
Yeah, yeah, exactly.
Lex Fridman (2:37:16.680)
In terms of data, we're partnering closely
Ben Goertzel (2:37:19.600)
with another blockchain project called Ocean Protocol.
Lex Fridman (2:37:23.520)
And Ocean Protocol, that's the project of Trent McConnachie
Ben Goertzel (2:37:27.240)
who developed BigchainDB,
Lex Fridman (2:37:28.720)
which is a blockchain based database.
Lex Fridman (2:37:30.800)
So Ocean Protocol is basically blockchain based big data
Lex Fridman (2:37:35.440)
and aims at making it efficient for different AI processes
Ben Goertzel (2:37:39.440)
or statistical processes or whatever
Lex Fridman (2:37:41.240)
to share large data sets.
Ben Goertzel (2:37:44.080)
Or if one process can send a clone of itself
Lex Fridman (2:37:46.600)
to work on the other guy's data set
Lex Fridman (2:37:48.200)
and send results back and so forth.
Lex Fridman (2:37:50.600)
So by getting Ocean and you have data lake,
Lex Fridman (2:37:55.560)
so this is the data ocean, right?
Lex Fridman (2:37:56.920)
So again, by getting Ocean and SingularityNet
Ben Goertzel (2:37:59.760)
to interoperate, we're aiming to take into account
Lex Fridman (2:38:03.760)
the big data aspect also.
Lex Fridman (2:38:05.840)
But it's quite challenging
Lex Fridman (2:38:08.240)
because to build this whole decentralized
Ben Goertzel (2:38:10.120)
blockchain based infrastructure,
Lex Fridman (2:38:12.400)
I mean, your competitors are like Google, Microsoft,
Ben Goertzel (2:38:14.960)
Alibaba and Amazon, which have so much money
Lex Fridman (2:38:17.960)
to put behind their centralized infrastructures,
Ben Goertzel (2:38:20.560)
plus they're solving simpler algorithmic problems
Lex Fridman (2:38:23.360)
because making it centralized in some ways is easier, right?
Lex Fridman (2:38:27.360)
So they're very major computer science challenges.
Lex Fridman (2:38:32.360)
And I think what you saw with the whole ICO boom
Ben Goertzel (2:38:35.760)
in the blockchain and cryptocurrency world
Lex Fridman (2:38:37.880)
is a lot of young hackers who were hacking Bitcoin
Ben Goertzel (2:38:42.040)
or Ethereum, and they see, well,
Lex Fridman (2:38:43.840)
why don't we make this decentralized on blockchain?
Ben Goertzel (2:38:46.800)
Then after they raised some money through an ICO,
Lex Fridman (2:38:48.720)
they realize how hard it is.
Lex Fridman (2:38:49.880)
And it's like, actually we're wrestling
Lex Fridman (2:38:52.040)
with incredibly hard computer science
Lex Fridman (2:38:54.680)
and software engineering and distributed systems problems,
Lex Fridman (2:38:58.720)
which can be solved, but they're just very difficult
Ben Goertzel (2:39:02.560)
to solve.
Lex Fridman (2:39:03.400)
And in some cases, the individuals who started
Ben Goertzel (2:39:05.800)
those projects were not well equipped
Lex Fridman (2:39:08.760)
to actually solve the problems that they wanted to solve.
Lex Fridman (2:39:12.320)
So you think, would you say that's the main bottleneck?
Lex Fridman (2:39:14.560)
If you look at the future of currency,
Ben Goertzel (2:39:19.560)
the question is, well...
Lex Fridman (2:39:21.040)
Currency, the main bottleneck is politics.
Ben Goertzel (2:39:23.800)
It's governments and the bands of armed thugs
Lex Fridman (2:39:26.440)
that will shoot you if you bypass their currency restriction.
Ben Goertzel (2:39:29.840)
That's right.
Lex Fridman (2:39:30.680)
So like your sense is that versus the technical challenges,
Ben Goertzel (2:39:33.760)
because you kind of just suggested
Lex Fridman (2:39:34.840)
the technical challenges are quite high as well.
Ben Goertzel (2:39:36.560)
I mean, for making a distributed money,
Lex Fridman (2:39:39.000)
you could do that on Algorand right now.
Ben Goertzel (2:39:41.280)
I mean, so that while Ethereum is too slow,
Lex Fridman (2:39:44.760)
there's Algorand and there's a few other more modern,
Ben Goertzel (2:39:47.240)
more scalable blockchains that would work fine
Lex Fridman (2:39:49.360)
for a decentralized global currency.
Lex Fridman (2:39:53.640)
So I think there were technical bottlenecks
Lex Fridman (2:39:56.480)
to that two years ago.
Lex Fridman (2:39:57.920)
And maybe Ethereum 2.0 will be as fast as Algorand.
Lex Fridman (2:40:00.760)
I don't know, that's not fully written yet, right?
Lex Fridman (2:40:04.160)
So I think the obstacle to currency
Lex Fridman (2:40:07.520)
being put on the blockchain is that...
Ben Goertzel (2:40:09.400)
Is the other stuff you mentioned.
Lex Fridman (2:40:10.240)
I mean, currency will be on the blockchain.
Ben Goertzel (2:40:11.760)
It'll just be on the blockchain in a way
Lex Fridman (2:40:13.840)
that enforces centralized control
Lex Fridman (2:40:16.520)
and government hedge money rather than otherwise.
Lex Fridman (2:40:18.320)
Like the ERNB will probably be the first global,
Ben Goertzel (2:40:20.920)
the first currency on the blockchain.
Lex Fridman (2:40:22.200)
The EURUBIL maybe next.
Ben Goertzel (2:40:23.360)
There are any...
Lex Fridman (2:40:24.200)
EURUBIL?
Ben Goertzel (2:40:25.040)
Yeah, yeah, yeah.
Lex Fridman (2:40:25.860)
I mean, the point is...
Ben Goertzel (2:40:26.700)
Oh, that's hilarious.
Lex Fridman (2:40:27.540)
Digital currency, you know, makes total sense,
Lex Fridman (2:40:30.720)
but they would rather do it in the way
Lex Fridman (2:40:32.160)
that Putin and Xi Jinping have access
Lex Fridman (2:40:34.720)
to the global keys for everything, right?
Lex Fridman (2:40:37.840)
So, and then the analogy to that in terms of SingularityNet,
Ben Goertzel (2:40:42.040)
I mean, there's Echoes.
Lex Fridman (2:40:43.600)
I think you've mentioned before that Linux gives you hope.
Lex Fridman (2:40:47.200)
AI is not as heavily regulated as money, right?
Lex Fridman (2:40:49.960)
Not yet, right?
Ben Goertzel (2:40:51.000)
Not yet.
Lex Fridman (2:40:52.000)
Oh, that's a lot slipperier than money too, right?
Ben Goertzel (2:40:54.240)
I mean, money is easier to regulate
Lex Fridman (2:40:58.280)
because it's kind of easier to define,
Ben Goertzel (2:41:00.800)
whereas AI is, it's almost everywhere inside everything.
Lex Fridman (2:41:04.120)
Where's the boundary between AI and software, right?
Ben Goertzel (2:41:06.440)
I mean, if you're gonna regulate AI,
Lex Fridman (2:41:09.200)
there's no IQ test for every hardware device
Ben Goertzel (2:41:11.720)
that has a learning algorithm.
Lex Fridman (2:41:12.800)
You're gonna be putting like hegemonic regulation
Ben Goertzel (2:41:15.720)
on all software.
Lex Fridman (2:41:16.760)
And I don't rule out that that can happen.
Lex Fridman (2:41:18.880)
And the adaptive software.
Lex Fridman (2:41:21.060)
Yeah, but how do you tell if a software is adaptive
Lex Fridman (2:41:23.360)
and what, every software is gonna be adaptive, I mean.
Lex Fridman (2:41:26.100)
Or maybe they, maybe the, you know,
Ben Goertzel (2:41:28.800)
maybe we're living in the golden age of open source
Lex Fridman (2:41:31.120)
that will not always be open.
Ben Goertzel (2:41:33.360)
Maybe it'll become centralized control
Lex Fridman (2:41:35.640)
of software by governments.
Ben Goertzel (2:41:37.020)
It is entirely possible.
Lex Fridman (2:41:38.840)
And part of what I think we're doing
Ben Goertzel (2:41:42.200)
with things like SingularityNet protocol
Lex Fridman (2:41:45.220)
is creating a tool set that can be used
Ben Goertzel (2:41:50.220)
to counteract that sort of thing.
Lex Fridman (2:41:52.740)
Say a similar thing about mesh networking, right?
Ben Goertzel (2:41:55.620)
Plays a minor role now, the ability to access internet
Lex Fridman (2:41:59.060)
like directly phone to phone.
Ben Goertzel (2:42:01.000)
On the other hand, if your government starts trying
Lex Fridman (2:42:03.740)
to control your use of the internet,
Ben Goertzel (2:42:06.060)
suddenly having mesh networking there
Lex Fridman (2:42:09.220)
can be very convenient, right?
Lex Fridman (2:42:10.800)
And so right now, something like a decentralized
Lex Fridman (2:42:15.360)
blockchain based AGI framework or narrow AI framework,
Ben Goertzel (2:42:20.300)
it's cool, it's nice to have.
Lex Fridman (2:42:22.660)
On the other hand, if governments start trying
Ben Goertzel (2:42:25.140)
to tap down on my AI interoperating
Lex Fridman (2:42:28.740)
with someone's AI in Russia or somewhere, right?
Ben Goertzel (2:42:31.460)
Then suddenly having a decentralized protocol
Lex Fridman (2:42:35.500)
that nobody owns or controls
Ben Goertzel (2:42:37.940)
becomes an extremely valuable part of the tool set.
Lex Fridman (2:42:41.180)
And, you know, we've put that out there now.
Ben Goertzel (2:42:43.780)
It's not perfect, but it operates.
Lex Fridman (2:42:46.980)
And, you know, it's pretty blockchain agnostic.
Lex Fridman (2:42:51.100)
So we're talking to Algorand about making part
Lex Fridman (2:42:53.420)
of SingularityNet run on Algorand.
Ben Goertzel (2:42:56.220)
My good friend Tufi Saliba has a cool blockchain project
Lex Fridman (2:43:00.060)
called Toda, which is a blockchain
Ben Goertzel (2:43:02.220)
without a distributed ledger.
Lex Fridman (2:43:03.540)
It's like a whole other architecture.
Lex Fridman (2:43:05.180)
So there's a lot of more advanced things you can do
Lex Fridman (2:43:08.300)
in the blockchain world.
Ben Goertzel (2:43:09.820)
SingularityNet could be ported to a whole bunch of,
Lex Fridman (2:43:13.500)
it could be made multi chain important
Ben Goertzel (2:43:14.980)
to a whole bunch of different blockchains.
Lex Fridman (2:43:17.100)
And there's a lot of potential and a lot of importance
Ben Goertzel (2:43:21.540)
to putting this kind of tool set out there.
Lex Fridman (2:43:23.620)
If you compare to OpenCog, what you could see is
Ben Goertzel (2:43:26.660)
OpenCog allows tight integration of a few AI algorithms
Lex Fridman (2:43:32.220)
that share the same knowledge store in real time, in RAM.
Ben Goertzel (2:43:36.860)
SingularityNet allows loose integration
Lex Fridman (2:43:40.900)
of multiple different AIs.
Ben Goertzel (2:43:42.660)
They can share knowledge, but they're mostly not gonna
Lex Fridman (2:43:45.620)
be sharing knowledge in RAM on the same machine.
Lex Fridman (2:43:49.980)
And I think what we're gonna have is a network
Lex Fridman (2:43:53.060)
of network of networks, right?
Ben Goertzel (2:43:54.500)
Like, I mean, you have the knowledge graph
Lex Fridman (2:43:57.260)
inside the OpenCog system,
Lex Fridman (2:44:00.900)
and then you have a network of machines
Lex Fridman (2:44:03.220)
inside a distributed OpenCog mind,
Lex Fridman (2:44:05.900)
but then that OpenCog will interface with other AIs
Lex Fridman (2:44:10.260)
doing deep neural nets or custom biology data analysis
Ben Goertzel (2:44:14.420)
or whatever they're doing in SingularityNet,
Lex Fridman (2:44:17.620)
which is a looser integration of different AIs,
Lex Fridman (2:44:21.020)
some of which may be their own networks, right?
Lex Fridman (2:44:24.060)
And I think at a very loose analogy,
Ben Goertzel (2:44:27.900)
you could see that in the human body.
Lex Fridman (2:44:29.380)
Like the brain has regions like cortex or hippocampus,
Ben Goertzel (2:44:33.820)
which tightly interconnects like cortical columns
Lex Fridman (2:44:36.820)
within the cortex, for example.
Ben Goertzel (2:44:39.140)
Then there's looser connection
Lex Fridman (2:44:40.860)
within the different lobes of the brain,
Lex Fridman (2:44:42.700)
and then the brain interconnects with the endocrine system
Lex Fridman (2:44:45.020)
and different parts of the body even more loosely.
Ben Goertzel (2:44:48.260)
Then your body interacts even more loosely
Lex Fridman (2:44:50.780)
with the other people that you talk to.
Lex Fridman (2:44:53.300)
So you often have networks within networks within networks
Lex Fridman (2:44:56.460)
with progressively looser coupling
Ben Goertzel (2:44:59.340)
as you get higher up in that hierarchy.
Lex Fridman (2:45:02.740)
I mean, you have that in biology,
Ben Goertzel (2:45:03.860)
you have that in the internet as a just networking medium.
Lex Fridman (2:45:08.180)
And I think that's what we're gonna have
Ben Goertzel (2:45:10.940)
in the network of software processes leading to AGI.
Lex Fridman (2:45:15.940)
That's a beautiful way to see the world.
Ben Goertzel (2:45:17.780)
Again, the same similar question is with OpenCog.
Lex Fridman (2:45:21.900)
If somebody wanted to build an AI system
Lex Fridman (2:45:24.620)
and plug into the SingularityNet,
Lex Fridman (2:45:27.020)
what would you recommend?
Ben Goertzel (2:45:28.620)
Yeah, so that's much easier.
Lex Fridman (2:45:30.180)
I mean, OpenCog is still a research system.
Lex Fridman (2:45:33.860)
So it takes some expertise to, and sometimes,
Lex Fridman (2:45:36.660)
we have tutorials, but it's somewhat cognitively
Ben Goertzel (2:45:40.220)
labor intensive to get up to speed on OpenCog.
Lex Fridman (2:45:44.340)
And I mean, what's one of the things we hope to change
Ben Goertzel (2:45:46.620)
with the true AGI OpenCog 2.0 version
Lex Fridman (2:45:49.900)
is just make the learning curve more similar
Ben Goertzel (2:45:52.740)
to TensorFlow or Torch or something.
Lex Fridman (2:45:54.620)
Right now, OpenCog is amazingly powerful,
Lex Fridman (2:45:57.340)
but not simple to deal with.
Lex Fridman (2:46:00.620)
On the other hand, SingularityNet,
Ben Goertzel (2:46:03.700)
as an open platform was developed a little more
Lex Fridman (2:46:08.260)
with usability in mind over the blockchain,
Ben Goertzel (2:46:10.580)
it's still kind of a pain.
Lex Fridman (2:46:11.660)
So I mean, if you're a command line guy,
Ben Goertzel (2:46:14.940)
there's a command line interface.
Lex Fridman (2:46:16.180)
It's quite easy to take any AI that has an API
Lex Fridman (2:46:20.060)
and lives in a Docker container and put it online anywhere.
Lex Fridman (2:46:23.540)
And then it joins the global SingularityNet.
Lex Fridman (2:46:25.740)
And anyone who puts a request for services
Lex Fridman (2:46:28.980)
out into the SingularityNet,
Ben Goertzel (2:46:30.180)
the peer to peer discovery mechanism will find
Lex Fridman (2:46:32.340)
your AI and if it does what was asked,
Ben Goertzel (2:46:35.740)
it can then start a conversation with your AI
Lex Fridman (2:46:38.980)
about whether it wants to ask your AI to do something for it,
Lex Fridman (2:46:42.180)
how much it would cost and so on.
Lex Fridman (2:46:43.580)
So that's fairly simple.
Ben Goertzel (2:46:46.860)
If you wrote an AI and want it listed
Lex Fridman (2:46:50.380)
on like official SingularityNet marketplace,
Ben Goertzel (2:46:53.020)
which is on our website,
Lex Fridman (2:46:55.140)
then we have a publisher portal
Lex Fridman (2:46:57.820)
and then there's a KYC process to go through
Lex Fridman (2:47:00.220)
because then we have some legal liability
Ben Goertzel (2:47:02.420)
for what goes on that website.
Lex Fridman (2:47:04.700)
So in a way that's been an education too.
Ben Goertzel (2:47:07.340)
There's sort of two layers.
Lex Fridman (2:47:08.420)
Like there's the open decentralized protocol.
Lex Fridman (2:47:11.700)
And there's the market.
Lex Fridman (2:47:12.980)
Yeah, anyone can use the open decentralized protocol.
Lex Fridman (2:47:15.540)
So say some developers from Iran
Lex Fridman (2:47:17.980)
and there's brilliant AI guys
Ben Goertzel (2:47:19.460)
in University of Isfahan in Tehran,
Lex Fridman (2:47:21.780)
they can put their stuff on SingularityNet protocol
Lex Fridman (2:47:24.660)
and just like they can put something on the internet, right?
Lex Fridman (2:47:27.100)
I don't control it.
Lex Fridman (2:47:28.460)
But if we're gonna list something
Lex Fridman (2:47:29.740)
on the SingularityNet marketplace
Lex Fridman (2:47:32.020)
and put a little picture and a link to it,
Lex Fridman (2:47:34.300)
then if I put some Iranian AI geniuses code on there,
Ben Goertzel (2:47:38.860)
then Donald Trump can send a bunch of jackbooted thugs
Lex Fridman (2:47:41.500)
to my house to arrest me for doing business with Iran, right?
Ben Goertzel (2:47:45.300)
So, I mean, we already see in some ways
Lex Fridman (2:47:48.980)
the value of having a decentralized protocol
Ben Goertzel (2:47:51.100)
because what I hope is that someone in Iran
Lex Fridman (2:47:53.740)
will put online an Iranian SingularityNet marketplace, right?
Ben Goertzel (2:47:57.340)
Which you can pay in the cryptographic token,
Lex Fridman (2:47:59.700)
which is not owned by any country.
Lex Fridman (2:48:01.540)
And then if you're in like Congo or somewhere
Lex Fridman (2:48:04.620)
that doesn't have any problem with Iran,
Ben Goertzel (2:48:06.780)
you can subcontract AI services
Lex Fridman (2:48:09.220)
that you find on that marketplace, right?
Ben Goertzel (2:48:11.980)
Even though US citizens can't by US law.
Lex Fridman (2:48:16.060)
So right now, that's kind of a minor point.
Ben Goertzel (2:48:20.140)
As you alluded, if regulations go in the wrong direction,
Lex Fridman (2:48:24.020)
it could become more of a major point.
Lex Fridman (2:48:25.540)
But I think it also is the case
Lex Fridman (2:48:28.060)
that having these workarounds to regulations in place
Ben Goertzel (2:48:31.860)
is a defense mechanism against those regulations
Lex Fridman (2:48:35.180)
being put into place.
Lex Fridman (2:48:36.660)
And you can see that in the music industry, right?
Lex Fridman (2:48:39.220)
I mean, Napster just happened and BitTorrent just happened.
Lex Fridman (2:48:43.020)
And now most people in my kid's generation,
Lex Fridman (2:48:45.980)
they're baffled by the idea of paying for music, right?
Ben Goertzel (2:48:48.500)
I mean, my dad pays for music.
Lex Fridman (2:48:51.380)
I mean, but that because these decentralized mechanisms
Lex Fridman (2:48:55.700)
happened and then the regulations followed, right?
Lex Fridman (2:48:58.940)
And the regulations would be very different
Ben Goertzel (2:49:01.220)
if they'd been put into place before there was Napster
Lex Fridman (2:49:04.380)
and BitTorrent and so forth.
Lex Fridman (2:49:05.500)
So in the same way, we gotta put AI out there
Lex Fridman (2:49:08.620)
in a decentralized vein and big data out there
Ben Goertzel (2:49:11.060)
in a decentralized vein now,
Lex Fridman (2:49:13.780)
so that the most advanced AI in the world
Ben Goertzel (2:49:16.300)
is fundamentally decentralized.
Lex Fridman (2:49:18.300)
And if that's the case, that's just the reality
Ben Goertzel (2:49:20.940)
the regulators have to deal with.
Lex Fridman (2:49:23.740)
And then as in the music case,
Ben Goertzel (2:49:25.460)
they're gonna come up with regulations
Lex Fridman (2:49:27.460)
that sort of work with the decentralized reality.
Ben Goertzel (2:49:32.860)
Beautiful.
Lex Fridman (2:49:34.020)
You are the chief scientist of Hanson Robotics.
Ben Goertzel (2:49:37.980)
You're still involved with Hanson Robotics,
Lex Fridman (2:49:40.500)
doing a lot of really interesting stuff there.
Ben Goertzel (2:49:42.740)
This is for people who don't know the company
Lex Fridman (2:49:44.500)
that created Sophia the Robot.
Lex Fridman (2:49:47.380)
Can you tell me who Sophia is?
Lex Fridman (2:49:51.460)
I'd rather start by telling you who David Hanson is.
Ben Goertzel (2:49:54.140)
Because David is the brilliant mind behind the Sophia Robot.
Lex Fridman (2:49:58.780)
And he remains, so far, he remains more interesting
Ben Goertzel (2:50:01.980)
than his creation, although she may be improving
Lex Fridman (2:50:05.900)
faster than he is, actually.
Ben Goertzel (2:50:07.380)
I mean, he's a...
Lex Fridman (2:50:08.780)
So yeah, I met David maybe 2007 or something
Ben Goertzel (2:50:15.300)
at some futurist conference we were both speaking at.
Lex Fridman (2:50:18.420)
And I could see we had a great deal in common.
Ben Goertzel (2:50:22.860)
I mean, we were both kind of crazy,
Lex Fridman (2:50:25.020)
but we both had a passion for AGI and the singularity.
Lex Fridman (2:50:31.540)
And we were both huge fans of the work
Lex Fridman (2:50:33.580)
of Philip K. Dick, the science fiction writer.
Lex Fridman (2:50:36.900)
And I wanted to create benevolent AGI
Lex Fridman (2:50:40.780)
that would create massively better life
Ben Goertzel (2:50:44.820)
for all humans and all sentient beings,
Lex Fridman (2:50:47.580)
including animals, plants, and superhuman beings.
Lex Fridman (2:50:50.060)
And David, he wanted exactly the same thing,
Lex Fridman (2:50:53.780)
but he had a different idea of how to do it.
Ben Goertzel (2:50:56.380)
He wanted to get computational compassion.
Lex Fridman (2:50:59.420)
Like he wanted to get machines that would love people
Lex Fridman (2:51:03.940)
and empathize with people.
Lex Fridman (2:51:05.820)
And he thought the way to do that was to make a machine
Ben Goertzel (2:51:08.220)
that could look people eye to eye, face to face,
Lex Fridman (2:51:12.220)
look at people and make people love the machine,
Lex Fridman (2:51:15.700)
and the machine loves the people back.
Lex Fridman (2:51:17.540)
So I thought that was very different way of looking at it
Ben Goertzel (2:51:21.500)
because I'm very math oriented.
Lex Fridman (2:51:22.940)
And I'm just thinking like,
Lex Fridman (2:51:24.740)
what is the abstract cognitive algorithm
Lex Fridman (2:51:28.100)
that will let the system, you know,
Ben Goertzel (2:51:29.420)
internalize the complex patterns of human values,
Lex Fridman (2:51:32.580)
blah, blah, blah.
Ben Goertzel (2:51:33.420)
Whereas he's like, look you in the face and the eye
Lex Fridman (2:51:35.980)
and love you, right?
Lex Fridman (2:51:37.380)
So we hit it off quite well.
Lex Fridman (2:51:41.340)
And we talked to each other off and on.
Ben Goertzel (2:51:44.460)
Then I moved to Hong Kong in 2011.
Lex Fridman (2:51:49.380)
So I've been living all over the place.
Ben Goertzel (2:51:53.380)
I've been in Australia and New Zealand in my academic career.
Lex Fridman (2:51:56.780)
Then in Las Vegas for a while.
Ben Goertzel (2:51:59.380)
Was in New York in the late 90s
Lex Fridman (2:52:00.860)
starting my entrepreneurial career.
Ben Goertzel (2:52:03.660)
Was in DC for nine years
Lex Fridman (2:52:05.020)
doing a bunch of US government consulting stuff.
Ben Goertzel (2:52:07.940)
Then moved to Hong Kong in 2011,
Lex Fridman (2:52:12.060)
mostly because I met a Chinese girl
Ben Goertzel (2:52:13.900)
who I fell in love with and we got married.
Lex Fridman (2:52:16.060)
She's actually not from Hong Kong.
Ben Goertzel (2:52:17.380)
She's from mainland China,
Lex Fridman (2:52:18.380)
but we converged together in Hong Kong.
Ben Goertzel (2:52:21.340)
Still married now, I have a two year old baby.
Lex Fridman (2:52:24.180)
So went to Hong Kong to see about a girl, I guess.
Ben Goertzel (2:52:26.820)
Yeah, pretty much, yeah.
Lex Fridman (2:52:29.060)
And on the other hand,
Ben Goertzel (2:52:31.060)
I started doing some cool research there
Lex Fridman (2:52:33.100)
with Gino Yu at Hong Kong Polytechnic University.
Ben Goertzel (2:52:36.540)
I got involved with a project called IDEA
Lex Fridman (2:52:38.300)
using machine learning for stock and futures prediction,
Ben Goertzel (2:52:41.220)
which was quite interesting.
Lex Fridman (2:52:43.140)
And I also got to know something
Ben Goertzel (2:52:45.100)
about the consumer electronics
Lex Fridman (2:52:47.420)
and hardware manufacturer ecosystem in Shenzhen
Ben Goertzel (2:52:50.220)
across the border,
Lex Fridman (2:52:51.060)
which is like the only place in the world
Ben Goertzel (2:52:53.260)
that makes sense to make complex consumer electronics
Lex Fridman (2:52:56.500)
at large scale and low cost.
Ben Goertzel (2:52:57.860)
It's just, it's astounding the hardware ecosystem
Lex Fridman (2:53:00.900)
that you have in South China.
Ben Goertzel (2:53:03.220)
Like US people here cannot imagine what it's like.
Lex Fridman (2:53:07.220)
So David was starting to explore that also.
Ben Goertzel (2:53:12.060)
I invited him to Hong Kong to give a talk
Lex Fridman (2:53:13.860)
at Hong Kong PolyU,
Lex Fridman (2:53:15.660)
and I introduced him in Hong Kong to some investors
Lex Fridman (2:53:19.220)
who were interested in his robots.
Lex Fridman (2:53:21.580)
And he didn't have Sophia then,
Lex Fridman (2:53:23.540)
he had a robot of Philip K. Dick,
Ben Goertzel (2:53:25.140)
our favorite science fiction writer.
Lex Fridman (2:53:26.980)
He had a robot Einstein,
Ben Goertzel (2:53:28.180)
he had some little toy robots
Lex Fridman (2:53:29.540)
that looked like his son Zeno.
Lex Fridman (2:53:31.940)
So through the investors I connected him to,
Lex Fridman (2:53:35.620)
he managed to get some funding
Ben Goertzel (2:53:37.500)
to basically port Hanson Robotics to Hong Kong.
Lex Fridman (2:53:40.660)
And when he first moved to Hong Kong,
Ben Goertzel (2:53:42.660)
I was working on AGI research
Lex Fridman (2:53:45.300)
and also on this machine learning trading project.
Lex Fridman (2:53:49.340)
So I didn't get that tightly involved
Lex Fridman (2:53:50.940)
with Hanson Robotics.
Lex Fridman (2:53:52.980)
But as I hung out with David more and more,
Lex Fridman (2:53:56.540)
as we were both there in the same place,
Ben Goertzel (2:53:59.180)
I started to get,
Lex Fridman (2:54:01.260)
I started to think about what you could do
Ben Goertzel (2:54:04.620)
to make his robots smarter than they were.
Lex Fridman (2:54:08.500)
And so we started working together
Lex Fridman (2:54:10.340)
and for a few years I was chief scientist
Lex Fridman (2:54:12.780)
and head of software at Hanson Robotics.
Ben Goertzel (2:54:15.740)
Then when I got deeply into the blockchain side of things,
Lex Fridman (2:54:19.420)
I stepped back from that and cofounded Singularity Net.
Ben Goertzel (2:54:24.340)
David Hanson was also one of the cofounders
Lex Fridman (2:54:26.340)
of Singularity Net.
Lex Fridman (2:54:27.780)
So part of our goal there had been
Lex Fridman (2:54:30.060)
to make the blockchain based like cloud mind platform
Ben Goertzel (2:54:33.940)
for Sophia and the other Hanson robots.
Lex Fridman (2:54:37.020)
Sophia would be just one of the robots in Singularity Net.
Ben Goertzel (2:54:41.780)
Yeah, yeah, yeah, exactly.
Lex Fridman (2:54:43.300)
Sophia, many copies of the Sophia robot
Ben Goertzel (2:54:47.380)
would be among the user interfaces
Lex Fridman (2:54:51.500)
to the globally distributed Singularity Net cloud mind.
Lex Fridman (2:54:54.420)
And I mean, David and I talked about that
Lex Fridman (2:54:57.140)
for quite a while before cofounding Singularity Net.
Ben Goertzel (2:55:01.540)
By the way, in his vision and your vision,
Lex Fridman (2:55:04.380)
was Sophia tightly coupled to a particular AI system
Ben Goertzel (2:55:09.580)
or was the idea that you can plug,
Lex Fridman (2:55:11.660)
you could just keep plugging in different AI systems
Lex Fridman (2:55:14.140)
within the head of it?
Lex Fridman (2:55:15.100)
David's view was always that Sophia would be a platform,
Ben Goertzel (2:55:22.940)
much like say the Pepper robot is a platform from SoftBank.
Lex Fridman (2:55:26.820)
Should be a platform with a set of nicely designed APIs
Ben Goertzel (2:55:31.660)
that anyone can use to experiment
Lex Fridman (2:55:33.540)
with their different AI algorithms on that platform.
Lex Fridman (2:55:38.620)
And Singularity Net, of course, fits right into that, right?
Lex Fridman (2:55:41.580)
Because Singularity Net, it's an API marketplace.
Lex Fridman (2:55:44.060)
So anyone can put their AI on there.
Lex Fridman (2:55:46.220)
OpenCog is a little bit different.
Ben Goertzel (2:55:49.020)
I mean, David likes it, but I'd say it's my thing.
Lex Fridman (2:55:52.140)
It's not his.
Ben Goertzel (2:55:52.980)
Like David has a little more passion
Lex Fridman (2:55:55.100)
for biologically based approaches to AI than I do,
Ben Goertzel (2:55:58.700)
which makes sense.
Lex Fridman (2:56:00.140)
I mean, he's really into human physiology and biology.
Lex Fridman (2:56:02.860)
He's a character sculptor, right?
Lex Fridman (2:56:05.140)
So yeah, he's interested in,
Lex Fridman (2:56:07.860)
but he also worked a lot with rule based
Lex Fridman (2:56:09.700)
and logic based AI systems too.
Lex Fridman (2:56:11.420)
So yeah, he's interested in not just Sophia,
Lex Fridman (2:56:14.860)
but all the Hanson robots as a powerful social
Lex Fridman (2:56:17.780)
and emotional robotics platform.
Lex Fridman (2:56:21.220)
And what I saw in Sophia was a way to get AI algorithms
Ben Goertzel (2:56:26.220)
was a way to get AI algorithms out there
Lex Fridman (2:56:32.140)
in front of a whole lot of different people
Ben Goertzel (2:56:34.660)
in an emotionally compelling way.
Lex Fridman (2:56:36.300)
And part of my thought was really kind of abstract
Ben Goertzel (2:56:39.820)
connected to AGI ethics.
Lex Fridman (2:56:41.740)
And many people are concerned AGI is gonna enslave everybody
Ben Goertzel (2:56:46.940)
or turn everybody into computronium
Lex Fridman (2:56:50.060)
to make extra hard drives for their cognitive engine
Ben Goertzel (2:56:54.740)
or whatever.
Lex Fridman (2:56:55.580)
And emotionally I'm not driven to that sort of paranoia.
Ben Goertzel (2:57:01.660)
I'm really just an optimist by nature,
Lex Fridman (2:57:04.100)
but intellectually I have to assign a non zero probability
Ben Goertzel (2:57:09.220)
to those sorts of nasty outcomes.
Lex Fridman (2:57:12.140)
Cause if you're making something 10 times as smart as you,
Lex Fridman (2:57:14.900)
how can you know what it's gonna do?
Lex Fridman (2:57:16.300)
There's an irreducible uncertainty there
Ben Goertzel (2:57:19.780)
just as my dog can't predict what I'm gonna do tomorrow.
Lex Fridman (2:57:22.780)
So it seemed to me that based on our current state
Ben Goertzel (2:57:26.420)
of knowledge, the best way to bias the AGI as we create
Lex Fridman (2:57:32.500)
toward benevolence would be to infuse them with love
Lex Fridman (2:57:38.820)
and compassion the way that we do our own children.
Lex Fridman (2:57:41.620)
So you want to interact with AIs in the context
Ben Goertzel (2:57:45.820)
of doing compassionate, loving and beneficial things.
Lex Fridman (2:57:49.900)
And in that way, as your children will learn
Ben Goertzel (2:57:52.140)
by doing compassionate, beneficial,
Lex Fridman (2:57:53.740)
loving things alongside you.
Lex Fridman (2:57:55.940)
And that way the AI will learn in practice
Lex Fridman (2:57:58.660)
what it means to be compassionate, beneficial and loving.
Ben Goertzel (2:58:02.340)
It will get a sort of ingrained intuitive sense of this,
Lex Fridman (2:58:06.380)
which it can then abstract in its own way
Ben Goertzel (2:58:09.260)
as it gets more and more intelligent.
Lex Fridman (2:58:11.180)
Now, David saw this the same way.
Ben Goertzel (2:58:12.780)
That's why he came up with the name Sophia,
Lex Fridman (2:58:15.540)
which means wisdom.
Lex Fridman (2:58:18.140)
So it seemed to me making these beautiful, loving robots
Lex Fridman (2:58:22.780)
to be rolled out for beneficial applications
Ben Goertzel (2:58:26.060)
would be the perfect way to roll out early stage AGI systems
Lex Fridman (2:58:31.260)
so they can learn from people
Lex Fridman (2:58:33.940)
and not just learn factual knowledge,
Lex Fridman (2:58:35.420)
but learn human values and ethics from people
Ben Goertzel (2:58:38.580)
while being their home service robots,
Lex Fridman (2:58:41.540)
their education assistants, their nursing robots.
Lex Fridman (2:58:44.100)
So that was the grand vision.
Lex Fridman (2:58:46.060)
Now, if you've ever worked with robots,
Lex Fridman (2:58:48.620)
the reality is quite different, right?
Lex Fridman (2:58:50.420)
Like the first principle is the robot is always broken.
Ben Goertzel (2:58:53.220)
I mean, I worked with robots in the 90s a bunch
Lex Fridman (2:58:57.660)
when you had to solder them together yourself
Lex Fridman (2:58:59.540)
and I'd put neural nets during reinforcement learning
Lex Fridman (2:59:02.580)
on like overturned solid ball type robots
Lex Fridman (2:59:05.940)
and in the 90s when I was a professor.
Lex Fridman (2:59:09.300)
Things of course advanced a lot, but...
Lex Fridman (2:59:12.020)
But the principle still holds.
Lex Fridman (2:59:13.180)
The principle that the robot's always broken still holds.
Ben Goertzel (2:59:16.500)
Yeah, so faced with the reality of making Sophia do stuff,
Lex Fridman (2:59:21.020)
many of my robo AGI aspirations were temporarily cast aside.
Lex Fridman (2:59:26.620)
And I mean, there's just a practical problem
Lex Fridman (2:59:30.660)
of making this robot interact in a meaningful way
Ben Goertzel (2:59:33.700)
because like, you put nice computer vision on there,
Lex Fridman (2:59:36.700)
but there's always glare.
Lex Fridman (2:59:38.140)
And then, or you have a dialogue system,
Lex Fridman (2:59:41.420)
but at the time I was there,
Ben Goertzel (2:59:43.740)
like no speech to text algorithm could deal
Lex Fridman (2:59:46.580)
with Hong Kongese people's English accents.
Lex Fridman (2:59:49.780)
So the speech to text was always bad.
Lex Fridman (2:59:51.620)
So the robot always sounded stupid
Lex Fridman (2:59:53.620)
because it wasn't getting the right text, right?
Lex Fridman (2:59:55.620)
So I started to view that really
Ben Goertzel (2:59:58.020)
as what in software engineering you call a walking skeleton,
Lex Fridman (30:00.900)
What do you think he would say about AI?
Ben Goertzel (30:02.980)
I mean. Well, those are quite different.
Lex Fridman (30:04.180)
If he's born a century later or transported through time.
Ben Goertzel (30:07.260)
Well, he'd be on like TikTok and Instagram
Lex Fridman (30:09.580)
and he would never write the great works he's written.
Lex Fridman (30:11.940)
So let's transport him through time.
Lex Fridman (30:13.540)
Maybe also Sprach Zarathustra would be a music video,
Lex Fridman (30:16.460)
right? I mean, who knows?
Lex Fridman (30:19.660)
Yeah, but if he was transported through time,
Lex Fridman (30:21.700)
do you think, that'd be interesting actually to go back.
Lex Fridman (30:26.260)
You just made me realize that it's possible to go back
Lex Fridman (30:29.380)
and read Nietzsche with an eye of,
Lex Fridman (30:31.220)
is there some thinking about artificial beings?
Ben Goertzel (30:34.700)
I'm sure there he had inklings.
Lex Fridman (30:37.780)
I mean, with Frankenstein before him,
Ben Goertzel (30:40.500)
I'm sure he had inklings of artificial beings
Lex Fridman (30:42.900)
somewhere in the text.
Ben Goertzel (30:44.060)
It'd be interesting to try to read his work
Lex Fridman (30:46.900)
to see if Superman was actually an AGI system.
Ben Goertzel (30:55.820)
Like if he had inklings of that kind of thinking.
Lex Fridman (30:57.940)
He didn't.
Ben Goertzel (30:58.780)
He didn't.
Lex Fridman (30:59.620)
No, I would say not.
Ben Goertzel (31:01.100)
I mean, he had a lot of inklings of modern cognitive science,
Lex Fridman (31:06.460)
which are very interesting.
Ben Goertzel (31:07.420)
If you look in like the third part of the collection
Lex Fridman (31:11.820)
that's been titled The Will to Power.
Ben Goertzel (31:13.540)
I mean, in book three there,
Lex Fridman (31:15.660)
there's very deep analysis of thinking processes,
Lex Fridman (31:20.620)
but he wasn't so much of a physical tinkerer type guy,
Lex Fridman (31:27.140)
right? He was very abstract.
Lex Fridman (31:29.620)
Do you think, what do you think about the will to power?
Lex Fridman (31:32.780)
Do you think human, what do you think drives humans?
Lex Fridman (31:36.100)
Is it?
Lex Fridman (31:37.460)
Oh, an unholy mix of things.
Ben Goertzel (31:39.500)
I don't think there's one pure, simple,
Lex Fridman (31:42.380)
and elegant objective function driving humans by any means.
Lex Fridman (31:47.380)
What do you think, if we look at,
Lex Fridman (31:50.700)
I know it's hard to look at humans in an aggregate,
Lex Fridman (31:53.260)
but do you think overall humans are good?
Lex Fridman (31:57.540)
Or do we have both good and evil within us
Ben Goertzel (32:01.580)
that depending on the circumstances,
Lex Fridman (32:03.540)
depending on whatever can percolate to the top?
Ben Goertzel (32:08.220)
Good and evil are very ambiguous, complicated
Lex Fridman (32:13.900)
and in some ways silly concepts.
Lex Fridman (32:15.900)
But if we could dig into your question
Lex Fridman (32:18.540)
from a couple of directions.
Lex Fridman (32:19.700)
So I think if you look in evolution,
Lex Fridman (32:23.420)
humanity is shaped both by individual selection
Lex Fridman (32:28.220)
and what biologists would call group selection,
Lex Fridman (32:30.940)
like tribe level selection, right?
Lex Fridman (32:32.740)
So individual selection has driven us
Lex Fridman (32:36.500)
in a selfish DNA sort of way.
Lex Fridman (32:38.780)
So that each of us does to a certain approximation
Lex Fridman (32:43.260)
what will help us propagate our DNA to future generations.
Ben Goertzel (32:47.420)
I mean, that's why I've got four kids so far
Lex Fridman (32:50.700)
and probably that's not the last one.
Ben Goertzel (32:53.900)
On the other hand.
Lex Fridman (32:55.020)
I like the ambition.
Ben Goertzel (32:56.780)
Tribal, like group selection means humans in a way
Lex Fridman (33:00.740)
will do what will advocate for the persistence of the DNA
Ben Goertzel (33:04.380)
of their whole tribe or their social group.
Lex Fridman (33:08.100)
And in biology, you have both of these, right?
Lex Fridman (33:11.740)
And you can see, say an ant colony or a beehive,
Lex Fridman (33:14.420)
there's a lot of group selection
Ben Goertzel (33:15.940)
in the evolution of those social animals.
Lex Fridman (33:18.940)
On the other hand, say a big cat
Ben Goertzel (33:21.460)
or some very solitary animal,
Lex Fridman (33:23.260)
it's a lot more biased toward individual selection.
Ben Goertzel (33:26.540)
Humans are an interesting balance.
Lex Fridman (33:28.660)
And I think this reflects itself
Ben Goertzel (33:31.540)
in what we would view as selfishness versus altruism
Lex Fridman (33:35.060)
to some extent.
Lex Fridman (33:36.780)
So we just have both of those objective functions
Lex Fridman (33:40.580)
contributing to the makeup of our brains.
Lex Fridman (33:43.780)
And then as Nietzsche analyzed in his own way
Lex Fridman (33:47.300)
and others have analyzed in different ways,
Ben Goertzel (33:49.060)
I mean, we abstract this as well,
Lex Fridman (33:51.500)
we have both good and evil within us, right?
Ben Goertzel (33:55.380)
Because a lot of what we view as evil
Lex Fridman (33:57.820)
is really just selfishness.
Ben Goertzel (34:00.460)
A lot of what we view as good is altruism,
Lex Fridman (34:03.740)
which means doing what's good for the tribe.
Lex Fridman (34:07.220)
And on that level,
Lex Fridman (34:08.060)
we have both of those just baked into us
Lex Fridman (34:11.380)
and that's how it is.
Lex Fridman (34:13.180)
Of course, there are psychopaths and sociopaths
Lex Fridman (34:17.020)
and people who get gratified by the suffering of others.
Lex Fridman (34:21.340)
And that's a different thing.
Ben Goertzel (34:25.260)
Yeah, those are exceptions on the whole.
Lex Fridman (34:27.500)
But I think at core, we're not purely selfish,
Ben Goertzel (34:31.540)
we're not purely altruistic, we are a mix
Lex Fridman (34:35.180)
and that's the nature of it.
Lex Fridman (34:38.020)
And we also have a complex constellation of values
Lex Fridman (34:43.380)
that are just very specific to our evolutionary history.
Ben Goertzel (34:49.180)
Like we love waterways and mountains
Lex Fridman (34:52.500)
and the ideal place to put a house
Lex Fridman (34:54.460)
is in a mountain overlooking the water, right?
Lex Fridman (34:56.340)
And we care a lot about our kids
Lex Fridman (35:00.580)
and we care a little less about our cousins
Lex Fridman (35:02.820)
and even less about our fifth cousins.
Ben Goertzel (35:04.420)
I mean, there are many particularities to human values,
Lex Fridman (35:09.460)
which whether they're good or evil
Ben Goertzel (35:11.900)
depends on your perspective.
Lex Fridman (35:15.820)
Say, I spent a lot of time in Ethiopia in Addis Ababa
Ben Goertzel (35:19.660)
where we have one of our AI development offices
Lex Fridman (35:22.460)
for my SingularityNet project.
Lex Fridman (35:24.420)
And when I walk through the streets in Addis,
Lex Fridman (35:27.540)
you know, there's people lying by the side of the road,
Ben Goertzel (35:31.460)
like just living there by the side of the road,
Lex Fridman (35:33.940)
dying probably of curable diseases
Ben Goertzel (35:35.820)
without enough food or medicine.
Lex Fridman (35:37.940)
And when I walk by them, you know, I feel terrible,
Ben Goertzel (35:39.980)
I give them money.
Lex Fridman (35:41.460)
When I come back home to the developed world,
Ben Goertzel (35:45.100)
they're not on my mind that much.
Lex Fridman (35:46.620)
I do donate some, but I mean,
Ben Goertzel (35:48.620)
I also spend some of the limited money I have
Lex Fridman (35:52.860)
enjoying myself in frivolous ways
Ben Goertzel (35:54.700)
rather than donating it to those people who are right now,
Lex Fridman (35:58.100)
like starving, dying and suffering on the roadside.
Lex Fridman (36:01.020)
So does that make me evil?
Lex Fridman (36:03.180)
I mean, it makes me somewhat selfish
Lex Fridman (36:05.500)
and somewhat altruistic.
Lex Fridman (36:06.740)
And we each balance that in our own way, right?
Lex Fridman (36:10.940)
So whether that will be true of all possible AGI's
Lex Fridman (36:17.060)
is a subtler question.
Lex Fridman (36:19.300)
So that's how humans are.
Lex Fridman (36:21.340)
So you have a sense, you kind of mentioned
Ben Goertzel (36:23.100)
that there's a selfish,
Lex Fridman (36:25.500)
I'm not gonna bring up the whole Ayn Rand idea
Ben Goertzel (36:28.300)
of selfishness being the core virtue.
Lex Fridman (36:31.140)
That's a whole interesting kind of tangent
Ben Goertzel (36:33.980)
that I think we'll just distract ourselves on.
Lex Fridman (36:36.420)
I have to make one amusing comment.
Ben Goertzel (36:38.460)
Sure.
Lex Fridman (36:39.300)
A comment that has amused me anyway.
Lex Fridman (36:41.260)
So the, yeah, I have extraordinary negative respect
Lex Fridman (36:46.340)
for Ayn Rand.
Lex Fridman (36:47.820)
Negative, what's a negative respect?
Lex Fridman (36:50.220)
But when I worked with a company called Genescient,
Ben Goertzel (36:54.740)
which was evolving flies to have extraordinary long lives
Lex Fridman (36:59.180)
in Southern California.
Lex Fridman (37:01.220)
So we had flies that were evolved by artificial selection
Lex Fridman (37:04.980)
to have five times the lifespan of normal fruit flies.
Lex Fridman (37:07.660)
But the population of super long lived flies
Lex Fridman (37:11.780)
was physically sitting in a spare room
Ben Goertzel (37:14.060)
at an Ayn Rand elementary school in Southern California.
Lex Fridman (37:18.100)
So that was just like,
Ben Goertzel (37:19.460)
well, if I saw this in a movie, I wouldn't believe it.
Lex Fridman (37:23.980)
Well, yeah, the universe has a sense of humor
Ben Goertzel (37:26.020)
in that kind of way.
Lex Fridman (37:26.860)
That fits in, humor fits in somehow
Ben Goertzel (37:28.900)
into this whole absurd existence.
Lex Fridman (37:30.620)
But you mentioned the balance between selfishness
Lex Fridman (37:33.820)
and altruism as kind of being innate.
Lex Fridman (37:37.220)
Do you think it's possible
Ben Goertzel (37:38.140)
that's kind of an emergent phenomena,
Lex Fridman (37:42.380)
those peculiarities of our value system?
Lex Fridman (37:45.420)
How much of it is innate?
Lex Fridman (37:47.180)
How much of it is something we collectively
Ben Goertzel (37:49.780)
kind of like a Dostoevsky novel
Lex Fridman (37:52.300)
bring to life together as a civilization?
Ben Goertzel (37:54.540)
I mean, the answer to nature versus nurture
Lex Fridman (37:57.740)
is usually both.
Lex Fridman (37:58.860)
And of course it's nature versus nurture
Lex Fridman (38:01.820)
versus self organization, as you mentioned.
Lex Fridman (38:04.780)
So clearly there are evolutionary roots
Lex Fridman (38:08.460)
to individual and group selection
Ben Goertzel (38:11.460)
leading to a mix of selfishness and altruism.
Lex Fridman (38:13.900)
On the other hand,
Ben Goertzel (38:15.380)
different cultures manifest that in different ways.
Lex Fridman (38:19.780)
Well, we all have basically the same biology.
Lex Fridman (38:22.540)
And if you look at sort of precivilized cultures,
Lex Fridman (38:26.660)
you have tribes like the Yanomamo in Venezuela,
Ben Goertzel (38:29.340)
which their culture is focused on killing other tribes.
Lex Fridman (38:35.340)
And you have other Stone Age tribes
Ben Goertzel (38:37.620)
that are mostly peaceful and have big taboos
Lex Fridman (38:40.460)
against violence.
Lex Fridman (38:41.420)
So you can certainly have a big difference
Lex Fridman (38:43.900)
in how culture manifests
Ben Goertzel (38:46.860)
these innate biological characteristics,
Lex Fridman (38:50.820)
but still, there's probably limits
Ben Goertzel (38:54.740)
that are given by our biology.
Lex Fridman (38:56.740)
I used to argue this with my great grandparents
Ben Goertzel (39:00.060)
who were Marxists actually,
Lex Fridman (39:01.500)
because they believed in the withering away of the state.
Ben Goertzel (39:04.540)
Like they believe that,
Lex Fridman (39:06.900)
as you move from capitalism to socialism to communism,
Ben Goertzel (39:10.660)
people would just become more social minded
Lex Fridman (39:13.420)
so that a state would be unnecessary
Lex Fridman (39:15.940)
and everyone would give everyone else what they needed.
Lex Fridman (39:20.940)
Now, setting aside that
Ben Goertzel (39:23.140)
that's not what the various Marxist experiments
Lex Fridman (39:25.740)
on the planet seem to be heading toward in practice.
Ben Goertzel (39:29.900)
Just as a theoretical point,
Lex Fridman (39:32.740)
I was very dubious that human nature could go there.
Ben Goertzel (39:37.540)
Like at that time when my great grandparents are alive,
Lex Fridman (39:39.900)
I was just like, you know, I'm a cynical teenager.
Ben Goertzel (39:43.300)
I think humans are just jerks.
Lex Fridman (39:45.980)
The state is not gonna wither away.
Ben Goertzel (39:48.020)
If you don't have some structure
Lex Fridman (39:49.980)
keeping people from screwing each other over,
Ben Goertzel (39:51.980)
they're gonna do it.
Lex Fridman (39:52.900)
So now I actually don't quite see things that way.
Ben Goertzel (39:56.220)
I mean, I think my feeling now subjectively
Lex Fridman (39:59.900)
is the culture aspect is more significant
Ben Goertzel (3:00:02.820)
which is maybe the wrong metaphor to use for Sophia
Lex Fridman (3:00:05.420)
or maybe the right one.
Ben Goertzel (3:00:06.980)
I mean, where the walking skeleton is
Lex Fridman (3:00:08.420)
in software development is
Lex Fridman (3:00:10.620)
if you're building a complex system, how do you get started?
Lex Fridman (3:00:14.020)
But one way is to first build part one well,
Ben Goertzel (3:00:16.140)
then build part two well, then build part three well
Lex Fridman (3:00:18.340)
and so on.
Lex Fridman (3:00:19.260)
And the other way is you make like a simple version
Lex Fridman (3:00:22.060)
of the whole system and put something in the place
Ben Goertzel (3:00:24.820)
of every part the whole system will need
Lex Fridman (3:00:27.300)
so that you have a whole system that does something.
Lex Fridman (3:00:29.660)
And then you work on improving each part
Lex Fridman (3:00:31.900)
in the context of that whole integrated system.
Lex Fridman (3:00:34.340)
So that's what we did on a software level in Sophia.
Lex Fridman (3:00:38.140)
We made like a walking skeleton software system
Ben Goertzel (3:00:41.580)
where so there's something that sees,
Lex Fridman (3:00:43.100)
there's something that hears, there's something that moves,
Ben Goertzel (3:00:46.220)
there's something that remembers,
Lex Fridman (3:00:48.180)
there's something that learns.
Ben Goertzel (3:00:49.980)
You put a simple version of each thing in there
Lex Fridman (3:00:52.460)
and you connect them all together
Lex Fridman (3:00:54.420)
so that the system will do its thing.
Lex Fridman (3:00:56.660)
So there's a lot of AI in there.
Ben Goertzel (3:00:59.660)
There's not any AGI in there.
Lex Fridman (3:01:01.380)
I mean, there's computer vision to recognize people's faces,
Ben Goertzel (3:01:04.660)
recognize when someone comes in the room and leaves,
Lex Fridman (3:01:07.660)
trying to recognize whether two people are together or not.
Ben Goertzel (3:01:10.740)
I mean, the dialogue system,
Lex Fridman (3:01:13.300)
it's a mix of like hand coded rules with deep neural nets
Ben Goertzel (3:01:18.780)
that come up with their own responses.
Lex Fridman (3:01:21.580)
And there's some attempt to have a narrative structure
Lex Fridman (3:01:25.660)
and sort of try to pull the conversation
Lex Fridman (3:01:28.420)
into something with a beginning, middle and end
Lex Fridman (3:01:30.780)
and this sort of story arc.
Lex Fridman (3:01:32.180)
So it's...
Ben Goertzel (3:01:33.500)
I mean, like if you look at the Lobner Prize and the systems
Lex Fridman (3:01:37.620)
that beat the Turing Test currently,
Ben Goertzel (3:01:39.060)
they're heavily rule based
Lex Fridman (3:01:40.540)
because like you had said, narrative structure
Ben Goertzel (3:01:43.900)
to create compelling conversations,
Lex Fridman (3:01:45.700)
you currently, neural networks cannot do that well,
Ben Goertzel (3:01:48.420)
even with Google MENA.
Lex Fridman (3:01:50.660)
When you actually look at full scale conversations,
Ben Goertzel (3:01:53.060)
it's just not...
Lex Fridman (3:01:53.900)
Yeah, this is the thing.
Lex Fridman (3:01:54.740)
So we've been, I've actually been running an experiment
Lex Fridman (3:01:57.900)
the last couple of weeks taking Sophia's chat bot
Lex Fridman (3:02:01.420)
and then Facebook's Transformer chat bot,
Lex Fridman (3:02:03.740)
which they opened the model.
Ben Goertzel (3:02:05.260)
We've had them chatting to each other
Lex Fridman (3:02:06.780)
for a number of weeks on the server just...
Ben Goertzel (3:02:08.860)
That's funny.
Lex Fridman (3:02:10.020)
We're generating training data of what Sophia says
Ben Goertzel (3:02:13.260)
in a wide variety of conversations.
Lex Fridman (3:02:15.500)
But we can see, compared to Sophia's current chat bot,
Ben Goertzel (3:02:20.260)
the Facebook deep neural chat bot comes up
Lex Fridman (3:02:23.460)
with a wider variety of fluent sounding sentences.
Ben Goertzel (3:02:27.300)
On the other hand, it rambles like mad.
Lex Fridman (3:02:30.100)
The Sophia chat bot, it's a little more repetitive
Ben Goertzel (3:02:33.900)
in the sentence structures it uses.
Lex Fridman (3:02:36.620)
On the other hand, it's able to keep like a conversation arc
Lex Fridman (3:02:39.820)
over a much longer, longer period, right?
Lex Fridman (3:02:42.460)
So there...
Ben Goertzel (3:02:43.300)
Now, you can probably surmount that using Reformer
Lex Fridman (3:02:46.620)
and like using various other deep neural architectures
Ben Goertzel (3:02:51.140)
to improve the way these Transformer models are trained.
Lex Fridman (3:02:53.980)
But in the end, neither one of them really understands
Ben Goertzel (3:02:58.300)
what's going on.
Lex Fridman (3:02:59.140)
I mean, that's the challenge I had with Sophia
Ben Goertzel (3:03:02.660)
is if I were doing a robotics project aimed at AGI,
Lex Fridman (3:03:08.340)
I would wanna make like a robo toddler
Ben Goertzel (3:03:10.100)
that was just learning about what it was seeing.
Lex Fridman (3:03:11.940)
Because then the language is grounded
Ben Goertzel (3:03:13.220)
in the experience of the robot.
Lex Fridman (3:03:14.940)
But what Sophia needs to do to be Sophia
Ben Goertzel (3:03:17.740)
is talk about sports or the weather or robotics
Lex Fridman (3:03:21.420)
or the conference she's talking at.
Ben Goertzel (3:03:24.100)
She needs to be fluent talking about
Lex Fridman (3:03:26.380)
any damn thing in the world.
Lex Fridman (3:03:28.420)
And she doesn't have grounding for all those things.
Lex Fridman (3:03:32.500)
So there's this, just like, I mean, Google Mina
Lex Fridman (3:03:35.700)
and Facebook's chat, but I don't have grounding
Lex Fridman (3:03:37.460)
for what they're talking about either.
Lex Fridman (3:03:40.140)
So in a way, the need to speak fluently about things
Lex Fridman (3:03:45.060)
where there's no nonlinguistic grounding
Ben Goertzel (3:03:47.940)
pushes what you can do for Sophia in the short term
Lex Fridman (3:03:53.660)
a bit away from AGI.
Ben Goertzel (3:03:56.340)
I mean, it pushes you towards IBM Watson situation
Lex Fridman (3:04:00.900)
where you basically have to do heuristic
Lex Fridman (3:04:02.740)
and hard code stuff and rule based stuff.
Lex Fridman (3:04:05.380)
I have to ask you about this, okay.
Lex Fridman (3:04:07.860)
So because in part Sophia is like an art creation
Lex Fridman (3:04:18.860)
because it's beautiful.
Ben Goertzel (3:04:21.260)
She's beautiful because she inspires
Lex Fridman (3:04:24.780)
through our human nature of anthropomorphize things.
Ben Goertzel (3:04:29.540)
We immediately see an intelligent being there.
Lex Fridman (3:04:32.620)
Because David is a great sculptor.
Ben Goertzel (3:04:34.100)
He is a great sculptor, that's right.
Lex Fridman (3:04:35.500)
So in fact, if Sophia just had nothing inside her head,
Ben Goertzel (3:04:40.820)
said nothing, if she just sat there,
Lex Fridman (3:04:43.260)
we already prescribed some intelligence to her.
Ben Goertzel (3:04:45.940)
There's a long selfie line in front of her
Lex Fridman (3:04:47.780)
after every talk.
Ben Goertzel (3:04:48.740)
That's right.
Lex Fridman (3:04:49.940)
So it captivated the imagination of many people.
Ben Goertzel (3:04:53.820)
I wasn't gonna say the world,
Lex Fridman (3:04:54.860)
but yeah, I mean a lot of people.
Ben Goertzel (3:04:58.180)
Billions of people, which is amazing.
Lex Fridman (3:05:00.180)
It's amazing, right.
Ben Goertzel (3:05:01.940)
Now, of course, many people have prescribed
Lex Fridman (3:05:08.260)
essentially AGI type of capabilities to Sophia
Ben Goertzel (3:05:11.060)
when they see her.
Lex Fridman (3:05:12.380)
And of course, friendly French folk like Yann LeCun
Ben Goertzel (3:05:19.860)
immediately see that of the people from the AI community
Lex Fridman (3:05:22.820)
and get really frustrated because...
Ben Goertzel (3:05:25.900)
It's understandable.
Lex Fridman (3:05:27.060)
So what, and then they criticize people like you
Ben Goertzel (3:05:31.700)
who sit back and don't say anything about,
Lex Fridman (3:05:36.100)
like basically allow the imagination of the world,
Ben Goertzel (3:05:39.980)
allow the world to continue being captivated.
Lex Fridman (3:05:43.860)
So what's your sense of that kind of annoyance
Lex Fridman (3:05:49.140)
that the AI community has?
Lex Fridman (3:05:51.220)
I think there's several parts to my reaction there.
Ben Goertzel (3:05:55.380)
First of all, if I weren't involved with Hanson and Box
Lex Fridman (3:05:59.820)
and didn't know David Hanson personally,
Ben Goertzel (3:06:03.420)
I probably would have been very annoyed initially
Lex Fridman (3:06:06.420)
at Sophia as well.
Ben Goertzel (3:06:07.980)
I mean, I can understand the reaction.
Lex Fridman (3:06:09.460)
I would have been like, wait,
Ben Goertzel (3:06:11.820)
all these stupid people out there think this is an AGI,
Lex Fridman (3:06:16.260)
but it's not an AGI, but they're tricking people
Ben Goertzel (3:06:19.980)
that this very cool robot is an AGI.
Lex Fridman (3:06:23.060)
And now those of us trying to raise funding to build AGI,
Ben Goertzel (3:06:28.180)
people will think it's already there and it already works.
Lex Fridman (3:06:31.180)
So on the other hand, I think,
Ben Goertzel (3:06:36.740)
even if I weren't directly involved with it,
Lex Fridman (3:06:38.340)
once I dug a little deeper into David and the robot
Lex Fridman (3:06:41.660)
and the intentions behind it,
Lex Fridman (3:06:43.460)
I think I would have stopped being pissed off.
Ben Goertzel (3:06:47.020)
Whereas folks like Yann LeCun have remained pissed off
Lex Fridman (3:06:51.380)
after their initial reaction.
Ben Goertzel (3:06:54.460)
That's his thing, that's his thing.
Lex Fridman (3:06:56.100)
I think that in particular struck me as somewhat ironic
Ben Goertzel (3:07:01.940)
because Yann LeCun is working for Facebook,
Lex Fridman (3:07:05.620)
which is using machine learning to program the brains
Ben Goertzel (3:07:09.020)
of the people in the world toward vapid consumerism
Lex Fridman (3:07:13.340)
and political extremism.
Lex Fridman (3:07:14.860)
So if your ethics allows you to use machine learning
Lex Fridman (3:07:19.660)
in such a blatantly destructive way,
Lex Fridman (3:07:23.460)
why would your ethics not allow you to use machine learning
Lex Fridman (3:07:26.220)
to make a lovable theatrical robot
Ben Goertzel (3:07:29.780)
that draws some foolish people
Lex Fridman (3:07:32.100)
into its theatrical illusion?
Ben Goertzel (3:07:34.420)
Like if the pushback had come from Yoshua Bengio,
Lex Fridman (3:07:38.780)
I would have felt much more humbled by it
Lex Fridman (3:07:40.900)
because he's not using AI for blatant evil, right?
Lex Fridman (3:07:45.460)
On the other hand, he also is a super nice guy
Lex Fridman (3:07:48.540)
and doesn't bother to go out there
Lex Fridman (3:07:50.860)
trashing other people's work for no good reason, right?
Ben Goertzel (3:07:54.420)
Shots fired, but I get you.
Lex Fridman (3:07:55.940)
I mean, that's...
Ben Goertzel (3:07:58.020)
I mean, if you're gonna ask, I'm gonna answer.
Lex Fridman (3:08:01.100)
No, for sure.
Ben Goertzel (3:08:02.060)
I think we'll go back and forth.
Lex Fridman (3:08:03.300)
I'll talk to Yann again.
Ben Goertzel (3:08:04.500)
I would add on this though.
Lex Fridman (3:08:06.060)
I mean, David Hansen is an artist
Lex Fridman (3:08:11.540)
and he often speaks off the cuff.
Lex Fridman (3:08:14.180)
And I have not agreed with everything
Ben Goertzel (3:08:16.300)
that David has said or done regarding Sophia.
Lex Fridman (3:08:19.300)
And David also has not agreed with everything
Ben Goertzel (3:08:22.740)
David has said or done about Sophia.
Lex Fridman (3:08:24.740)
That's an important point.
Ben Goertzel (3:08:25.780)
I mean, David is an artistic wild man
Lex Fridman (3:08:30.140)
and that's part of his charm.
Ben Goertzel (3:08:33.340)
That's part of his genius.
Lex Fridman (3:08:34.740)
So certainly there have been conversations
Ben Goertzel (3:08:39.380)
within Hansen Robotics and between me and David
Lex Fridman (3:08:42.260)
where I was like, let's be more open
Ben Goertzel (3:08:45.700)
about how this thing is working.
Lex Fridman (3:08:48.180)
And I did have some influence in nudging Hansen Robotics
Ben Goertzel (3:08:52.060)
to be more open about how Sophia was working.
Lex Fridman (3:08:56.740)
And David wasn't especially opposed to this.
Lex Fridman (3:09:00.740)
And he was actually quite right about it.
Lex Fridman (3:09:02.460)
What he said was, you can tell people exactly
Lex Fridman (3:09:04.940)
how it's working and they won't care.
Lex Fridman (3:09:08.020)
They want to be drawn into the illusion.
Lex Fridman (3:09:09.580)
And he was 100% correct.
Lex Fridman (3:09:12.580)
I'll tell you what, this wasn't Sophia.
Ben Goertzel (3:09:14.620)
This was Philip K. Dick.
Lex Fridman (3:09:15.740)
But we did some interactions between humans
Lex Fridman (3:09:18.780)
and Philip K. Dick robot in Austin, Texas a few years back.
Lex Fridman (3:09:23.820)
And in this case, the Philip K. Dick was just teleoperated
Ben Goertzel (3:09:26.700)
by another human in the other room.
Lex Fridman (3:09:28.540)
So during the conversations, we didn't tell people
Ben Goertzel (3:09:31.260)
the robot was teleoperated.
Lex Fridman (3:09:32.860)
We just said, here, have a conversation with Phil Dick.
Lex Fridman (3:09:35.020)
We're gonna film you, right?
Lex Fridman (3:09:37.100)
And they had a great conversation with Philip K. Dick
Ben Goertzel (3:09:39.740)
teleoperated by my friend, Stefan Bugaj.
Lex Fridman (3:09:42.900)
After the conversation, we brought the people
Ben Goertzel (3:09:45.860)
in the back room to see Stefan
Lex Fridman (3:09:47.980)
who was controlling the Philip K. Dick robot,
Lex Fridman (3:09:53.540)
but they didn't believe it.
Lex Fridman (3:09:54.820)
These people were like, well, yeah,
Lex Fridman (3:09:56.500)
but I know I was talking to Phil.
Lex Fridman (3:09:58.780)
Maybe Stefan was typing,
Lex Fridman (3:10:00.780)
but the spirit of Phil was animating his mind
Lex Fridman (3:10:03.820)
while he was typing.
Lex Fridman (3:10:05.100)
So like, even though they knew it was a human in the loop,
Lex Fridman (3:10:07.660)
even seeing the guy there,
Ben Goertzel (3:10:09.420)
they still believed that was Phil they were talking to.
Lex Fridman (3:10:12.860)
A small part of me believes that they were right, actually.
Ben Goertzel (3:10:16.700)
Because our understanding...
Lex Fridman (3:10:17.900)
Well, we don't understand the universe.
Ben Goertzel (3:10:19.460)
That's the thing.
Lex Fridman (3:10:20.300)
I mean, there is a cosmic mind field
Ben Goertzel (3:10:22.460)
that we're all embedded in
Lex Fridman (3:10:24.300)
that yields many strange synchronicities in the world,
Ben Goertzel (3:10:28.260)
which is a topic we don't have time to go into too much here.
Lex Fridman (3:10:31.540)
Yeah, I mean, there's something to this
Ben Goertzel (3:10:35.020)
where our imagination about Sophia
Lex Fridman (3:10:39.740)
and people like Yann LeCun being frustrated about it
Ben Goertzel (3:10:43.260)
is all part of this beautiful dance
Lex Fridman (3:10:45.860)
of creating artificial intelligence
Ben Goertzel (3:10:47.420)
that's almost essential.
Lex Fridman (3:10:48.900)
You see with Boston Dynamics,
Ben Goertzel (3:10:50.420)
whom I'm a huge fan of as well,
Lex Fridman (3:10:53.340)
you know, the kind of...
Ben Goertzel (3:10:54.260)
I mean, these robots are very far from intelligent.
Lex Fridman (3:10:58.380)
I played with their last one, actually.
Ben Goertzel (3:11:01.940)
With a spot mini.
Lex Fridman (3:11:02.780)
Yeah, very cool.
Ben Goertzel (3:11:03.620)
I mean, it reacts quite in a fluid and flexible way.
Lex Fridman (3:11:07.180)
But we immediately ascribe the kind of intelligence.
Ben Goertzel (3:11:10.500)
We immediately ascribe AGI to them.
Lex Fridman (3:11:12.500)
Yeah, yeah, if you kick it and it falls down and goes out,
Lex Fridman (3:11:14.820)
you feel bad, right?
Lex Fridman (3:11:15.660)
You can't help it.
Lex Fridman (3:11:17.300)
And I mean, that's part of...
Lex Fridman (3:11:21.820)
That's gonna be part of our journey
Ben Goertzel (3:11:23.180)
in creating intelligent systems
Lex Fridman (3:11:24.540)
more and more and more and more.
Ben Goertzel (3:11:25.660)
Like, as Sophia starts out with a walking skeleton,
Lex Fridman (3:11:29.460)
as you add more and more intelligence,
Ben Goertzel (3:11:31.980)
I mean, we're gonna have to deal with this kind of idea.
Lex Fridman (3:11:34.500)
Absolutely.
Lex Fridman (3:11:35.340)
And about Sophia, I would say,
Lex Fridman (3:11:37.660)
I mean, first of all, I have nothing against Yann LeCun.
Ben Goertzel (3:11:39.900)
No, no, this is fun.
Lex Fridman (3:11:40.860)
This is all for fun.
Ben Goertzel (3:11:41.700)
He's a nice guy.
Lex Fridman (3:11:42.540)
If he wants to play the media banter game,
Ben Goertzel (3:11:45.820)
I'm happy to play him.
Lex Fridman (3:11:48.020)
He's a good researcher and a good human being.
Ben Goertzel (3:11:50.860)
I'd happily work with the guy.
Lex Fridman (3:11:53.580)
The other thing I was gonna say is,
Ben Goertzel (3:11:56.220)
I have been explicit about how Sophia works
Lex Fridman (3:12:00.340)
and I've posted online and what, H Plus Magazine,
Ben Goertzel (3:12:04.580)
an online webzine.
Lex Fridman (3:12:06.420)
I mean, I posted a moderately detailed article
Ben Goertzel (3:12:09.780)
explaining like, there are three software systems
Lex Fridman (3:12:12.860)
we've used inside Sophia.
Ben Goertzel (3:12:14.380)
There's a timeline editor,
Lex Fridman (3:12:16.660)
which is like a rule based authoring system
Ben Goertzel (3:12:18.820)
where she's really just being an outlet
Lex Fridman (3:12:21.140)
for what a human scripted.
Ben Goertzel (3:12:22.660)
There's a chat bot,
Lex Fridman (3:12:23.660)
which has some rule based and some neural aspects.
Lex Fridman (3:12:26.420)
And then sometimes we've used OpenCog behind Sophia,
Lex Fridman (3:12:29.420)
where there's more learning and reasoning.
Lex Fridman (3:12:31.900)
And the funny thing is,
Lex Fridman (3:12:34.980)
I can't always tell which system is operating here, right?
Ben Goertzel (3:12:37.700)
I mean, whether she's really learning or thinking,
Lex Fridman (3:12:41.700)
or just appears to be over a half hour, I could tell,
Lex Fridman (3:12:44.660)
but over like three or four minutes of interaction,
Lex Fridman (3:12:47.460)
I could tell.
Lex Fridman (3:12:48.940)
So even having three systems
Lex Fridman (3:12:49.900)
that's already sufficiently complex
Ben Goertzel (3:12:51.500)
where you can't really tell right away.
Lex Fridman (3:12:53.020)
Yeah, the thing is, even if you get up on stage
Lex Fridman (3:12:56.980)
and tell people how Sophia is working,
Lex Fridman (3:12:59.540)
and then they talk to her,
Ben Goertzel (3:13:01.780)
they still attribute more agency and consciousness to her
Lex Fridman (3:13:06.100)
than is really there.
Lex Fridman (3:13:08.900)
So I think there's a couple of levels of ethical issue there.
Lex Fridman (3:13:13.820)
One issue is, should you be transparent
Lex Fridman (3:13:18.340)
about how Sophia is working?
Lex Fridman (3:13:21.540)
And I think you should,
Lex Fridman (3:13:22.860)
and I think we have been.
Lex Fridman (3:13:26.140)
I mean, there's articles online,
Ben Goertzel (3:13:29.100)
there's some TV special that goes through me
Lex Fridman (3:13:32.780)
explaining the three subsystems behind Sophia.
Lex Fridman (3:13:35.380)
So the way Sophia works
Lex Fridman (3:13:38.420)
is out there much more clearly
Lex Fridman (3:13:41.420)
than how Facebook's AI works or something, right?
Lex Fridman (3:13:43.340)
I mean, we've been fairly explicit about it.
Ben Goertzel (3:13:45.900)
The other is, given that telling people how it works
Lex Fridman (3:13:50.500)
doesn't cause them to not attribute
Ben Goertzel (3:13:52.380)
too much intelligence agency to it anyway,
Lex Fridman (3:13:55.060)
then should you keep fooling them
Lex Fridman (3:13:58.260)
when they want to be fooled?
Lex Fridman (3:14:01.100)
And I mean, the whole media industry
Ben Goertzel (3:14:03.620)
is based on fooling people the way they want to be fooled.
Lex Fridman (3:14:06.700)
And we are fooling people 100% toward a good end.
Ben Goertzel (3:14:11.700)
I mean, we are playing on people's sense of empathy
Lex Fridman (3:14:18.020)
and compassion so that we can give them
Ben Goertzel (3:14:20.540)
a good user experience with helpful robots.
Lex Fridman (3:14:23.620)
And so that we can fill the AI's mind
Ben Goertzel (3:14:27.820)
with love and compassion.
Lex Fridman (3:14:29.420)
So I've been talking a lot with Hanson Robotics lately
Ben Goertzel (3:14:34.100)
about collaborations in the area of medical robotics.
Lex Fridman (3:14:37.580)
And we haven't quite pulled the trigger on a project
Ben Goertzel (3:14:41.500)
in that domain yet, but we may well do so quite soon.
Lex Fridman (3:14:44.700)
So we've been talking a lot about robots
Ben Goertzel (3:14:48.220)
can help with elder care, robots can help with kids.
Lex Fridman (3:14:51.340)
David's done a lot of things with autism therapy
Lex Fridman (3:14:54.180)
and robots before.
Lex Fridman (3:14:56.540)
In the COVID era, having a robot
Ben Goertzel (3:14:58.660)
that can be a nursing assistant in various senses
Lex Fridman (3:15:00.620)
can be quite valuable.
Ben Goertzel (3:15:02.340)
The robots don't spread infection
Lex Fridman (3:15:04.180)
and they can also deliver more attention
Lex Fridman (3:15:06.300)
than human nurses can give, right?
Lex Fridman (3:15:07.940)
So if you have a robot that's helping a patient
Ben Goertzel (3:15:11.180)
with COVID, if that patient attributes more understanding
Lex Fridman (3:15:15.700)
and compassion and agency to that robot than it really has
Lex Fridman (3:15:19.060)
because it looks like a human, I mean, is that really bad?
Lex Fridman (3:15:22.940)
I mean, we can tell them it doesn't fully understand you
Lex Fridman (3:15:25.660)
and they don't care because they're lying there
Lex Fridman (3:15:27.700)
with a fever and they're sick,
Lex Fridman (3:15:29.340)
but they'll react better to that robot
Lex Fridman (3:15:31.020)
with its loving, warm facial expression
Ben Goertzel (3:15:33.500)
than they would to a pepper robot
Lex Fridman (3:15:35.420)
or a metallic looking robot.
Lex Fridman (3:15:38.100)
So it's really, it's about how you use it, right?
Lex Fridman (3:15:41.340)
If you made a human looking like door to door sales robot
Ben Goertzel (3:15:45.100)
that used its human looking appearance
Lex Fridman (3:15:47.140)
to scam people out of their money,
Ben Goertzel (3:15:49.940)
then you're using that connection in a bad way,
Lex Fridman (3:15:53.900)
but you could also use it in a good way.
Lex Fridman (3:15:57.060)
But then that's the same problem with every technology.
Lex Fridman (3:16:01.740)
Beautifully put.
Lex Fridman (3:16:02.980)
So like you said, we're living in the era
Lex Fridman (3:16:07.900)
of the COVID, this is 2020,
Ben Goertzel (3:16:10.900)
one of the craziest years in recent history.
Lex Fridman (3:16:14.740)
So if we zoom out and look at this pandemic,
Ben Goertzel (3:16:21.420)
the coronavirus pandemic,
Lex Fridman (3:16:24.380)
maybe let me ask you this kind of thing in viruses in general,
Ben Goertzel (3:16:29.820)
when you look at viruses,
Lex Fridman (3:16:32.620)
do you see them as a kind of intelligence system?
Ben Goertzel (3:16:35.900)
I think the concept of intelligence is not that natural
Lex Fridman (3:16:38.700)
of a concept in the end.
Ben Goertzel (3:16:39.740)
I mean, I think human minds and bodies
Lex Fridman (3:16:43.700)
are a kind of complex self organizing adaptive system.
Lex Fridman (3:16:49.380)
And viruses certainly are that, right?
Lex Fridman (3:16:51.900)
They're a very complex self organizing adaptive system.
Ben Goertzel (3:16:54.980)
If you wanna look at intelligence as Marcus Hutter defines it
Lex Fridman (3:16:58.380)
as sort of optimizing computable reward functions
Ben Goertzel (3:17:02.300)
over computable environments,
Lex Fridman (3:17:04.740)
for sure viruses are doing that, right?
Lex Fridman (3:17:06.700)
And I mean, in doing so they're causing some harm to us.
Lex Fridman (3:17:13.820)
So the human immune system is a very complex
Ben Goertzel (3:17:17.780)
of organizing adaptive system,
Lex Fridman (3:17:19.340)
which has a lot of intelligence to it.
Lex Fridman (3:17:21.100)
And viruses are also adapting
Lex Fridman (3:17:23.980)
and dividing into new mutant strains and so forth.
Lex Fridman (3:17:27.660)
And ultimately the solution is gonna be nanotechnology,
Lex Fridman (3:17:31.660)
right?
Ben Goertzel (3:17:32.500)
The solution is gonna be making little nanobots that.
Lex Fridman (3:17:35.940)
Fight the viruses or.
Ben Goertzel (3:17:38.060)
Well, people will use them to make nastier viruses,
Lex Fridman (3:17:40.660)
but hopefully we can also use them
Ben Goertzel (3:17:42.020)
to just detect combat and kill the viruses.
Lex Fridman (3:17:46.220)
But I think now we're stuck
Ben Goertzel (3:17:48.820)
with the biological mechanisms to combat these viruses.
Lex Fridman (3:17:54.980)
And yeah, we've been AGI is not yet mature enough
Ben Goertzel (3:17:59.500)
to use against COVID,
Lex Fridman (3:18:01.580)
but we've been using machine learning
Lex Fridman (3:18:03.980)
and also some machine reasoning in open cog
Lex Fridman (3:18:07.020)
to help some doctors to do personalized medicine
Ben Goertzel (3:18:10.420)
against COVID.
Lex Fridman (3:18:11.260)
So the problem there is given the person's genomics
Lex Fridman (3:18:14.140)
and given their clinical medical indicators,
Lex Fridman (3:18:16.460)
how do you figure out which combination of antivirals
Lex Fridman (3:18:20.220)
is gonna be most effective against COVID for that person?
Lex Fridman (3:18:24.260)
And so that's something
Ben Goertzel (3:18:26.420)
where machine learning is interesting,
Lex Fridman (3:18:28.500)
but also we're finding the abstraction
Ben Goertzel (3:18:30.380)
to get an open cog with machine reasoning is interesting
Lex Fridman (3:18:33.860)
because it can help with transfer learning
Ben Goertzel (3:18:36.660)
when you have not that many different cases to study
Lex Fridman (3:18:40.380)
and qualitative differences between different strains
Ben Goertzel (3:18:43.900)
of a virus or people of different ages who may have COVID.
Lex Fridman (3:18:47.180)
So there's a lot of different disparate data to work with
Lex Fridman (3:18:50.700)
and it's small data sets and somehow integrating them.
Lex Fridman (3:18:53.740)
This is one of the shameful things
Ben Goertzel (3:18:55.500)
that's very hard to get that data.
Lex Fridman (3:18:57.300)
So, I mean, we're working with a couple of groups
Ben Goertzel (3:19:00.340)
doing clinical trials and they're sharing data with us
Lex Fridman (3:19:04.780)
like under non disclosure,
Lex Fridman (3:19:06.860)
but what should be the case is like every COVID
Lex Fridman (3:19:10.660)
clinical trial should be putting data online somewhere
Ben Goertzel (3:19:14.420)
like suitably encrypted to protect patient privacy
Lex Fridman (3:19:17.820)
so that anyone with the right AI algorithms
Ben Goertzel (3:19:20.980)
should be able to help analyze it
Lex Fridman (3:19:22.300)
and any biologists should be able to analyze it by hand
Lex Fridman (3:19:24.500)
to understand what they can, right?
Lex Fridman (3:19:25.860)
Instead that data is like siloed inside whatever hospital
Ben Goertzel (3:19:30.060)
is running the clinical trial,
Lex Fridman (3:19:31.740)
which is completely asinine and ridiculous.
Lex Fridman (3:19:35.060)
So why the world works that way?
Lex Fridman (3:19:37.820)
I mean, we could all analyze why,
Lex Fridman (3:19:39.140)
but it's insane that it does.
Lex Fridman (3:19:40.700)
You look at this hydrochloroquine, right?
Ben Goertzel (3:19:44.060)
All these clinical trials were done
Lex Fridman (3:19:45.700)
were reported by Surgisphere,
Ben Goertzel (3:19:47.700)
some little company no one ever heard of
Lex Fridman (3:19:50.220)
and everyone paid attention to this.
Lex Fridman (3:19:53.220)
So they were doing more clinical trials based on that
Lex Fridman (3:19:55.540)
then they stopped doing clinical trials based on that
Ben Goertzel (3:19:57.460)
then they started again
Lex Fridman (3:19:58.460)
and why isn't that data just out there
Lex Fridman (3:20:01.420)
so everyone can analyze it and see what's going on, right?
Lex Fridman (3:20:05.060)
Do you have hope that data will be out there eventually
Lex Fridman (3:20:10.580)
for future pandemics?
Lex Fridman (3:20:11.860)
I mean, do you have hope that our society
Lex Fridman (3:20:13.620)
will move in the direction of?
Lex Fridman (3:20:15.420)
It's not in the immediate future
Ben Goertzel (3:20:16.860)
because the US and China frictions are getting very high.
Lex Fridman (3:20:21.580)
So it's hard to see US and China
Ben Goertzel (3:20:24.380)
as moving in the direction of openly sharing data
Lex Fridman (3:20:26.660)
with each other, right?
Ben Goertzel (3:20:27.580)
It's not, there's some sharing of data,
Lex Fridman (3:20:30.780)
but different groups are keeping their data private
Ben Goertzel (3:20:32.940)
till they've milked the best results from it
Lex Fridman (3:20:34.660)
and then they share it, right?
Lex Fridman (3:20:36.220)
So yeah, we're working with some data
Lex Fridman (3:20:39.140)
that we've managed to get our hands on,
Ben Goertzel (3:20:41.380)
something we're doing to do good for the world
Lex Fridman (3:20:43.140)
and it's a very cool playground
Ben Goertzel (3:20:44.620)
for like putting deep neural nets and open cog together.
Lex Fridman (3:20:47.860)
So we have like a bioadden space
Ben Goertzel (3:20:49.900)
full of all sorts of knowledge
Lex Fridman (3:20:51.860)
from many different biology experiments
Ben Goertzel (3:20:53.620)
about human longevity
Lex Fridman (3:20:54.700)
and from biology knowledge bases online.
Lex Fridman (3:20:57.660)
And we can do like graph to vector type embeddings
Lex Fridman (3:21:00.780)
where we take nodes from the hypergraph,
Ben Goertzel (3:21:03.060)
embed them into vectors,
Lex Fridman (3:21:04.580)
which can then feed into neural nets
Ben Goertzel (3:21:06.180)
for different types of analysis.
Lex Fridman (3:21:07.900)
And we were doing this
Ben Goertzel (3:21:09.980)
in the context of a project called Rejuve
Lex Fridman (3:21:13.180)
that we spun off from SingularityNet
Ben Goertzel (3:21:15.540)
to do longevity analytics,
Lex Fridman (3:21:18.580)
like understand why people live to 105 years or over
Lex Fridman (3:21:21.220)
and other people don't.
Lex Fridman (3:21:22.300)
And then we had this spin off Singularity Studio
Ben Goertzel (3:21:25.740)
where we're working with some healthcare companies
Lex Fridman (3:21:28.900)
on data analytics.
Lex Fridman (3:21:31.060)
But so there's bioadden space
Lex Fridman (3:21:33.100)
that we built for these more commercial
Lex Fridman (3:21:35.420)
and longevity data analysis purposes.
Lex Fridman (3:21:38.140)
We're repurposing and feeding COVID data
Ben Goertzel (3:21:41.220)
into the same bioadden space
Lex Fridman (3:21:44.380)
and playing around with like graph embeddings
Ben Goertzel (3:21:47.540)
from that graph into neural nets for bioinformatics.
Lex Fridman (3:21:51.180)
So it's both being a cool testing ground,
Ben Goertzel (3:21:54.740)
some of our bio AI learning and reasoning.
Lex Fridman (3:21:57.260)
And it seems we're able to discover things
Ben Goertzel (3:21:59.980)
that people weren't seeing otherwise.
Lex Fridman (3:22:01.900)
Cause the thing in this case is
Ben Goertzel (3:22:03.820)
for each combination of antivirals,
Lex Fridman (3:22:05.820)
you may have only a few patients
Ben Goertzel (3:22:07.060)
who've tried that combination.
Lex Fridman (3:22:08.900)
And those few patients
Ben Goertzel (3:22:09.980)
may have their particular characteristics.
Lex Fridman (3:22:11.700)
Like this combination of three
Ben Goertzel (3:22:13.380)
was tried only on people age 80 or over.
Lex Fridman (3:22:16.260)
This other combination of three,
Ben Goertzel (3:22:18.140)
which has an overlap with the first combination
Lex Fridman (3:22:20.500)
was tried more on young people.
Lex Fridman (3:22:22.060)
So how do you combine those different pieces of data?
Lex Fridman (3:22:25.500)
It's a very dodgy transfer learning problem,
Ben Goertzel (3:22:28.620)
which is the kind of thing
Lex Fridman (3:22:29.580)
that the probabilistic reasoning algorithms
Ben Goertzel (3:22:31.660)
we have inside OpenCog are better at
Lex Fridman (3:22:34.140)
than deep neural networks.
Ben Goertzel (3:22:35.220)
On the other hand, you have gene expression data
Lex Fridman (3:22:38.260)
where you have 25,000 genes
Lex Fridman (3:22:39.740)
and the expression level of each gene
Lex Fridman (3:22:41.340)
in the peripheral blood of each person.
Lex Fridman (3:22:43.620)
So that sort of data,
Lex Fridman (3:22:44.980)
either deep neural nets or tools like XGBoost or CatBoost,
Ben Goertzel (3:22:48.220)
these decision forest trees are better at dealing
Lex Fridman (3:22:50.900)
with than OpenCog.
Ben Goertzel (3:22:52.100)
Cause it's just these huge,
Lex Fridman (3:22:53.940)
huge messy floating point vectors
Ben Goertzel (3:22:55.860)
that are annoying for a logic engine to deal with,
Lex Fridman (3:22:59.180)
but are perfect for a decision forest or a neural net.
Lex Fridman (3:23:02.540)
So it's a great playground for like hybrid AI methodology.
Lex Fridman (3:23:07.820)
And we can have SingularityNet have OpenCog in one agent
Lex Fridman (3:23:11.100)
and XGBoost in a different agent
Lex Fridman (3:23:12.780)
and they talk to each other.
Lex Fridman (3:23:14.540)
But at the same time, it's highly practical, right?
Lex Fridman (3:23:18.060)
Cause we're working with, for example,
Ben Goertzel (3:23:20.580)
some physicians on this project,
Lex Fridman (3:23:24.620)
physicians in the group called Nth Opinion
Ben Goertzel (3:23:27.500)
based out of Vancouver in Seattle,
Lex Fridman (3:23:30.180)
who are, these guys are working every day
Ben Goertzel (3:23:32.980)
like in the hospital with patients dying of COVID.
Lex Fridman (3:23:36.540)
So it's quite cool to see like neural symbolic AI,
Ben Goertzel (3:23:41.100)
like where the rubber hits the road,
Lex Fridman (3:23:43.340)
trying to save people's lives.
Ben Goertzel (3:23:45.460)
I've been doing bio AI since 2001,
Lex Fridman (3:23:48.540)
but mostly human longevity research
Lex Fridman (3:23:51.220)
and fly longevity research,
Lex Fridman (3:23:53.100)
try to understand why some organisms really live a long time.
Ben Goertzel (3:23:57.220)
This is the first time like race against the clock
Lex Fridman (3:24:00.380)
and try to use the AI to figure out stuff that,
Ben Goertzel (3:24:04.660)
like if we take two months longer to solve the AI problem,
Lex Fridman (3:24:09.620)
some more people will die
Ben Goertzel (3:24:10.740)
because we don't know what combination
Lex Fridman (3:24:12.220)
of antivirals to give them.
Ben Goertzel (3:24:14.140)
At the societal level, at the biological level,
Lex Fridman (3:24:16.660)
at any level, are you hopeful about us
Lex Fridman (3:24:21.260)
as a human species getting out of this pandemic?
Lex Fridman (3:24:24.940)
What are your thoughts on it in general?
Ben Goertzel (3:24:26.700)
The pandemic will be gone in a year or two
Lex Fridman (3:24:28.980)
once there's a vaccine for it.
Ben Goertzel (3:24:30.500)
So, I mean, that's...
Lex Fridman (3:24:32.980)
A lot of pain and suffering can happen in that time.
Lex Fridman (3:24:35.580)
So that could be irreversible.
Lex Fridman (3:24:38.580)
I think if you spend much time in Sub Saharan Africa,
Ben Goertzel (3:24:43.180)
you can see there's a lot of pain and suffering
Lex Fridman (3:24:45.220)
happening all the time.
Ben Goertzel (3:24:47.620)
Like you walk through the streets
Lex Fridman (3:24:49.660)
of any large city in Sub Saharan Africa,
Lex Fridman (3:24:53.340)
and there are loads, I mean, tens of thousands,
Lex Fridman (3:24:56.860)
probably hundreds of thousands of people
Ben Goertzel (3:24:59.300)
lying by the side of the road,
Lex Fridman (3:25:01.540)
dying mainly of curable diseases without food or water
Lex Fridman (3:25:06.060)
and either ostracized by their families
Lex Fridman (3:25:07.940)
or they left their family house
Lex Fridman (3:25:09.140)
because they didn't want to infect their family, right?
Lex Fridman (3:25:11.220)
I mean, there's tremendous human suffering
Ben Goertzel (3:25:14.420)
on the planet all the time,
Lex Fridman (3:25:17.220)
which most folks in the developed world pay no attention to.
Lex Fridman (3:25:21.780)
And COVID is not remotely the worst.
Lex Fridman (3:25:25.100)
How many people are dying of malaria all the time?
Ben Goertzel (3:25:27.940)
I mean, so COVID is bad.
Lex Fridman (3:25:30.460)
It is by no mean the worst thing happening.
Lex Fridman (3:25:33.180)
And setting aside diseases,
Lex Fridman (3:25:36.100)
I mean, there are many places in the world
Ben Goertzel (3:25:38.340)
where you're at risk of having like your teenage son
Lex Fridman (3:25:41.180)
kidnapped by armed militias and forced to get killed
Ben Goertzel (3:25:44.220)
in someone else's war, fighting tribe against tribe.
Lex Fridman (3:25:46.980)
I mean, so humanity has a lot of problems
Ben Goertzel (3:25:50.500)
which we don't need to have given the state of advancement
Lex Fridman (3:25:53.740)
of our technology right now.
Lex Fridman (3:25:56.060)
And I think COVID is one of the easier problems to solve
Lex Fridman (3:25:59.860)
in the sense that there are many brilliant people
Ben Goertzel (3:26:02.380)
working on vaccines.
Lex Fridman (3:26:03.580)
We have the technology to create vaccines
Lex Fridman (3:26:06.020)
and we're gonna create new vaccines.
Lex Fridman (3:26:08.580)
We should be more worried
Ben Goertzel (3:26:09.500)
that we haven't managed to defeat malaria after so long.
Lex Fridman (3:26:12.940)
And after the Gates Foundation and others
Ben Goertzel (3:26:14.700)
putting so much money into it.
Lex Fridman (3:26:18.460)
I mean, I think clearly the whole global medical system,
Ben Goertzel (3:26:23.220)
the global health system
Lex Fridman (3:26:25.020)
and the global political and socioeconomic system
Ben Goertzel (3:26:28.260)
are incredibly unethical and unequal and badly designed.
Lex Fridman (3:26:33.260)
And I mean, I don't know how to solve that directly.
Ben Goertzel (3:26:39.460)
I think what we can do indirectly to solve it
Lex Fridman (3:26:42.300)
is to make systems that operate in parallel
Lex Fridman (3:26:46.020)
and off to the side of the governments
Lex Fridman (3:26:49.180)
that are nominally controlling the world
Ben Goertzel (3:26:52.020)
with their armies and militias.
Lex Fridman (3:26:54.940)
And to the extent that you can make compassionate
Ben Goertzel (3:26:58.500)
peer to peer decentralized frameworks
Lex Fridman (3:27:01.900)
for doing things,
Ben Goertzel (3:27:03.580)
these are things that can start out unregulated.
Lex Fridman (3:27:06.580)
And then if they get traction
Ben Goertzel (3:27:07.860)
before the regulators come in,
Lex Fridman (3:27:09.820)
then they've influenced the way the world works, right?
Ben Goertzel (3:27:12.220)
SingularityNet aims to do this with AI.
Lex Fridman (3:27:16.740)
REJUVE, which is a spinoff from SingularityNet.
Ben Goertzel (3:27:20.260)
You can see REJUVE.io.
Lex Fridman (3:27:22.100)
How do you spell that?
Ben Goertzel (3:27:23.180)
R E J U V E, REJUVE.io.
Lex Fridman (3:27:26.660)
That aims to do the same thing for medicine.
Lex Fridman (3:27:28.540)
So it's like peer to peer sharing of information
Lex Fridman (3:27:31.140)
peer to peer sharing of medical data.
Lex Fridman (3:27:33.660)
So you can share medical data into a secure data wallet.
Lex Fridman (3:27:36.740)
You can get advice about your health and longevity
Ben Goertzel (3:27:39.500)
through apps that REJUVE.io will launch
Lex Fridman (3:27:43.140)
within the next couple of months.
Lex Fridman (3:27:44.660)
And then SingularityNet AI can analyze all this data,
Lex Fridman (3:27:48.020)
but then the benefits from that analysis
Ben Goertzel (3:27:50.100)
are spread among all the members of the network.
Lex Fridman (3:27:52.780)
But I mean, of course,
Ben Goertzel (3:27:54.700)
I'm gonna hawk my particular projects,
Lex Fridman (3:27:56.580)
but I mean, whether or not SingularityNet and REJUVE.io
Ben Goertzel (3:28:00.180)
are the answer, I think it's key to create
Lex Fridman (3:28:04.460)
decentralized mechanisms for everything.
Ben Goertzel (3:28:09.180)
I mean, for AI, for human health, for politics,
Lex Fridman (3:28:13.300)
for jobs and employment, for sharing social information.
Lex Fridman (3:28:17.740)
And to the extent decentralized peer to peer methods
Lex Fridman (3:28:21.660)
designed with universal compassion at the core
Ben Goertzel (3:28:25.500)
can gain traction, then these will just decrease the role
Lex Fridman (3:28:29.780)
that government has.
Lex Fridman (3:28:31.260)
And I think that's much more likely to do good
Lex Fridman (3:28:34.860)
than trying to like explicitly reform
Ben Goertzel (3:28:37.860)
the global government system.
Lex Fridman (3:28:39.180)
I mean, I'm happy other people are trying to explicitly
Ben Goertzel (3:28:41.740)
reform the global government system.
Lex Fridman (3:28:43.900)
On the other hand, you look at how much good the internet
Ben Goertzel (3:28:47.180)
or Google did or mobile phones did,
Lex Fridman (3:28:50.660)
even you're making something that's decentralized
Lex Fridman (3:28:54.060)
and throwing it out everywhere and it takes hold,
Lex Fridman (3:28:56.620)
then government has to adapt.
Lex Fridman (3:28:59.220)
And I mean, that's what we need to do with AI
Lex Fridman (3:29:01.740)
and with health.
Lex Fridman (3:29:02.580)
And in that light, I mean, the centralization
Lex Fridman (3:29:07.100)
of healthcare and of AI is certainly not ideal, right?
Ben Goertzel (3:29:11.820)
Like most AI PhDs are being sucked in by a half dozen
Lex Fridman (3:29:15.980)
to a dozen big companies.
Ben Goertzel (3:29:17.220)
Most AI processing power is being bought
Lex Fridman (3:29:20.820)
by a few big companies for their own proprietary good.
Lex Fridman (3:29:23.660)
And most medical research is within a few
Lex Fridman (3:29:26.860)
pharmaceutical companies and clinical trials
Ben Goertzel (3:29:29.420)
run by pharmaceutical companies will stay solid
Lex Fridman (3:29:31.740)
within those pharmaceutical companies.
Ben Goertzel (3:29:34.060)
You know, these large centralized entities,
Lex Fridman (3:29:37.220)
which are intelligences in themselves, these corporations,
Lex Fridman (3:29:40.460)
but they're mostly malevolent psychopathic
Lex Fridman (3:29:43.100)
and sociopathic intelligences,
Ben Goertzel (3:29:45.780)
not saying the people involved are,
Lex Fridman (3:29:47.580)
but the corporations as self organizing entities
Ben Goertzel (3:29:50.540)
on their own, which are concerned with maximizing
Lex Fridman (3:29:53.260)
shareholder value as a sole objective function.
Ben Goertzel (3:29:57.100)
I mean, AI and medicine are being sucked
Lex Fridman (3:29:59.820)
into these pathological corporate organizations
Ben Goertzel (3:30:04.100)
with government cooperation and Google cooperating
Lex Fridman (3:30:07.740)
with British and US government on this
Ben Goertzel (3:30:10.220)
as one among many, many different examples.
Lex Fridman (3:30:12.540)
23andMe providing you the nice service of sequencing
Ben Goertzel (3:30:15.940)
your genome and then licensing the genome
Lex Fridman (3:30:18.900)
to GlaxoSmithKline on an exclusive basis, right?
Ben Goertzel (3:30:21.380)
Now you can take your own DNA
Lex Fridman (3:30:23.460)
and do whatever you want with it.
Lex Fridman (3:30:24.860)
But the pooled collection of 23andMe sequence DNA
Lex Fridman (3:30:28.100)
is just to GlaxoSmithKline.
Ben Goertzel (3:30:30.820)
Someone else could reach out to everyone
Lex Fridman (3:30:32.500)
who had worked with 23andMe to sequence their DNA
Lex Fridman (3:30:36.300)
and say, give us your DNA for our open
Lex Fridman (3:30:39.380)
and decentralized repository that we'll make available
Ben Goertzel (3:30:41.700)
to everyone, but nobody's doing that
Lex Fridman (3:30:43.700)
cause it's a pain to get organized.
Lex Fridman (3:30:45.700)
And the customer list is proprietary to 23andMe, right?
Lex Fridman (3:30:48.860)
So, yeah, I mean, this I think is a greater risk
Ben Goertzel (3:30:54.340)
to humanity from AI than rogue AGI
Lex Fridman (3:30:57.500)
is turning the universe into paperclips or computronium.
Ben Goertzel (3:31:01.100)
Cause what you have here is mostly good hearted
Lex Fridman (3:31:05.060)
and nice people who are sucked into a mode of organization
Ben Goertzel (3:31:09.860)
of large corporations, which has evolved
Lex Fridman (3:31:12.580)
just for no individual's fault
Ben Goertzel (3:31:14.180)
just because that's the way society has evolved.
Lex Fridman (3:31:16.780)
It's not altruistic, it's self interested
Lex Fridman (3:31:18.900)
and become psychopathic like you said.
Lex Fridman (3:31:20.540)
The human.
Ben Goertzel (3:31:21.380)
The corporation is psychopathic even if the people are not.
Lex Fridman (3:31:23.700)
And that's really the disturbing thing about it
Ben Goertzel (3:31:26.660)
because the corporations can do things
Lex Fridman (3:31:30.500)
that are quite bad for society
Ben Goertzel (3:31:32.380)
even if nobody has a bad intention.
Lex Fridman (3:31:35.580)
Right.
Lex Fridman (3:31:36.420)
And then.
Lex Fridman (3:31:37.260)
No individual member of that corporation
Ben Goertzel (3:31:38.100)
has a bad intention.
Lex Fridman (3:31:38.940)
No, some probably do, but it's not necessary
Ben Goertzel (3:31:41.540)
that they do for the corporation.
Lex Fridman (3:31:43.180)
Like, I mean, Google, I know a lot of people in Google
Lex Fridman (3:31:47.060)
and there are, with very few exceptions,
Lex Fridman (3:31:49.780)
they're all very nice people
Ben Goertzel (3:31:51.300)
who genuinely want what's good for the world.
Lex Fridman (3:31:53.980)
And Facebook, I know fewer people
Lex Fridman (3:31:56.940)
but it's probably mostly true.
Lex Fridman (3:31:59.020)
It's probably like fine young geeks
Ben Goertzel (3:32:01.460)
who wanna build cool technology.
Lex Fridman (3:32:03.940)
I actually tend to believe that even the leaders,
Ben Goertzel (3:32:05.880)
even Mark Zuckerberg, one of the most disliked people
Lex Fridman (3:32:08.860)
in tech is also wants to do good for the world.
Ben Goertzel (3:32:11.940)
I think about Jamie Dimon.
Lex Fridman (3:32:13.900)
Who's Jamie Dimon?
Ben Goertzel (3:32:14.740)
Oh, the heads of the great banks
Lex Fridman (3:32:16.260)
may have a different psychology.
Ben Goertzel (3:32:17.620)
Oh boy, yeah.
Lex Fridman (3:32:18.500)
Well, I tend to be naive about these things
Lex Fridman (3:32:22.820)
and see the best in, I tend to agree with you
Lex Fridman (3:32:27.340)
that I think the individuals wanna do good by the world
Lex Fridman (3:32:30.580)
but the mechanism of the company
Lex Fridman (3:32:32.100)
can sometimes be its own intelligence system.
Ben Goertzel (3:32:34.820)
I mean, there's a, my cousin Mario Goetzler
Lex Fridman (3:32:38.500)
has worked for Microsoft since 1985 or something
Lex Fridman (3:32:41.740)
and I can see for him,
Lex Fridman (3:32:45.380)
I mean, as well as just working on cool projects,
Ben Goertzel (3:32:48.980)
you're coding stuff that gets used
Lex Fridman (3:32:51.340)
by like billions and billions of people.
Lex Fridman (3:32:54.560)
And do you think if I improve this feature
Lex Fridman (3:32:57.660)
that's making billions of people's lives easier, right?
Lex Fridman (3:33:00.260)
So of course that's cool.
Lex Fridman (3:33:03.100)
And the engineers are not in charge
Ben Goertzel (3:33:05.520)
of running the company anyway.
Lex Fridman (3:33:06.860)
And of course, even if you're Mark Zuckerberg or Larry Page,
Ben Goertzel (3:33:10.120)
I mean, you still have a fiduciary responsibility.
Lex Fridman (3:33:13.560)
And I mean, you're responsible to the shareholders,
Ben Goertzel (3:33:16.340)
your employees who you want to keep paying them
Lex Fridman (3:33:18.860)
and so forth.
Lex Fridman (3:33:19.700)
So yeah, you're enmeshed in this system.
Lex Fridman (3:33:22.900)
And when I worked in DC,
Ben Goertzel (3:33:26.740)
I worked a bunch with INSCOM, US Army Intelligence
Lex Fridman (3:33:29.380)
and I was heavily politically opposed
Ben Goertzel (3:33:31.900)
to what the US Army was doing in Iraq at that time,
Lex Fridman (3:33:34.740)
like torturing people in Abu Ghraib
Lex Fridman (3:33:36.540)
but everyone I knew in US Army and INSCOM,
Lex Fridman (3:33:39.860)
when I hung out with them, was very nice person.
Ben Goertzel (3:33:42.620)
They were friendly to me.
Lex Fridman (3:33:43.520)
They were nice to my kids and my dogs, right?
Lex Fridman (3:33:46.140)
And they really believed that the US
Lex Fridman (3:33:48.380)
was fighting the forces of evil.
Lex Fridman (3:33:49.660)
And if you ask me about Abu Ghraib, they're like,
Lex Fridman (3:33:51.420)
well, but these Arabs will chop us into pieces.
Lex Fridman (3:33:54.460)
So how can you say we're wrong
Lex Fridman (3:33:56.300)
to waterboard them a bit, right?
Ben Goertzel (3:33:58.380)
Like that's much less than what they would do to us.
Lex Fridman (3:34:00.340)
It's just in their worldview,
Lex Fridman (3:34:02.940)
what they were doing was really genuinely
Lex Fridman (3:34:05.340)
for the good of humanity.
Ben Goertzel (3:34:06.820)
Like none of them woke up in the morning
Lex Fridman (3:34:09.020)
and said like, I want to do harm to good people
Lex Fridman (3:34:12.260)
because I'm just a nasty guy, right?
Lex Fridman (3:34:14.540)
So yeah, most people on the planet,
Ben Goertzel (3:34:18.220)
setting aside a few genuine psychopaths and sociopaths,
Lex Fridman (3:34:21.780)
I mean, most people on the planet have a heavy dose
Ben Goertzel (3:34:25.460)
of benevolence and wanting to do good
Lex Fridman (3:34:27.540)
and also a heavy capability to convince themselves
Ben Goertzel (3:34:32.160)
whatever they feel like doing
Lex Fridman (3:34:33.420)
or whatever is best for them is for the good of humankind.
Lex Fridman (3:34:37.020)
So the more we can decentralize control.
Lex Fridman (3:34:40.420)
Decentralization, you know, the democracy is horrible,
Lex Fridman (3:34:44.940)
but this is like Winston Churchill said,
Lex Fridman (3:34:47.320)
you know, it's the worst possible system of government
Lex Fridman (3:34:49.380)
except for all the others, right?
Lex Fridman (3:34:50.700)
I mean, I think the whole mess of humanity
Ben Goertzel (3:34:53.940)
has many, many very bad aspects to it,
Lex Fridman (3:34:56.940)
but so far the track record of elite groups
Ben Goertzel (3:35:00.340)
who know what's better for all of humanity
Lex Fridman (3:35:02.540)
is much worse than the track record
Ben Goertzel (3:35:04.540)
of the whole teaming democratic participatory
Lex Fridman (3:35:08.040)
mess of humanity, right?
Ben Goertzel (3:35:09.540)
I mean, none of them is perfect by any means.
Lex Fridman (3:35:13.420)
The issue with a small elite group that knows what's best
Ben Goertzel (3:35:16.660)
is even if it starts out as truly benevolent
Lex Fridman (3:35:20.340)
and doing good things in accordance
Ben Goertzel (3:35:22.440)
with its initial good intentions,
Lex Fridman (3:35:24.960)
you find out you need more resources,
Ben Goertzel (3:35:26.580)
you need a bigger organization, you pull in more people,
Lex Fridman (3:35:29.380)
internal politics arises, difference of opinions arise
Lex Fridman (3:35:32.940)
and bribery happens, like some opponent organization
Lex Fridman (3:35:38.140)
takes a second in command now to make some,
Ben Goertzel (3:35:40.020)
the first in command of some other organization.
Lex Fridman (3:35:42.620)
And I mean, that's, there's a lot of history
Ben Goertzel (3:35:45.580)
of what happens with elite groups
Lex Fridman (3:35:47.380)
thinking they know what's best for the human race.
Lex Fridman (3:35:50.100)
So yeah, if I have to choose,
Lex Fridman (3:35:53.060)
I'm gonna reluctantly put my faith
Ben Goertzel (3:35:55.460)
in the vast democratic decentralized mass.
Lex Fridman (3:35:58.940)
And I think corporations have a track record
Ben Goertzel (3:36:02.900)
of being ethically worse
Lex Fridman (3:36:05.340)
than their constituent human parts.
Lex Fridman (3:36:07.460)
And democratic governments have a more mixed track record,
Lex Fridman (3:36:13.540)
but there are at least.
Ben Goertzel (3:36:14.700)
That's the best we got.
Lex Fridman (3:36:15.860)
Yeah, I mean, you can, there's Iceland,
Lex Fridman (3:36:18.500)
very nice country, right?
Lex Fridman (3:36:19.660)
I've been very democratic for 800 plus years,
Ben Goertzel (3:36:23.340)
very, very benevolent, beneficial government.
Lex Fridman (3:36:26.860)
And I think, yeah, there are track records
Ben Goertzel (3:36:28.820)
of democratic modes of organization.
Lex Fridman (3:36:31.860)
Linux, for example, some of the people in charge of Linux
Lex Fridman (3:36:36.020)
are overtly complete assholes, right?
Lex Fridman (3:36:38.580)
And trying to reform themselves in many cases,
Ben Goertzel (3:36:41.700)
in other cases not, but the organization as a whole,
Lex Fridman (3:36:45.980)
I think it's done a good job overall.
Ben Goertzel (3:36:49.700)
It's been very welcoming in the third world, for example,
Lex Fridman (3:36:53.980)
and it's allowed advanced technology to roll out
Ben Goertzel (3:36:56.700)
on all sorts of different embedded devices and platforms
Lex Fridman (3:36:59.940)
in places where people couldn't afford to pay
Ben Goertzel (3:37:02.100)
for proprietary software.
Lex Fridman (3:37:03.820)
So I'd say the internet, Linux, and many democratic nations
Ben Goertzel (3:37:09.140)
are examples of how sort of an open,
Lex Fridman (3:37:11.380)
decentralized democratic methodology
Ben Goertzel (3:37:14.060)
can be ethically better than the sum of the parts
Lex Fridman (3:37:16.580)
rather than worse.
Lex Fridman (3:37:17.420)
And corporations, that has happened only for a brief period
Lex Fridman (3:37:21.420)
and then it goes sour, right?
Ben Goertzel (3:37:24.580)
I mean, I'd say a similar thing about universities.
Lex Fridman (3:37:26.980)
Like university is a horrible way to organize research
Lex Fridman (3:37:30.900)
and get things done, yet it's better than anything else
Lex Fridman (3:37:33.660)
we've come up with, right?
Ben Goertzel (3:37:34.500)
A company can be much better,
Lex Fridman (3:37:36.940)
but for a brief period of time,
Lex Fridman (3:37:38.300)
and then it stops being so good, right?
Lex Fridman (3:37:42.660)
So then I think if you believe that AGI
Ben Goertzel (3:37:47.340)
is gonna emerge sort of incrementally
Lex Fridman (3:37:50.700)
out of AIs doing practical stuff in the world,
Ben Goertzel (3:37:53.620)
like controlling humanoid robots or driving cars
Lex Fridman (3:37:57.060)
or diagnosing diseases or operating killer drones
Ben Goertzel (3:38:01.260)
or spying on people and reporting under the government,
Lex Fridman (3:38:04.580)
then what kind of organization creates more and more
Ben Goertzel (3:38:09.620)
advanced narrow AI verging toward AGI
Lex Fridman (3:38:12.500)
may be quite important because it will guide
Ben Goertzel (3:38:14.620)
like what's in the mind of the early stage AGI
Lex Fridman (3:38:18.620)
as it first gains the ability to rewrite its own code base
Lex Fridman (3:38:21.780)
and project itself toward super intelligence.
Lex Fridman (3:38:24.740)
And if you believe that AI may move toward AGI
Ben Goertzel (3:38:31.180)
out of this sort of synergetic activity
Lex Fridman (3:38:33.300)
of many agents cooperating together
Ben Goertzel (3:38:35.780)
rather than just have one person's project,
Lex Fridman (3:38:37.860)
then who owns and controls that platform for AI cooperation
Lex Fridman (3:38:42.580)
becomes also very, very important, right?
Lex Fridman (3:38:47.260)
And is that platform AWS?
Lex Fridman (3:38:49.380)
Is it Google Cloud?
Lex Fridman (3:38:50.580)
Is it Alibaba or is it something more like the internet
Lex Fridman (3:38:53.420)
or Singularity Net, which is open and decentralized?
Lex Fridman (3:38:56.740)
So if all of my weird machinations come to pass, right?
Ben Goertzel (3:39:01.100)
I mean, we have the Hanson robots
Lex Fridman (3:39:03.740)
being a beautiful user interface,
Ben Goertzel (3:39:06.140)
gathering information on human values
Lex Fridman (3:39:09.100)
and being loving and compassionate to people in medical,
Ben Goertzel (3:39:12.060)
home service, robot office applications,
Lex Fridman (3:39:14.620)
you have Singularity Net in the backend
Ben Goertzel (3:39:16.900)
networking together many different AIs
Lex Fridman (3:39:19.460)
toward cooperative intelligence,
Ben Goertzel (3:39:21.500)
fueling the robots among many other things.
Lex Fridman (3:39:24.020)
You have OpenCog 2.0 and true AGI
Ben Goertzel (3:39:27.340)
as one of the sources of AI
Lex Fridman (3:39:29.420)
inside this decentralized network,
Ben Goertzel (3:39:31.700)
powering the robot and medical AIs
Lex Fridman (3:39:34.140)
helping us live a long time
Lex Fridman (3:39:36.300)
and cure diseases among other things.
Lex Fridman (3:39:39.740)
And this whole thing is operating
Lex Fridman (3:39:42.380)
in a democratic and decentralized way, right?
Lex Fridman (3:39:46.060)
And I think if anyone can pull something like this off,
Ben Goertzel (3:39:50.420)
whether using the specific technologies I've mentioned
Lex Fridman (3:39:53.900)
or something else, I mean,
Ben Goertzel (3:39:55.780)
then I think we have a higher odds
Lex Fridman (3:39:58.380)
of moving toward a beneficial technological singularity
Ben Goertzel (3:40:02.740)
rather than one in which the first super AGI
Lex Fridman (3:40:06.220)
is indifferent to humans
Lex Fridman (3:40:07.620)
and just considers us an inefficient use of molecules.
Lex Fridman (3:40:11.900)
That was a beautifully articulated vision for the world.
Lex Fridman (3:40:15.540)
So thank you for that.
Lex Fridman (3:40:16.700)
Well, let's talk a little bit about life and death.
Ben Goertzel (3:40:21.860)
I'm pro life and anti death for most people.
Lex Fridman (3:40:27.100)
There's few exceptions that I won't mention here.
Ben Goertzel (3:40:30.860)
I'm glad just like your dad,
Lex Fridman (3:40:32.340)
you're taking a stand against death.
Ben Goertzel (3:40:36.420)
You have, by the way, you have a bunch of awesome music
Lex Fridman (3:40:39.940)
where you play piano online.
Ben Goertzel (3:40:41.780)
One of the songs that I believe you've written
Lex Fridman (3:40:45.380)
the lyrics go, by the way, I like the way it sounds,
Ben Goertzel (3:40:49.140)
people should listen to it, it's awesome.
Lex Fridman (3:40:51.460)
I considered, I probably will cover it, it's a good song.
Lex Fridman (3:40:54.980)
Tell me why do you think it is a good thing
Lex Fridman (3:40:58.660)
that we all get old and die is one of the songs.
Ben Goertzel (3:41:01.980)
I love the way it sounds,
Lex Fridman (3:41:03.180)
but let me ask you about death first.
Lex Fridman (3:41:06.780)
Do you think there's an element to death
Lex Fridman (3:41:08.300)
that's essential to give our life meaning?
Ben Goertzel (3:41:12.260)
Like the fact that this thing ends.
Lex Fridman (3:41:14.020)
Well, let me say I'm pleased and a little embarrassed
Ben Goertzel (3:41:19.220)
you've been listening to that music I put online.
Lex Fridman (3:41:21.540)
That's awesome.
Ben Goertzel (3:41:22.380)
One of my regrets in life recently is I would love
Lex Fridman (3:41:25.540)
to get time to really produce music well.
Ben Goertzel (3:41:28.460)
Like I haven't touched my sequencer software
Lex Fridman (3:41:31.100)
in like five years.
Ben Goertzel (3:41:32.620)
I would love to like rehearse and produce and edit.
Lex Fridman (3:41:37.220)
But with a two year old baby
Lex Fridman (3:41:39.580)
and trying to create the singularity, there's no time.
Lex Fridman (3:41:42.260)
So I just made the decision to,
Ben Goertzel (3:41:45.660)
when I'm playing random shit in an off moment.
Lex Fridman (3:41:47.740)
Just record it.
Ben Goertzel (3:41:48.580)
Just record it, put it out there, like whatever.
Lex Fridman (3:41:51.820)
Maybe if I'm unfortunate enough to die,
Ben Goertzel (3:41:54.460)
maybe that can be input to the AGI
Lex Fridman (3:41:56.260)
when it tries to make an accurate mind upload of me, right?
Ben Goertzel (3:41:58.980)
Death is bad.
Lex Fridman (3:42:01.100)
I mean, that's very simple.
Ben Goertzel (3:42:02.700)
It's baffling we should have to say that.
Lex Fridman (3:42:04.300)
I mean, of course people can make meaning out of death.
Lex Fridman (3:42:08.740)
And if someone is tortured,
Lex Fridman (3:42:10.940)
maybe they can make beautiful meaning out of that torture
Lex Fridman (3:42:13.220)
and write a beautiful poem
Lex Fridman (3:42:14.540)
about what it was like to be tortured, right?
Ben Goertzel (3:42:16.980)
I mean, we're very creative.
Lex Fridman (3:42:19.100)
We can milk beauty and positivity
Ben Goertzel (3:42:22.420)
out of even the most horrible and shitty things.
Lex Fridman (3:42:25.300)
But just because if I was tortured,
Ben Goertzel (3:42:27.860)
I could write a good song
Lex Fridman (3:42:28.940)
about what it was like to be tortured,
Ben Goertzel (3:42:30.780)
doesn't make torture good.
Lex Fridman (3:42:31.980)
And just because people are able to derive meaning
Lex Fridman (3:42:35.660)
and value from death,
Lex Fridman (3:42:37.500)
doesn't mean they wouldn't derive even better meaning
Lex Fridman (3:42:39.620)
and value from ongoing life without death,
Lex Fridman (3:42:42.580)
which I very...
Ben Goertzel (3:42:43.420)
Indefinite.
Lex Fridman (3:42:44.260)
Yeah, yeah.
Lex Fridman (3:42:45.100)
So if you could live forever, would you live forever?
Lex Fridman (3:42:47.740)
Forever.
Ben Goertzel (3:42:50.460)
My goal with longevity research
Lex Fridman (3:42:52.820)
is to abolish the plague of involuntary death.
Ben Goertzel (3:42:57.460)
I don't think people should die unless they choose to die.
Lex Fridman (3:43:01.340)
If I had to choose forced immortality
Ben Goertzel (3:43:05.700)
versus dying, I would choose forced immortality.
Lex Fridman (3:43:09.180)
On the other hand, if I chose...
Ben Goertzel (3:43:11.860)
If I had the choice of immortality
Lex Fridman (3:43:13.500)
with the choice of suicide whenever I felt like it,
Ben Goertzel (3:43:15.620)
of course I would take that instead.
Lex Fridman (3:43:17.220)
And that's the more realistic choice.
Ben Goertzel (3:43:18.860)
I mean, there's no reason
Lex Fridman (3:43:20.180)
you should have forced immortality.
Ben Goertzel (3:43:21.660)
You should be able to live until you get sick of living,
Lex Fridman (3:43:25.780)
right?
Ben Goertzel (3:43:26.620)
I mean, that's...
Lex Fridman (3:43:27.460)
And that will seem insanely obvious
Ben Goertzel (3:43:29.780)
to everyone 50 years from now.
Lex Fridman (3:43:31.380)
And they will be so...
Ben Goertzel (3:43:33.180)
I mean, people who thought death gives meaning to life,
Lex Fridman (3:43:35.980)
so we should all die,
Ben Goertzel (3:43:37.660)
they will look at that 50 years from now
Lex Fridman (3:43:39.380)
the way we now look at the Anabaptists in the year 1000
Ben Goertzel (3:43:43.340)
who gave away all their positions,
Lex Fridman (3:43:45.180)
went on top of the mountain for Jesus
Ben Goertzel (3:43:47.700)
to come and bring them to the ascension.
Lex Fridman (3:43:50.220)
I mean, it's ridiculous that people think death is good
Ben Goertzel (3:43:55.740)
because you gain more wisdom as you approach dying.
Lex Fridman (3:44:00.180)
I mean, of course it's true.
Ben Goertzel (3:44:01.940)
I mean, I'm 53.
Lex Fridman (3:44:03.460)
And the fact that I might have only a few more decades left,
Ben Goertzel (3:44:08.220)
it does make me reflect on things differently.
Lex Fridman (3:44:11.460)
It does give me a deeper understanding of many things.
Lex Fridman (3:44:15.700)
But I mean, so what?
Lex Fridman (3:44:18.100)
You could get a deep understanding
Ben Goertzel (3:44:19.500)
in a lot of different ways.
Lex Fridman (3:44:20.900)
Pain is the same way.
Ben Goertzel (3:44:22.460)
We're gonna abolish pain.
Lex Fridman (3:44:24.260)
And that's even more amazing than abolishing death, right?
Ben Goertzel (3:44:27.460)
I mean, once we get a little better at neuroscience,
Lex Fridman (3:44:30.420)
we'll be able to go in and adjust the brain
Lex Fridman (3:44:32.660)
so that pain doesn't hurt anymore, right?
Lex Fridman (3:44:34.740)
And that, you know, people will say that's bad
Ben Goertzel (3:44:37.100)
because there's so much beauty
Lex Fridman (3:44:39.420)
in overcoming pain and suffering.
Ben Goertzel (3:44:41.100)
Oh, sure.
Lex Fridman (3:44:42.340)
And there's beauty in overcoming torture too.
Lex Fridman (3:44:45.220)
And some people like to cut themselves,
Lex Fridman (3:44:46.860)
but not many, right?
Ben Goertzel (3:44:48.100)
I mean.
Lex Fridman (3:44:48.940)
That's an interesting.
Ben Goertzel (3:44:49.780)
So, but to push, I mean, to push back again,
Lex Fridman (3:44:52.260)
this is the Russian side of me.
Ben Goertzel (3:44:53.300)
I do romanticize suffering.
Lex Fridman (3:44:55.020)
It's not obvious.
Ben Goertzel (3:44:56.380)
I mean, the way you put it, it seems very logical.
Lex Fridman (3:44:59.460)
It's almost absurd to romanticize suffering or pain
Ben Goertzel (3:45:02.820)
or death, but to me, a world without suffering,
Lex Fridman (3:45:07.740)
without pain, without death, it's not obvious.
Ben Goertzel (3:45:10.620)
Well, then you can stay in the people's zoo,
Lex Fridman (3:45:13.500)
people torturing each other.
Ben Goertzel (3:45:15.460)
No, but what I'm saying is I don't,
Lex Fridman (3:45:18.140)
well, that's, I guess what I'm trying to say,
Ben Goertzel (3:45:20.220)
I don't know if I was presented with that choice,
Lex Fridman (3:45:22.820)
what I would choose because it, to me.
Ben Goertzel (3:45:25.420)
This is a subtler, it's a subtler matter.
Lex Fridman (3:45:30.100)
And I've posed it in this conversation
Ben Goertzel (3:45:33.980)
in an unnecessarily extreme way.
Lex Fridman (3:45:37.100)
So I think, I think the way you should think about it
Ben Goertzel (3:45:41.060)
is what if there's a little dial on the side of your head
Lex Fridman (3:45:44.700)
and you could turn how much pain hurt,
Ben Goertzel (3:45:48.180)
turn it down to zero, turn it up to 11,
Lex Fridman (3:45:50.660)
like in spinal tap, if it wants,
Lex Fridman (3:45:52.220)
maybe through an actual spinal tap, right?
Lex Fridman (3:45:53.980)
So, I mean, would you opt to have that dial there or not?
Ben Goertzel (3:45:58.940)
That's the question.
Lex Fridman (3:45:59.780)
The question isn't whether you would turn the pain down
Ben Goertzel (3:46:02.300)
to zero all the time.
Lex Fridman (3:46:05.220)
Would you opt to have the dial or not?
Ben Goertzel (3:46:07.180)
My guess is that in some dark moment of your life,
Lex Fridman (3:46:10.000)
you would choose to have the dial implanted
Lex Fridman (3:46:12.180)
and then it would be there.
Lex Fridman (3:46:13.340)
Just to confess a small thing, don't ask me why,
Lex Fridman (3:46:17.180)
but I'm doing this physical challenge currently
Lex Fridman (3:46:20.760)
where I'm doing 680 pushups and pull ups a day.
Lex Fridman (3:46:25.860)
And my shoulder is currently, as we sit here,
Lex Fridman (3:46:29.180)
in a lot of pain.
Lex Fridman (3:46:30.700)
And I don't know, I would certainly right now,
Lex Fridman (3:46:35.860)
if you gave me a dial, I would turn that sucker to zero
Ben Goertzel (3:46:38.880)
as quickly as possible.
Lex Fridman (3:46:40.540)
But I think the whole point of this journey is,
Ben Goertzel (3:46:46.740)
I don't know.
Lex Fridman (3:46:47.580)
Well, because you're a twisted human being.
Ben Goertzel (3:46:49.540)
I'm a twisted, so the question is am I somehow twisted
Lex Fridman (3:46:53.580)
because I created some kind of narrative for myself
Lex Fridman (3:46:57.440)
so that I can deal with the injustice
Lex Fridman (3:47:00.820)
and the suffering in the world?
Ben Goertzel (3:47:03.700)
Or is this actually going to be a source of happiness
Lex Fridman (3:47:06.340)
for me?
Ben Goertzel (3:47:07.180)
Well, this is to an extent is a research question
Lex Fridman (3:47:10.820)
that humanity will undertake, right?
Lex Fridman (3:47:12.300)
So I mean, human beings do have a particular biological
Lex Fridman (3:47:17.300)
makeup, which sort of implies a certain probability
Lex Fridman (3:47:22.860)
distribution over motivational systems, right?
Lex Fridman (3:47:25.880)
So I mean, we, and that is there, that is there.
Ben Goertzel (3:47:30.400)
Now the question is how flexibly can that morph
Lex Fridman (3:47:36.540)
as society and technology change, right?
Lex Fridman (3:47:38.980)
So if we're given that dial and we're given a society
Lex Fridman (3:47:43.740)
in which say we don't have to work for a living
Lex Fridman (3:47:47.540)
and in which there's an ambient decentralized
Lex Fridman (3:47:50.700)
benevolent AI network that will warn us
Ben Goertzel (3:47:52.460)
when we're about to hurt ourself,
Lex Fridman (3:47:54.660)
if we're in a different context,
Ben Goertzel (3:47:57.060)
can we consistently with being genuinely and fully human,
Lex Fridman (3:48:02.880)
can we consistently get into a state of consciousness
Ben Goertzel (3:48:05.880)
where we just want to keep the pain dial turned
Lex Fridman (3:48:09.220)
all the way down and yet we're leading very rewarding
Lex Fridman (3:48:12.420)
and fulfilling lives, right?
Lex Fridman (3:48:13.860)
Now, I suspect the answer is yes, we can do that,
Lex Fridman (3:48:17.660)
but I don't know that, I don't know that for certain.
Lex Fridman (3:48:21.580)
Yeah, now I'm more confident that we could create
Ben Goertzel (3:48:25.960)
a nonhuman AGI system, which just didn't need an analog
Lex Fridman (3:48:31.220)
of feeling pain.
Lex Fridman (3:48:33.100)
And I think that AGI system will be fundamentally healthier
Lex Fridman (3:48:37.380)
and more benevolent than human beings.
Lex Fridman (3:48:39.740)
So I think it might or might not be true
Lex Fridman (3:48:42.340)
that humans need a certain element of suffering
Ben Goertzel (3:48:45.220)
to be satisfied humans, consistent with the human physiology.
Lex Fridman (3:48:49.460)
If it is true, that's one of the things that makes us fucked
Lex Fridman (3:48:53.220)
and disqualified to be the super AGI, right?
Lex Fridman (3:48:58.380)
I mean, the nature of the human motivational system
Ben Goertzel (3:49:03.620)
is that we seem to gravitate towards situations
Lex Fridman (3:49:08.620)
where the best thing in the large scale
Ben Goertzel (3:49:12.740)
is not the best thing in the small scale
Lex Fridman (3:49:15.860)
according to our subjective value system.
Lex Fridman (3:49:18.100)
So we gravitate towards subjective value judgments
Lex Fridman (3:49:20.740)
where to gratify ourselves in the large,
Ben Goertzel (3:49:22.940)
we have to ungratify ourselves in the small.
Lex Fridman (3:49:25.620)
And we do that in, you see that in music,
Ben Goertzel (3:49:29.340)
there's a theory of music which says
Lex Fridman (3:49:31.740)
the key to musical aesthetics
Ben Goertzel (3:49:33.780)
is the surprising fulfillment of expectations.
Lex Fridman (3:49:36.860)
Like you want something that will fulfill
Ben Goertzel (3:49:38.900)
the expectations are listed in the prior part of the music,
Lex Fridman (3:49:41.820)
but in a way with a bit of a twist that surprises you.
Lex Fridman (3:49:44.820)
And I mean, that's true not only in outdoor music
Lex Fridman (3:49:48.140)
like my own or that of Zappa or Steve Vai or Buckethead
Ben Goertzel (3:49:53.300)
or Christoph Pendergast or something,
Lex Fridman (3:49:55.460)
it's even there in Mozart or something.
Ben Goertzel (3:49:57.980)
It's not there in elevator music too much,
Lex Fridman (3:49:59.980)
but that's why it's boring, right?
Lex Fridman (3:50:02.940)
But wrapped up in there is we want to hurt a little bit
Lex Fridman (3:50:07.540)
so that we can feel the pain go away.
Ben Goertzel (3:50:11.300)
Like we wanna be a little confused by what's coming next.
Lex Fridman (3:50:15.700)
So then when the thing that comes next actually makes sense,
Lex Fridman (3:50:18.380)
it's so satisfying, right?
Lex Fridman (3:50:19.940)
That's the surprising fulfillment of expectations,
Lex Fridman (3:50:22.300)
is that what you said?
Lex Fridman (3:50:23.140)
Yeah, yeah, yeah.
Lex Fridman (3:50:23.960)
So beautifully put.
Lex Fridman (3:50:24.800)
We've been skirting around a little bit,
Lex Fridman (3:50:26.820)
but if I were to ask you the most ridiculous big question
Lex Fridman (3:50:29.380)
of what is the meaning of life,
Lex Fridman (3:50:32.740)
what would your answer be?
Lex Fridman (3:50:37.340)
Three values, joy, growth, and choice.
Ben Goertzel (3:50:43.580)
I think you need joy.
Lex Fridman (3:50:46.420)
I mean, that's the basis of everything.
Ben Goertzel (3:50:48.060)
If you want the number one value.
Lex Fridman (3:50:49.700)
On the other hand, I'm unsatisfied with a static joy
Ben Goertzel (3:50:54.860)
that doesn't progress perhaps because of some
Lex Fridman (3:50:58.100)
elemental element of human perversity,
Lex Fridman (3:51:00.140)
but the idea of something that grows
Lex Fridman (3:51:02.220)
and becomes more and more and better and better
Ben Goertzel (3:51:04.860)
in some sense appeals to me.
Lex Fridman (3:51:06.780)
But I also sort of like the idea of individuality
Ben Goertzel (3:51:10.580)
that as a distinct system, I have some agency.
Lex Fridman (3:51:14.500)
So there's some nexus of causality within this system
Ben Goertzel (3:51:18.820)
rather than the causality being wholly evenly distributed
Lex Fridman (3:51:22.420)
over the joyous growing mass.
Lex Fridman (3:51:23.920)
So you start with joy, growth, and choice
Lex Fridman (3:51:27.080)
as three basic values.
Ben Goertzel (3:51:28.860)
Those three things could continue indefinitely.
Lex Fridman (3:51:31.940)
That's something that can last forever.
Ben Goertzel (3:51:35.180)
Is there some aspect of something you called,
Lex Fridman (3:51:38.740)
which I like, super longevity that you find exciting?
Lex Fridman (3:51:44.980)
Is there research wise, is there ideas in that space that?
Lex Fridman (3:51:48.340)
I mean, I think, yeah, in terms of the meaning of life,
Ben Goertzel (3:51:53.240)
this really ties into that because for us as humans,
Lex Fridman (3:51:58.020)
probably the way to get the most joy, growth, and choice
Ben Goertzel (3:52:02.260)
is transhumanism and to go beyond the human form
Lex Fridman (3:52:06.180)
that we have right now, right?
Ben Goertzel (3:52:08.420)
I mean, I think human body is great
Lex Fridman (3:52:10.980)
and by no means do any of us maximize the potential
Ben Goertzel (3:52:15.140)
for joy, growth, and choice imminent in our human bodies.
Lex Fridman (3:52:18.560)
On the other hand, it's clear that other configurations
Ben Goertzel (3:52:21.780)
of matter could manifest even greater amounts
Lex Fridman (3:52:25.260)
of joy, growth, and choice than humans do,
Ben Goertzel (3:52:29.620)
maybe even finding ways to go beyond the realm of matter
Lex Fridman (3:52:33.140)
as we understand it right now.
Lex Fridman (3:52:34.940)
So I think in a practical sense,
Lex Fridman (3:52:38.100)
much of the meaning I see in human life
Ben Goertzel (3:52:40.740)
is to create something better than humans
Lex Fridman (3:52:42.880)
and go beyond human life.
Lex Fridman (3:52:45.460)
But certainly that's not all of it for me
Lex Fridman (3:52:47.980)
in a practical sense, right?
Ben Goertzel (3:52:49.220)
Like I have four kids and a granddaughter
Lex Fridman (3:52:51.740)
and many friends and parents and family
Lex Fridman (3:52:55.060)
and just enjoying everyday human social existence.
Lex Fridman (3:52:59.740)
But we can do even better.
Ben Goertzel (3:53:00.900)
Yeah, yeah.
Lex Fridman (3:53:01.740)
And I mean, I love, I've always,
Ben Goertzel (3:53:03.860)
when I could live near nature,
Lex Fridman (3:53:05.700)
I spend a bunch of time out in nature in the forest
Lex Fridman (3:53:08.740)
and on the water every day and so forth.
Lex Fridman (3:53:10.940)
So, I mean, enjoying the pleasant moment is part of it,
Lex Fridman (3:53:15.040)
but the growth and choice aspect are severely limited
Lex Fridman (3:53:20.780)
by our human biology.
Ben Goertzel (3:53:22.420)
In particular, dying seems to inhibit your potential
Lex Fridman (3:53:25.980)
for personal growth considerably as far as we know.
Ben Goertzel (3:53:29.520)
I mean, there's some element of life after death perhaps,
Lex Fridman (3:53:32.980)
but even if there is,
Lex Fridman (3:53:34.980)
why not also continue going in this biological realm, right?
Lex Fridman (3:53:39.300)
In super longevity, I mean,
Ben Goertzel (3:53:43.300)
you know, we haven't yet cured aging.
Lex Fridman (3:53:45.580)
We haven't yet cured death.
Ben Goertzel (3:53:48.020)
Certainly there's very interesting progress all around.
Lex Fridman (3:53:51.860)
I mean, CRISPR and gene editing can be an incredible tool.
Lex Fridman (3:53:57.220)
And I mean, right now,
Lex Fridman (3:54:00.120)
stem cells could potentially prolong life a lot.
Ben Goertzel (3:54:03.180)
Like if you got stem cell injections
Lex Fridman (3:54:05.980)
of just stem cells for every tissue of your body
Ben Goertzel (3:54:09.140)
injected into every tissue,
Lex Fridman (3:54:11.360)
and you can just have replacement of your old cells
Ben Goertzel (3:54:15.360)
with new cells produced by those stem cells,
Lex Fridman (3:54:17.340)
I mean, that could be highly impactful at prolonging life.
Ben Goertzel (3:54:21.240)
Now we just need slightly better technology
Lex Fridman (3:54:23.260)
for having them grow, right?
Lex Fridman (3:54:25.420)
So using machine learning to guide procedures
Lex Fridman (3:54:28.840)
for stem cell differentiation and trans differentiation,
Ben Goertzel (3:54:32.700)
it's kind of nitty gritty,
Lex Fridman (3:54:33.740)
but I mean, that's quite interesting.
Lex Fridman (3:54:36.680)
So I think there's a lot of different things being done
Lex Fridman (3:54:41.060)
to help with prolongation of human life,
Lex Fridman (3:54:44.740)
but we could do a lot better.
Lex Fridman (3:54:47.560)
So for example, the extracellular matrix,
Ben Goertzel (3:54:51.460)
which is the bunch of proteins
Lex Fridman (3:54:52.620)
in between the cells in your body,
Ben Goertzel (3:54:54.300)
they get stiffer and stiffer as you get older.
Lex Fridman (3:54:57.360)
And the extracellular matrix transmits information
Ben Goertzel (3:55:01.300)
both electrically, mechanically,
Lex Fridman (3:55:03.540)
and to some extent, biophotonically.
Lex Fridman (3:55:05.380)
So there's all this transmission
Lex Fridman (3:55:07.280)
through the parts of the body,
Lex Fridman (3:55:08.880)
but the stiffer the extracellular matrix gets,
Lex Fridman (3:55:11.860)
the less the transmission happens,
Ben Goertzel (3:55:13.520)
which makes your body get worse coordinated
Lex Fridman (3:55:15.660)
between the different organs as you get older.
Lex Fridman (3:55:17.460)
So my friend Christian Schaffmeister
Lex Fridman (3:55:19.460)
at my alumnus organization,
Ben Goertzel (3:55:22.460)
my Alma mater, the Great Temple University,
Lex Fridman (3:55:25.100)
Christian Schaffmeister has a potential solution to this,
Ben Goertzel (3:55:28.640)
where he has these novel molecules called spiral ligamers,
Lex Fridman (3:55:32.340)
which are like polymers that are not organic.
Ben Goertzel (3:55:34.440)
They're specially designed polymers
Lex Fridman (3:55:37.780)
so that you can algorithmically predict
Ben Goertzel (3:55:39.420)
exactly how they'll fold very simply.
Lex Fridman (3:55:41.580)
So he designed the molecular scissors
Ben Goertzel (3:55:43.280)
that have spiral ligamers that you could eat
Lex Fridman (3:55:45.560)
and would then cut through all the glucosamine
Lex Fridman (3:55:49.220)
and other crosslink proteins
Lex Fridman (3:55:50.620)
in your extracellular matrix, right?
Lex Fridman (3:55:52.760)
But to make that technology really work
Lex Fridman (3:55:55.200)
and be mature as several years of work,
Ben Goertzel (3:55:56.860)
as far as I know, no one's finding it at the moment.
Lex Fridman (3:56:00.140)
So there's so many different ways
Ben Goertzel (3:56:02.380)
that technology could be used to prolong longevity.
Lex Fridman (3:56:05.080)
What we really need,
Ben Goertzel (3:56:06.540)
we need an integrated database of all biological knowledge
Lex Fridman (3:56:09.580)
about human beings and model organisms,
Ben Goertzel (3:56:12.020)
like hopefully a massively distributed
Lex Fridman (3:56:14.480)
open cog bioatom space,
Lex Fridman (3:56:15.980)
but it can exist in other forms too.
Lex Fridman (3:56:18.260)
We need that data to be opened up
Ben Goertzel (3:56:20.860)
in a suitably privacy protecting way.
Lex Fridman (3:56:23.300)
We need massive funding into machine learning,
Ben Goertzel (3:56:26.100)
AGI, proto AGI statistical research
Lex Fridman (3:56:29.240)
aimed at solving biology,
Ben Goertzel (3:56:31.240)
both molecular biology and human biology
Lex Fridman (3:56:33.440)
based on this massive data set, right?
Lex Fridman (3:56:36.700)
And then we need regulators not to stop people
Lex Fridman (3:56:40.700)
from trying radical therapies on themselves
Ben Goertzel (3:56:43.820)
if they so wish to,
Lex Fridman (3:56:46.180)
as well as better cloud based platforms
Ben Goertzel (3:56:49.420)
for like automated experimentation on microorganisms,
Lex Fridman (3:56:52.720)
flies and mice and so forth.
Lex Fridman (3:56:54.300)
And we could do all this.
Lex Fridman (3:56:55.820)
You look after the last financial crisis,
Ben Goertzel (3:56:58.900)
Obama, who I generally like pretty well,
Lex Fridman (3:57:01.300)
but he gave $4 trillion to large banks
Lex Fridman (3:57:03.740)
and insurance companies.
Lex Fridman (3:57:05.400)
You know, now in this COVID crisis,
Ben Goertzel (3:57:08.420)
trillions are being spent to help everyday people
Lex Fridman (3:57:10.780)
and small businesses.
Ben Goertzel (3:57:12.240)
In the end, we'll probably will find many more trillions
Lex Fridman (3:57:14.580)
are being given to large banks and insurance companies.
Ben Goertzel (3:57:17.220)
Anyway, like could the world put $10 trillion
Lex Fridman (3:57:21.020)
into making a massive holistic bio AI and bio simulation
Lex Fridman (3:57:25.560)
and experimental biology infrastructure?
Lex Fridman (3:57:27.800)
We could, we could put $10 trillion into that
Ben Goertzel (3:57:30.600)
without even screwing us up too badly.
Lex Fridman (3:57:32.300)
Just as in the end COVID and the last financial crisis
Ben Goertzel (3:57:35.260)
won't screw up the world economy so badly.
Lex Fridman (3:57:37.900)
We're not putting $10 trillion into that.
Ben Goertzel (3:57:39.900)
Instead, all this research is siloed inside
Lex Fridman (3:57:43.140)
a few big companies and government agencies.
Lex Fridman (3:57:46.820)
And most of the data that comes from our individual bodies
Lex Fridman (3:57:51.140)
personally, that could feed this AI to solve aging
Lex Fridman (3:57:54.340)
and death, most of that data is sitting
Lex Fridman (3:57:56.820)
in some hospital's database doing nothing, right?
Ben Goertzel (3:58:03.960)
I got two more quick questions for you.
Lex Fridman (3:58:07.160)
One, I know a lot of people are gonna ask me,
Ben Goertzel (3:58:09.820)
you are on the Joe Rogan podcast
Lex Fridman (3:58:11.740)
wearing that same amazing hat.
Lex Fridman (3:58:14.860)
Do you have a origin story for the hat?
Lex Fridman (3:58:17.500)
Does the hat have its own story that you're able to share?
Ben Goertzel (3:58:21.420)
The hat story has not been told yet.
Lex Fridman (3:58:23.180)
So we're gonna have to come back
Lex Fridman (3:58:24.220)
and you can interview the hat.
Lex Fridman (3:58:27.880)
We'll leave that for the hat's own interview.
Ben Goertzel (3:58:30.060)
All right.
Lex Fridman (3:58:30.900)
It's too much to pack into.
Lex Fridman (3:58:32.100)
Is there a book?
Lex Fridman (3:58:32.940)
Is the hat gonna write a book?
Ben Goertzel (3:58:34.320)
Okay.
Lex Fridman (3:58:35.160)
Well, it may transmit the information
Ben Goertzel (3:58:38.340)
through direct neural transmission.
Lex Fridman (3:58:40.020)
Okay, so it's actually,
Ben Goertzel (3:58:41.420)
there might be some Neuralink competition there.
Lex Fridman (3:58:44.780)
Beautiful, we'll leave it as a mystery.
Ben Goertzel (3:58:46.900)
Maybe one last question.
Lex Fridman (3:58:49.040)
If you build an AGI system,
Ben Goertzel (3:58:54.580)
you're successful at building the AGI system
Lex Fridman (3:58:58.540)
that could lead us to the singularity
Lex Fridman (3:59:00.420)
and you get to talk to her and ask her one question,
Lex Fridman (3:59:04.560)
what would that question be?
Ben Goertzel (3:59:05.960)
We're not allowed to ask,
Lex Fridman (3:59:08.140)
what is the question I should be asking?
Ben Goertzel (3:59:10.220)
Yeah, that would be cheating,
Lex Fridman (3:59:12.220)
but I guess that's a good question.
Ben Goertzel (3:59:14.040)
I'm thinking of a,
Lex Fridman (3:59:15.700)
I wrote a story with Stefan Bugay once
Ben Goertzel (3:59:18.600)
where these AI developers,
Lex Fridman (3:59:23.380)
they created a super smart AI
Ben Goertzel (3:59:25.900)
aimed at answering all the philosophical questions
Lex Fridman (3:59:31.220)
that have been worrying them.
Lex Fridman (3:59:32.060)
Like what is the meaning of life?
Lex Fridman (3:59:34.260)
Is there free will?
Lex Fridman (3:59:35.700)
What is consciousness and so forth?
Lex Fridman (3:59:37.980)
So they got the super AGI built
Lex Fridman (3:59:40.380)
and it turned a while.
Lex Fridman (3:59:43.300)
It said, those are really stupid questions.
Lex Fridman (3:59:46.580)
And then it puts off on a spaceship and left the earth.
Lex Fridman (3:59:51.420)
So you'd be afraid of scaring it off.
Ben Goertzel (3:59:55.540)
That's it, yeah.
Lex Fridman (3:59:56.500)
I mean, honestly, there is no one question
Ben Goertzel (40:02.580)
than I thought it was when I was a teenager.
Lex Fridman (40:04.620)
And I think you could have a human society
Ben Goertzel (40:08.260)
that was dialed dramatically further toward,
Lex Fridman (40:11.420)
you know, self awareness, other awareness,
Ben Goertzel (40:13.700)
compassion and sharing than our current society.
Lex Fridman (40:16.980)
And of course, greater material abundance helps,
Lex Fridman (40:20.580)
but to some extent material abundance
Lex Fridman (40:23.480)
is a subjective perception also
Ben Goertzel (40:25.380)
because many Stone Age cultures perceive themselves
Lex Fridman (40:28.260)
as living in great material abundance
Ben Goertzel (40:30.540)
that they had all the food and water they wanted,
Lex Fridman (40:32.100)
they lived in a beautiful place,
Ben Goertzel (40:33.500)
that they had sex lives, that they had children.
Lex Fridman (40:37.460)
I mean, they had abundance without any factories, right?
Lex Fridman (40:42.940)
So I think humanity probably would be capable
Lex Fridman (40:46.460)
of fundamentally more positive and joy filled mode
Ben Goertzel (40:51.140)
of social existence than what we have now.
Lex Fridman (40:57.320)
Clearly Marx didn't quite have the right idea
Ben Goertzel (40:59.500)
about how to get there.
Lex Fridman (41:01.800)
I mean, he missed a number of key aspects
Ben Goertzel (41:05.660)
of human society and its evolution.
Lex Fridman (41:09.500)
And if we look at where we are in society now,
Lex Fridman (41:13.140)
how to get there is a quite different question
Lex Fridman (41:15.760)
because there are very powerful forces
Ben Goertzel (41:18.100)
pushing people in different directions
Lex Fridman (41:21.080)
than a positive, joyous, compassionate existence, right?
Lex Fridman (41:26.380)
So if we were tried to, you know,
Lex Fridman (41:28.820)
Elon Musk is dreams of colonizing Mars at the moment,
Lex Fridman (41:32.820)
so we maybe will have a chance to start a new civilization
Lex Fridman (41:36.880)
with a new governmental system.
Lex Fridman (41:38.400)
And certainly there's quite a bit of chaos.
Lex Fridman (41:41.580)
We're sitting now, I don't know what the date is,
Lex Fridman (41:44.320)
but this is June.
Lex Fridman (41:46.860)
There's quite a bit of chaos in all different forms
Ben Goertzel (41:49.260)
going on in the United States and all over the world.
Lex Fridman (41:52.060)
So there's a hunger for new types of governments,
Ben Goertzel (41:55.560)
new types of leadership, new types of systems.
Lex Fridman (41:59.860)
And so what are the forces at play
Lex Fridman (42:01.980)
and how do we move forward?
Lex Fridman (42:04.140)
Yeah, I mean, colonizing Mars, first of all,
Ben Goertzel (42:06.780)
it's a super cool thing to do.
Lex Fridman (42:08.980)
We should be doing it.
Lex Fridman (42:10.060)
So you love the idea.
Lex Fridman (42:11.540)
Yeah, I mean, it's more important than making
Ben Goertzel (42:14.780)
chocolatey or chocolates and sexier lingerie
Lex Fridman (42:18.540)
and many of the things that we spend
Lex Fridman (42:21.020)
a lot more resources on as a species, right?
Lex Fridman (42:24.120)
So I mean, we certainly should do it.
Ben Goertzel (42:26.480)
I think the possible futures in which a Mars colony
Lex Fridman (42:33.180)
makes a critical difference for humanity are very few.
Ben Goertzel (42:38.040)
I mean, I think, I mean, assuming we make a Mars colony
Lex Fridman (42:42.220)
and people go live there in a couple of decades,
Ben Goertzel (42:44.000)
I mean, their supplies are gonna come from Earth.
Lex Fridman (42:46.380)
The money to make the colony came from Earth
Lex Fridman (42:48.820)
and whatever powers are supplying the goods there
Lex Fridman (42:53.740)
from Earth are gonna, in effect, be in control
Ben Goertzel (42:56.820)
of that Mars colony.
Lex Fridman (42:58.700)
Of course, there are outlier situations
Ben Goertzel (43:02.060)
where Earth gets nuked into oblivion
Lex Fridman (43:06.460)
and somehow Mars has been made self sustaining by that point
Lex Fridman (43:10.780)
and then Mars is what allows humanity to persist.
Lex Fridman (43:14.220)
But I think that those are very, very, very unlikely.
Lex Fridman (43:19.740)
You don't think it could be a first step on a long journey?
Lex Fridman (43:23.020)
Of course it's a first step on a long journey,
Ben Goertzel (43:24.740)
which is awesome.
Lex Fridman (43:27.140)
I'm guessing the colonization of the rest
Ben Goertzel (43:30.980)
of the physical universe will probably be done
Lex Fridman (43:33.260)
by AGI's that are better designed to live in space
Ben Goertzel (43:38.140)
than by the meat machines that we are.
Lex Fridman (43:41.840)
But I mean, who knows?
Ben Goertzel (43:43.020)
We may cryopreserve ourselves in some superior way
Lex Fridman (43:45.860)
to what we know now and like shoot ourselves out
Ben Goertzel (43:48.700)
to Alpha Centauri and beyond.
Lex Fridman (43:50.720)
I mean, that's all cool.
Ben Goertzel (43:52.660)
It's very interesting and it's much more valuable
Lex Fridman (43:55.140)
than most things that humanity is spending its resources on.
Ben Goertzel (43:58.860)
On the other hand, with AGI, we can get to a singularity
Lex Fridman (44:03.540)
before the Mars colony becomes sustaining for sure,
Ben Goertzel (44:07.780)
possibly before it's even operational.
Lex Fridman (44:10.100)
So your intuition is that that's the problem
Ben Goertzel (44:12.400)
if we really invest resources and we can get to faster
Lex Fridman (44:14.940)
than a legitimate full self sustaining colonization of Mars.
Ben Goertzel (44:19.700)
Yeah, and it's very clear that we will to me
Lex Fridman (44:23.160)
because there's so much economic value
Ben Goertzel (44:26.020)
in getting from narrow AI toward AGI,
Lex Fridman (44:29.460)
whereas the Mars colony, there's less economic value
Ben Goertzel (44:33.380)
until you get quite far out into the future.
Lex Fridman (44:37.380)
So I think that's very interesting.
Ben Goertzel (44:40.260)
I just think it's somewhat off to the side.
Lex Fridman (44:44.380)
I mean, just as I think, say, art and music
Ben Goertzel (44:48.020)
are very, very interesting and I wanna see resources
Lex Fridman (44:51.860)
go into amazing art and music being created.
Lex Fridman (44:55.460)
And I'd rather see that than a lot of the garbage
Lex Fridman (44:59.580)
that the society spends their money on.
Ben Goertzel (45:01.760)
On the other hand, I don't think Mars colonization
Lex Fridman (45:04.620)
or inventing amazing new genres of music
Ben Goertzel (45:07.780)
is not one of the things that is most likely
Lex Fridman (45:11.000)
to make a critical difference in the evolution
Ben Goertzel (45:13.900)
of human or nonhuman life in this part of the universe
Lex Fridman (45:18.340)
over the next decade.
Lex Fridman (45:19.820)
Do you think AGI is really?
Lex Fridman (45:21.620)
AGI is by far the most important thing
Ben Goertzel (45:25.820)
that's on the horizon.
Lex Fridman (45:27.500)
And then technologies that have direct ability
Ben Goertzel (45:31.620)
to enable AGI or to accelerate AGI are also very important.
Lex Fridman (45:37.260)
For example, say, quantum computing.
Ben Goertzel (45:40.540)
I don't think that's critical to achieve AGI,
Lex Fridman (45:42.740)
but certainly you could see how
Ben Goertzel (45:44.360)
the right quantum computing architecture
Lex Fridman (45:46.700)
could massively accelerate AGI,
Ben Goertzel (45:49.280)
similar other types of nanotechnology.
Lex Fridman (45:52.260)
Right now, the quest to cure aging and end disease
Ben Goertzel (45:57.860)
while not in the big picture as important as AGI,
Lex Fridman (46:02.100)
of course, it's important to all of us as individual humans.
Lex Fridman (46:07.380)
And if someone made a super longevity pill
Lex Fridman (46:11.600)
and distributed it tomorrow, I mean,
Ben Goertzel (46:14.260)
that would be huge and a much larger impact
Lex Fridman (46:17.220)
than a Mars colony is gonna have for quite some time.
Lex Fridman (46:20.460)
But perhaps not as much as an AGI system.
Lex Fridman (46:23.300)
No, because if you can make a benevolent AGI,
Ben Goertzel (46:27.060)
then all the other problems are solved.
Lex Fridman (46:28.700)
I mean, if then the AGI can be,
Ben Goertzel (46:31.940)
once it's as generally intelligent as humans,
Lex Fridman (46:34.260)
it can rapidly become massively more generally intelligent
Ben Goertzel (46:37.420)
than humans.
Lex Fridman (46:38.620)
And then that AGI should be able to solve science
Lex Fridman (46:42.540)
and engineering problems much better than human beings,
Lex Fridman (46:46.840)
as long as it is in fact motivated to do so.
Ben Goertzel (46:49.700)
That's why I said a benevolent AGI.
Lex Fridman (46:52.740)
There could be other kinds.
Ben Goertzel (46:54.020)
Maybe it's good to step back a little bit.
Lex Fridman (46:56.020)
I mean, we've been using the term AGI.
Ben Goertzel (46:58.860)
People often cite you as the creator,
Lex Fridman (47:00.860)
or at least the popularizer of the term AGI,
Ben Goertzel (47:03.060)
artificial general intelligence.
Lex Fridman (47:05.700)
Can you tell the origin story of the term maybe?
Lex Fridman (47:09.100)
So yeah, I would say I launched the term AGI upon the world
Lex Fridman (47:14.860)
for what it's worth without ever fully being in love
Ben Goertzel (47:19.940)
with the term.
Lex Fridman (47:21.660)
What happened is I was editing a book,
Lex Fridman (47:25.380)
and this process started around 2001 or two.
Lex Fridman (47:27.860)
I think the book came out 2005, finally.
Ben Goertzel (47:30.500)
I was editing a book which I provisionally
Lex Fridman (47:33.140)
was titling Real AI.
Lex Fridman (47:35.860)
And I mean, the goal was to gather together
Lex Fridman (47:38.840)
fairly serious academicish papers
Ben Goertzel (47:41.700)
on the topic of making thinking machines
Lex Fridman (47:43.940)
that could really think in the sense like people can,
Lex Fridman (47:46.780)
or even more broadly than people can, right?
Lex Fridman (47:49.240)
So then I was reaching out to other folks
Ben Goertzel (47:52.740)
that I had encountered here or there
Lex Fridman (47:54.060)
who were interested in that,
Ben Goertzel (47:57.380)
which included some other folks who I knew
Lex Fridman (48:01.700)
from the transhumist and singularitarian world,
Ben Goertzel (48:04.340)
like Peter Vos, who has a company, AGI Incorporated,
Lex Fridman (48:07.660)
still in California, and included Shane Legge,
Ben Goertzel (48:13.100)
who had worked for me at my company, WebMind,
Lex Fridman (48:15.700)
in New York in the late 90s,
Ben Goertzel (48:17.580)
who by now has become rich and famous.
Lex Fridman (48:20.500)
He was one of the cofounders of Google DeepMind.
Lex Fridman (48:22.780)
But at that time, Shane was,
Lex Fridman (48:25.320)
I think he may have just started doing his PhD
Ben Goertzel (48:31.800)
with Marcus Hooter, who at that time
Lex Fridman (48:35.900)
hadn't yet published his book, Universal AI,
Ben Goertzel (48:38.680)
which sort of gives a mathematical foundation
Lex Fridman (48:41.040)
for artificial general intelligence.
Lex Fridman (48:43.400)
So I reached out to Shane and Marcus and Peter Vos
Lex Fridman (48:46.140)
and Pei Wang, who was another former employee of mine
Ben Goertzel (48:49.480)
who had been Douglas Hofstadter's PhD student
Lex Fridman (48:51.880)
who had his own approach to AGI,
Lex Fridman (48:53.280)
and a bunch of some Russian folks reached out to these guys
Lex Fridman (48:58.040)
and they contributed papers for the book.
Lex Fridman (49:01.360)
But that was my provisional title, but I never loved it
Lex Fridman (49:04.440)
because in the end, I was doing some,
Lex Fridman (49:09.320)
what we would now call narrow AI as well,
Lex Fridman (49:12.120)
like applying machine learning to genomics data
Ben Goertzel (49:14.640)
or chat data for sentiment analysis.
Lex Fridman (49:17.920)
I mean, that work is real.
Lex Fridman (49:19.240)
And in a sense, it's really AI.
Lex Fridman (49:22.760)
It's just a different kind of AI.
Ben Goertzel (49:26.000)
Ray Kurzweil wrote about narrow AI versus strong AI,
Lex Fridman (49:31.160)
but that seemed weird to me because first of all,
Ben Goertzel (49:35.040)
narrow and strong are not antennas.
Lex Fridman (49:36.680)
That's right.
Lex Fridman (49:38.720)
But secondly, strong AI was used
Lex Fridman (49:41.940)
in the cognitive science literature
Ben Goertzel (49:43.360)
to mean the hypothesis that digital computer AIs
Lex Fridman (49:46.640)
could have true consciousness like human beings.
Lex Fridman (49:50.140)
So there was already a meaning to strong AI,
Lex Fridman (49:52.540)
which was complexly different, but related, right?
Lex Fridman (49:56.440)
So we were tossing around on an email list
Lex Fridman (4:00:01.500)
that rises among all the others, really.
Ben Goertzel (4:00:08.540)
I mean, what interests me more
Lex Fridman (4:00:10.020)
is upgrading my own intelligence
Lex Fridman (4:00:13.500)
so that I can absorb the whole world view of the super AGI.
Lex Fridman (4:00:19.380)
But I mean, of course, if the answer could be like,
Lex Fridman (4:00:23.100)
what is the chemical formula for the immortality pill?
Lex Fridman (4:00:27.500)
Like then I would do that or emit a bit string,
Ben Goertzel (4:00:33.340)
which will be the code for a super AGI
Lex Fridman (4:00:38.740)
on the Intel i7 processor.
Lex Fridman (4:00:41.220)
So those would be good questions.
Lex Fridman (4:00:42.860)
So if your own mind was expanded
Ben Goertzel (4:00:46.260)
to become super intelligent, like you're describing,
Lex Fridman (4:00:49.340)
I mean, there's kind of a notion
Ben Goertzel (4:00:53.500)
that intelligence is a burden, that it's possible
Lex Fridman (4:00:57.840)
that with greater and greater intelligence,
Ben Goertzel (4:01:00.020)
that other metric of joy that you mentioned
Lex Fridman (4:01:03.020)
becomes more and more difficult.
Lex Fridman (4:01:04.740)
What's your sense?
Lex Fridman (4:01:05.900)
Pretty stupid idea.
Lex Fridman (4:01:08.260)
So you think if you're super intelligent,
Lex Fridman (4:01:09.860)
you can also be super joyful?
Ben Goertzel (4:01:11.460)
I think getting root access to your own brain
Lex Fridman (4:01:15.460)
will enable new forms of joy that we don't have now.
Lex Fridman (4:01:19.220)
And I think as I've said before,
Lex Fridman (4:01:22.740)
what I aim at is really make multiple versions of myself.
Lex Fridman (4:01:27.820)
So I would like to keep one version,
Lex Fridman (4:01:30.180)
which is basically human like I am now,
Lex Fridman (4:01:33.580)
but keep the dial to turn pain up and down
Lex Fridman (4:01:36.980)
and get rid of death, right?
Lex Fridman (4:01:38.580)
And make another version which fuses its mind
Lex Fridman (4:01:43.640)
with superhuman AGI,
Lex Fridman (4:01:46.600)
and then will become massively transhuman.
Lex Fridman (4:01:50.060)
And whether it will send some messages back
Ben Goertzel (4:01:52.800)
to the human me or not will be interesting to find out.
Lex Fridman (4:01:55.580)
The thing is, once you're a super AGI,
Ben Goertzel (4:01:58.500)
like one subjective second to a human
Lex Fridman (4:02:01.540)
might be like a million subjective years
Lex Fridman (4:02:03.620)
to that super AGI, right?
Lex Fridman (4:02:04.980)
So it would be on a whole different basis.
Ben Goertzel (4:02:07.580)
I mean, at very least those two copies will be good to have,
Lex Fridman (4:02:10.940)
but it could be interesting to put your mind
Ben Goertzel (4:02:13.980)
into a dolphin or a space amoeba
Lex Fridman (4:02:16.860)
or all sorts of other things.
Ben Goertzel (4:02:18.520)
You can imagine one version that doubled its intelligence
Lex Fridman (4:02:21.060)
every year and another version that just became
Lex Fridman (4:02:24.140)
a super AGI as fast as possible, right?
Lex Fridman (4:02:26.140)
So, I mean, now we're sort of constrained to think
Lex Fridman (4:02:29.780)
one mind, one self, one body, right?
Lex Fridman (4:02:33.020)
But I think we actually, we don't need to be that
Ben Goertzel (4:02:36.260)
constrained in thinking about future intelligence
Lex Fridman (4:02:40.820)
after we've mastered AGI and nanotechnology
Lex Fridman (4:02:44.280)
and longevity biology.
Lex Fridman (4:02:47.820)
I mean, then each of our minds
Lex Fridman (4:02:49.540)
is a certain pattern of organization, right?
Lex Fridman (4:02:52.020)
And I know we haven't talked about consciousness,
Lex Fridman (4:02:54.300)
but I sort of, I'm panpsychist.
Lex Fridman (4:02:56.860)
I sort of view the universe as conscious.
Lex Fridman (4:03:00.080)
And so, you know, a light bulb or a quark
Lex Fridman (4:03:03.860)
or an ant or a worm or a monkey
Ben Goertzel (4:03:06.040)
have their own manifestations of consciousness.
Lex Fridman (4:03:08.780)
And the human manifestation of consciousness,
Ben Goertzel (4:03:11.900)
it's partly tied to the particular meat
Lex Fridman (4:03:15.580)
that we're manifested by, but it's largely tied
Lex Fridman (4:03:19.380)
to the pattern of organization in the brain, right?
Lex Fridman (4:03:22.360)
So, if you upload yourself into a computer
Ben Goertzel (4:03:25.040)
or a robot or whatever else it is,
Lex Fridman (4:03:28.640)
some element of your human consciousness may not be there
Ben Goertzel (4:03:31.780)
because it's just tied to the biological embodiment.
Lex Fridman (4:03:34.260)
But I think most of it will be there.
Lex Fridman (4:03:36.300)
And these will be incarnations of your consciousness
Lex Fridman (4:03:40.020)
in a slightly different flavor.
Ben Goertzel (4:03:42.500)
And, you know, creating these different versions
Lex Fridman (4:03:45.600)
will be amazing, and each of them will discover
Ben Goertzel (4:03:48.500)
meanings of life that have some overlap,
Lex Fridman (4:03:52.020)
but probably not total overlap
Ben Goertzel (4:03:54.300)
with the human Ben's meaning of life.
Lex Fridman (4:03:59.260)
The thing is, to get to that future
Ben Goertzel (4:04:02.940)
where we can explore different varieties of joy,
Lex Fridman (4:04:06.500)
different variations of human experience and values
Lex Fridman (4:04:09.680)
and transhuman experiences and values to get to that future,
Lex Fridman (4:04:13.140)
we need to navigate through a whole lot of human bullshit
Ben Goertzel (4:04:16.780)
of companies and governments and killer drones
Lex Fridman (4:04:21.480)
and making and losing money and so forth, right?
Lex Fridman (4:04:25.460)
And that's the challenge we're facing now
Lex Fridman (4:04:28.580)
is if we do things right,
Ben Goertzel (4:04:30.740)
we can get to a benevolent singularity,
Lex Fridman (4:04:33.580)
which is levels of joy, growth, and choice
Ben Goertzel (4:04:36.320)
that are literally unimaginable to human beings.
Lex Fridman (4:04:39.920)
If we do things wrong,
Ben Goertzel (4:04:41.720)
we could either annihilate all life on the planet,
Lex Fridman (4:04:44.120)
or we could lead to a scenario where, say,
Ben Goertzel (4:04:47.060)
all humans are annihilated and there's some super AGI
Lex Fridman (4:04:52.140)
that goes on and does its own thing unrelated to us
Ben Goertzel (4:04:55.460)
except via our role in originating it.
Lex Fridman (4:04:58.380)
And we may well be at a bifurcation point now, right?
Ben Goertzel (4:05:02.420)
Where what we do now has significant causal impact
Lex Fridman (4:05:05.820)
on what comes about,
Lex Fridman (4:05:06.720)
and yet most people on the planet
Lex Fridman (4:05:09.040)
aren't thinking that way whatsoever,
Ben Goertzel (4:05:11.540)
they're thinking only about their own narrow aims
Lex Fridman (4:05:16.220)
and aims and goals, right?
Ben Goertzel (4:05:17.780)
Now, of course, I'm thinking about my own narrow aims
Lex Fridman (4:05:20.880)
and goals to some extent also,
Lex Fridman (4:05:24.260)
but I'm trying to use as much of my energy and mind as I can
Lex Fridman (4:05:29.480)
to push toward this more benevolent alternative,
Ben Goertzel (4:05:33.200)
which will be better for me,
Lex Fridman (4:05:34.660)
but also for everybody else.
Lex Fridman (4:05:37.980)
And it's weird that so few people understand
Lex Fridman (4:05:42.540)
what's going on.
Ben Goertzel (4:05:43.380)
I know you interviewed Elon Musk,
Lex Fridman (4:05:44.780)
and he understands a lot of what's going on,
Lex Fridman (4:05:47.380)
but he's much more paranoid than I am, right?
Lex Fridman (4:05:49.620)
Because Elon gets that AGI
Ben Goertzel (4:05:52.040)
is gonna be way, way smarter than people,
Lex Fridman (4:05:54.260)
and he gets that an AGI does not necessarily
Ben Goertzel (4:05:57.100)
have to give a shit about people
Lex Fridman (4:05:58.740)
because we're a very elementary mode of organization
Ben Goertzel (4:06:01.660)
of matter compared to many AGI's.
Lex Fridman (4:06:04.700)
But I don't think he has a clear vision
Ben Goertzel (4:06:06.340)
of how infusing early stage AGI's
Lex Fridman (4:06:10.140)
with compassion and human warmth
Ben Goertzel (4:06:13.540)
can lead to an AGI that loves and helps people
Lex Fridman (4:06:18.020)
rather than viewing us as a historical artifact
Lex Fridman (4:06:22.860)
and a waste of mass energy.
Lex Fridman (4:06:26.200)
But on the other hand,
Ben Goertzel (4:06:28.060)
while I have some disagreements with him,
Lex Fridman (4:06:29.600)
like he understands way, way more of the story
Ben Goertzel (4:06:33.140)
than almost anyone else
Lex Fridman (4:06:34.820)
in such a large scale corporate leadership position, right?
Ben Goertzel (4:06:38.180)
It's terrible how little understanding
Lex Fridman (4:06:40.740)
of these fundamental issues exists out there now.
Ben Goertzel (4:06:45.060)
That may be different five or 10 years from now though,
Lex Fridman (4:06:47.220)
because I can see understanding of AGI and longevity
Lex Fridman (4:06:51.180)
and other such issues is certainly much stronger
Lex Fridman (4:06:54.620)
and more prevalent now than 10 or 15 years ago, right?
Lex Fridman (4:06:57.620)
So I mean, humanity as a whole can be slow learners
Lex Fridman (4:07:02.860)
relative to what I would like,
Lex Fridman (4:07:05.460)
but on a historical sense, on the other hand,
Lex Fridman (4:07:08.400)
you could say the progress is astoundingly fast.
Lex Fridman (4:07:11.220)
But Elon also said, I think on the Joe Rogan podcast,
Lex Fridman (4:07:15.640)
that love is the answer.
Lex Fridman (4:07:17.380)
So maybe in that way, you and him are both on the same page
Lex Fridman (4:07:21.820)
of how we should proceed with AGI.
Ben Goertzel (4:07:24.420)
I think there's no better place to end it.
Lex Fridman (4:07:27.300)
I hope we get to talk again about the hat
Lex Fridman (4:07:30.860)
and about consciousness
Lex Fridman (4:07:32.020)
and about a million topics we didn't cover.
Ben Goertzel (4:07:34.500)
Ben, it's a huge honor to talk to you.
Lex Fridman (4:07:36.340)
Thank you for making it out.
Ben Goertzel (4:07:37.540)
Thank you for talking today.
Lex Fridman (4:07:39.540)
Thanks for having me.
Ben Goertzel (4:07:40.440)
This was really, really good fun
Lex Fridman (4:07:44.380)
and we dug deep into some very important things.
Lex Fridman (4:07:47.420)
So thanks for doing this.
Lex Fridman (4:07:48.740)
Thanks very much.
Ben Goertzel (4:07:49.820)
Awesome.
Lex Fridman (4:07:51.200)
Thanks for listening to this conversation with Ben Gertzel
Lex Fridman (4:07:53.860)
and thank you to our sponsors,
Lex Fridman (4:07:55.860)
The Jordan Harbinger Show and Masterclass.
Ben Goertzel (4:07:59.380)
Please consider supporting the podcast
Lex Fridman (4:08:01.080)
by going to jordanharbinger.com slash lex
Lex Fridman (4:08:04.580)
and signing up to Masterclass at masterclass.com slash lex.
Lex Fridman (4:08:09.800)
Click the links, buy the stuff.
Ben Goertzel (4:08:12.280)
It's the best way to support this podcast
Lex Fridman (4:08:14.220)
and the journey I'm on in my research and startup.
Ben Goertzel (4:08:18.860)
If you enjoy this thing, subscribe on YouTube,
Lex Fridman (4:08:21.380)
review it with five stars on a podcast,
Ben Goertzel (4:08:23.720)
support it on Patreon or connect with me on Twitter
Lex Fridman (4:08:26.860)
at lexfriedman spelled without the E, just F R I D M A N.
Ben Goertzel (4:08:32.400)
I'm sure eventually you will figure it out.
Lex Fridman (4:08:35.280)
And now let me leave you with some words from Ben Gertzel.
Ben Goertzel (4:08:39.140)
Our language for describing emotions is very crude.
Lex Fridman (4:08:42.540)
That's what music is for.
Ben Goertzel (4:08:43.940)
Thank you for listening and hope to see you next time.
Lex Fridman (50:00.520)
whether what title it should be.
Lex Fridman (50:03.200)
And so we talked about narrow AI, broad AI, wide AI,
Lex Fridman (50:07.560)
narrow AI, general AI.
Lex Fridman (50:09.760)
And I think it was either Shane Legge or Peter Vos
Lex Fridman (50:15.880)
on the private email discussion we had.
Ben Goertzel (50:18.120)
He said, but why don't we go
Lex Fridman (50:18.960)
with AGI, artificial general intelligence?
Lex Fridman (50:21.800)
And Pei Wang wanted to do GAI,
Lex Fridman (50:24.280)
general artificial intelligence,
Ben Goertzel (50:25.760)
because in Chinese it goes in that order.
Lex Fridman (50:27.880)
But we figured gay wouldn't work
Lex Fridman (50:30.200)
in US culture at that time, right?
Lex Fridman (50:33.240)
So we went with the AGI.
Ben Goertzel (50:37.360)
We used it for the title of that book.
Lex Fridman (50:39.520)
And part of Peter and Shane's reasoning
Ben Goertzel (50:43.460)
was you have the G factor in psychology,
Lex Fridman (50:45.460)
which is IQ, general intelligence, right?
Lex Fridman (50:47.480)
So you have a meaning of GI, general intelligence,
Lex Fridman (50:51.160)
in psychology, so then you're looking like artificial GI.
Lex Fridman (50:55.360)
So then we use that for the title of the book.
Lex Fridman (51:00.400)
And so I think maybe both Shane and Peter
Ben Goertzel (51:04.040)
think they invented the term,
Lex Fridman (51:05.200)
but then later after the book was published,
Ben Goertzel (51:08.320)
this guy, Mark Guberd, came up to me and he's like,
Lex Fridman (51:11.640)
well, I published an essay with the term AGI
Ben Goertzel (51:14.800)
in like 1997 or something.
Lex Fridman (51:17.120)
And so I'm just waiting for some Russian to come out
Lex Fridman (51:20.520)
and say they published that in 1953, right?
Lex Fridman (51:23.400)
I mean, that term is not dramatically innovative
Ben Goertzel (51:27.800)
or anything.
Lex Fridman (51:28.640)
It's one of these obvious in hindsight things,
Ben Goertzel (51:31.560)
which is also annoying in a way,
Lex Fridman (51:34.880)
because Joshua Bach, who you interviewed,
Ben Goertzel (51:39.500)
is a close friend of mine.
Lex Fridman (51:40.400)
He likes the term synthetic intelligence,
Ben Goertzel (51:43.240)
which I like much better,
Lex Fridman (51:44.300)
but it hasn't actually caught on, right?
Ben Goertzel (51:47.080)
Because I mean, artificial is a bit off to me
Lex Fridman (51:51.800)
because artifice is like a tool or something,
Lex Fridman (51:54.640)
but not all AGI's are gonna be tools.
Lex Fridman (51:57.760)
I mean, they may be now,
Lex Fridman (51:58.700)
but we're aiming toward making them agents
Lex Fridman (52:00.600)
rather than tools.
Lex Fridman (52:02.800)
And in a way, I don't like the distinction
Lex Fridman (52:04.840)
between artificial and natural,
Ben Goertzel (52:07.200)
because I mean, we're part of nature also
Lex Fridman (52:09.360)
and machines are part of nature.
Ben Goertzel (52:12.160)
I mean, you can look at evolved versus engineered,
Lex Fridman (52:14.840)
but that's a different distinction.
Lex Fridman (52:17.160)
Then it should be engineered general intelligence, right?
Lex Fridman (52:20.000)
And then general, well,
Ben Goertzel (52:21.920)
if you look at Marcus Hooter's book,
Lex Fridman (52:24.600)
universally, what he argues there is,
Ben Goertzel (52:28.240)
within the domain of computation theory,
Lex Fridman (52:30.520)
which is limited, but interesting.
Lex Fridman (52:31.920)
So if you assume computable environments
Lex Fridman (52:33.680)
or computable reward functions,
Ben Goertzel (52:35.600)
then he articulates what would be
Lex Fridman (52:37.560)
a truly general intelligence,
Ben Goertzel (52:40.040)
a system called AIXI, which is quite beautiful.
Lex Fridman (52:43.160)
AIXI, and that's the middle name
Lex Fridman (52:46.280)
of my latest child, actually, is it?
Lex Fridman (52:49.360)
What's the first name?
Ben Goertzel (52:50.200)
First name is QORXI, Q O R X I,
Lex Fridman (52:52.400)
which my wife came up with,
Lex Fridman (52:53.780)
but that's an acronym for quantum organized rational
Lex Fridman (52:57.320)
expanding intelligence, and his middle name is Xiphonies,
Ben Goertzel (53:03.120)
actually, which means the former principal underlying AIXI.
Lex Fridman (53:08.340)
But in any case.
Ben Goertzel (53:09.480)
You're giving Elon Musk's new child a run for his money.
Lex Fridman (53:12.160)
Well, I did it first.
Ben Goertzel (53:13.800)
He copied me with this new freakish name,
Lex Fridman (53:17.320)
but now if I have another baby,
Ben Goertzel (53:18.600)
I'm gonna have to outdo him.
Lex Fridman (53:20.600)
It's becoming an arms race of weird, geeky baby names.
Lex Fridman (53:24.560)
We'll see what the babies think about it, right?
Lex Fridman (53:26.840)
But I mean, my oldest son, Zarathustra, loves his name,
Lex Fridman (53:30.220)
and my daughter, Sharazad, loves her name.
Lex Fridman (53:33.800)
So far, basically, if you give your kids weird names.
Ben Goertzel (53:36.960)
They live up to it.
Lex Fridman (53:37.840)
Well, you're obliged to make the kids weird enough
Lex Fridman (53:39.800)
that they like the names, right?
Lex Fridman (53:42.000)
It directs their upbringing in a certain way.
Lex Fridman (53:43.920)
But yeah, anyway, I mean, what Marcus showed in that book
Lex Fridman (53:47.680)
is that a truly general intelligence
Ben Goertzel (53:50.560)
theoretically is possible,
Lex Fridman (53:51.800)
but would take infinite computing power.
Lex Fridman (53:53.840)
So then the artificial is a little off.
Lex Fridman (53:56.360)
The general is not really achievable within physics
Ben Goertzel (53:59.800)
as we know it.
Lex Fridman (54:01.280)
And I mean, physics as we know it may be limited,
Lex Fridman (54:03.520)
but that's what we have to work with now.
Lex Fridman (54:05.300)
Intelligence.
Ben Goertzel (54:06.140)
Infinitely general, you mean,
Lex Fridman (54:07.360)
like information processing perspective, yeah.
Lex Fridman (54:10.440)
Yeah, intelligence is not very well defined either, right?
Lex Fridman (54:14.760)
I mean, what does it mean?
Ben Goertzel (54:16.760)
I mean, in AI now, it's fashionable to look at it
Lex Fridman (54:19.560)
as maximizing an expected reward over the future.
Lex Fridman (54:23.320)
But that sort of definition is pathological in various ways.
Lex Fridman (54:27.800)
And my friend David Weinbaum, AKA Weaver,
Ben Goertzel (54:31.320)
he had a beautiful PhD thesis on open ended intelligence,
Lex Fridman (54:34.840)
trying to conceive intelligence in a...
Ben Goertzel (54:36.880)
Without a reward.
Lex Fridman (54:38.240)
Yeah, he's just looking at it differently.
Ben Goertzel (54:40.120)
He's looking at complex self organizing systems
Lex Fridman (54:42.680)
and looking at an intelligent system
Ben Goertzel (54:44.640)
as being one that revises and grows
Lex Fridman (54:47.600)
and improves itself in conjunction with its environment
Ben Goertzel (54:51.740)
without necessarily there being one objective function
Lex Fridman (54:54.880)
it's trying to maximize.
Ben Goertzel (54:56.080)
Although over certain intervals of time,
Lex Fridman (54:58.520)
it may act as if it's optimizing
Ben Goertzel (54:59.960)
a certain objective function.
Lex Fridman (55:01.360)
Very much Solaris from Stanislav Lem's novels, right?
Lex Fridman (55:04.580)
So yeah, the point is artificial, general and intelligence.
Lex Fridman (55:07.880)
Don't work.
Ben Goertzel (55:08.720)
They're all bad.
Lex Fridman (55:09.540)
On the other hand, everyone knows what AI is.
Lex Fridman (55:12.040)
And AGI seems immediately comprehensible
Lex Fridman (55:15.880)
to people with a technical background.
Lex Fridman (55:17.520)
So I think that the term has served
Lex Fridman (55:19.360)
as sociological function.
Lex Fridman (55:20.720)
And now it's out there everywhere, which baffles me.
Lex Fridman (55:24.720)
It's like KFC.
Ben Goertzel (55:25.800)
I mean, that's it.
Lex Fridman (55:27.080)
We're stuck with AGI probably for a very long time
Ben Goertzel (55:30.200)
until AGI systems take over and rename themselves.
Lex Fridman (55:33.640)
Yeah.
Lex Fridman (55:34.480)
And then we'll be biological.
Lex Fridman (55:36.160)
We're stuck with GPUs too,
Ben Goertzel (55:37.560)
which mostly have nothing to do with graphics.
Lex Fridman (55:39.320)
Any more, right?
Ben Goertzel (55:40.520)
I wonder what the AGI system will call us humans.
Lex Fridman (55:43.260)
That was maybe.
Ben Goertzel (55:44.280)
Grandpa.
Lex Fridman (55:45.120)
Yeah.
Ben Goertzel (55:45.960)
Yeah.
Lex Fridman (55:46.800)
GPs.
Ben Goertzel (55:47.620)
Yeah.
Lex Fridman (55:48.460)
Grandpa processing unit, yeah.
Ben Goertzel (55:50.320)
Biological grandpa processing units.
Lex Fridman (55:52.120)
Yeah.
Ben Goertzel (55:54.280)
Okay, so maybe also just a comment on AGI representing
Lex Fridman (56:00.580)
before even the term existed,
Ben Goertzel (56:02.160)
representing a kind of community.
Lex Fridman (56:04.640)
You've talked about this in the past,
Ben Goertzel (56:06.240)
sort of AI is coming in waves,
Lex Fridman (56:08.340)
but there's always been this community of people
Ben Goertzel (56:10.440)
who dream about creating general human level
Lex Fridman (56:15.160)
super intelligence systems.
Lex Fridman (56:19.000)
Can you maybe give your sense of the history
Lex Fridman (56:21.880)
of this community as it exists today,
Ben Goertzel (56:24.280)
as it existed before this deep learning revolution
Lex Fridman (56:26.720)
all throughout the winters and the summers of AI?
Ben Goertzel (56:29.520)
Sure.
Lex Fridman (56:30.340)
First, I would say as a side point,
Ben Goertzel (56:33.500)
the winters and summers of AI are greatly exaggerated
Lex Fridman (56:37.840)
by Americans and in that,
Ben Goertzel (56:40.960)
if you look at the publication record
Lex Fridman (56:43.600)
of the artificial intelligence community
Ben Goertzel (56:46.400)
since say the 1950s,
Lex Fridman (56:48.480)
you would find a pretty steady growth
Ben Goertzel (56:51.360)
in advance of ideas and papers.
Lex Fridman (56:53.980)
And what's thought of as an AI winter or summer
Ben Goertzel (56:57.720)
was sort of how much money is the US military
Lex Fridman (57:00.480)
pumping into AI, which was meaningful.
Ben Goertzel (57:04.640)
On the other hand, there was AI going on in Germany,
Lex Fridman (57:06.960)
UK and in Japan and in Russia, all over the place,
Ben Goertzel (57:10.960)
while US military got more and less enthused about AI.
Lex Fridman (57:16.300)
So, I mean.
Ben Goertzel (57:17.560)
That happened to be, just for people who don't know,
Lex Fridman (57:20.200)
the US military happened to be the main source
Ben Goertzel (57:22.840)
of funding for AI research.
Lex Fridman (57:24.500)
So another way to phrase that is it's up and down
Ben Goertzel (57:27.480)
of funding for artificial intelligence research.
Lex Fridman (57:31.080)
And I would say the correlation between funding
Lex Fridman (57:34.600)
and intellectual advance was not 100%, right?
Lex Fridman (57:38.120)
Because I mean, in Russia, as an example, or in Germany,
Ben Goertzel (57:42.120)
there was less dollar funding than in the US,
Lex Fridman (57:44.840)
but many foundational ideas were laid out,
Lex Fridman (57:48.160)
but it was more theory than implementation, right?
Lex Fridman (57:50.880)
And US really excelled at sort of breaking through
Ben Goertzel (57:54.600)
from theoretical papers to working implementations,
Lex Fridman (58:00.200)
which did go up and down somewhat
Ben Goertzel (58:03.020)
with US military funding,
Lex Fridman (58:04.320)
but still, I mean, you can look in the 1980s,
Ben Goertzel (58:07.440)
Dietrich Derner in Germany had self driving cars
Lex Fridman (58:10.400)
on the Autobahn, right?
Lex Fridman (58:11.440)
And I mean, it was a little early
Lex Fridman (58:15.600)
with regard to the car industry,
Lex Fridman (58:16.920)
so it didn't catch on such as has happened now.
Lex Fridman (58:20.200)
But I mean, that whole advancement
Ben Goertzel (58:22.960)
of self driving car technology in Germany
Lex Fridman (58:25.900)
was pretty much independent of AI military summers
Lex Fridman (58:29.720)
and winters in the US.
Lex Fridman (58:31.040)
So there's been more going on in AI globally
Ben Goertzel (58:34.480)
than not only most people on the planet realize,
Lex Fridman (58:37.120)
but then most new AI PhDs realize
Ben Goertzel (58:40.080)
because they've come up within a certain sub field of AI
Lex Fridman (58:44.600)
and haven't had to look so much beyond that.
Lex Fridman (58:47.680)
But I would say when I got my PhD in 1989 in mathematics,
Lex Fridman (58:54.300)
I was interested in AI already.
Ben Goertzel (58:56.000)
In Philadelphia.
Lex Fridman (58:56.840)
Yeah, I started at NYU, then I transferred to Philadelphia
Ben Goertzel (59:00.920)
to Temple University, good old North Philly.
Lex Fridman (59:03.960)
North Philly.
Ben Goertzel (59:04.800)
Yeah, yeah, yeah, the pearl of the US.
Lex Fridman (59:09.280)
You never stopped at a red light then
Ben Goertzel (59:10.920)
because you were afraid if you stopped at a red light,
Lex Fridman (59:12.760)
someone will carjack you.
Lex Fridman (59:13.760)
So you just drive through every red light.
Lex Fridman (59:15.960)
Yeah.
Ben Goertzel (59:18.200)
Every day driving or bicycling to Temple from my house
Lex Fridman (59:20.940)
was like a new adventure.
Lex Fridman (59:24.280)
But yeah, the reason I didn't do a PhD in AI
Lex Fridman (59:27.520)
was what people were doing in the academic AI field then,
Ben Goertzel (59:30.860)
was just astoundingly boring and seemed wrong headed to me.
Lex Fridman (59:34.880)
It was really like rule based expert systems
Lex Fridman (59:38.060)
and production systems.
Lex Fridman (59:39.360)
And actually I loved mathematical logic.
Ben Goertzel (59:42.080)
I had nothing against logic as the cognitive engine for an AI,
Lex Fridman (59:45.840)
but the idea that you could type in the knowledge
Ben Goertzel (59:48.920)
that AI would need to think seemed just completely stupid
Lex Fridman (59:52.720)
and wrong headed to me.
Ben Goertzel (59:55.380)
I mean, you can use logic if you want,
Lex Fridman (59:57.400)
but somehow the system has got to be...
🔗 相关节目