Max Tegmark: Life 3.0
生物与进化AI 与机器学习心理与人性音乐与艺术技术与编程
🤖
AI 智能总结
马克斯·泰格马克谈《生命3.0》与AI存在性风险
这是 Lex Fridman 与 MIT 物理学家、未来生命研究所联合创始人 Max Tegmark 的对话。Tegmark 探讨了他的《生命 3.0》框架、AI 存在性风险、数学宇宙假说，以及如何确保 AI 的未来对人类有益。
生命3.0AI存在性风险数学宇宙AI安全意识宇宙学
Max Tegmark 是 MIT 物理学教授，未来生命研究所（FLI）联合创始人，著有《数学宇宙》和《生命 3.0》，是 AI 安全研究和存在性风险领域最重要的倡导者之一。
📌 核心观点
- 生命 3.0 框架：Tegmark 将生命分为三个阶段——生命 1.0（生物进化决定硬件和软件）、生命 2.0（可以学习但硬件固定）、生命 3.0（可以重新设计自身的硬件和软件）。AI 代表了生命 3.0 的可能性，这是宇宙历史上的重大转折点。
- AI 存在性风险：Tegmark 认为超级智能 AI 是人类面临的最严重存在性风险，不是因为 AI 会「变坏」，而是因为目标对齐问题——一个追求错误目标的超级智能 AI 可能在实现目标的过程中消灭人类。
- 数学宇宙假说：Tegmark 提出宇宙本质上是一个数学结构，物理现实与数学现实是同一的。这意味着意识可能是任何足够复杂的信息处理系统的属性，不限于碳基生命。
- AI 安全研究的紧迫性：Tegmark 是 AI 安全研究的重要倡导者，他认为我们需要在 AI 变得足够强大之前解决对齐问题，这是人类历史上最重要的技术挑战。
- 对 AI 乐观主义的平衡：Tegmark 不是末日论者，他认为如果我们做对了，AI 可以带来人类历史上最美好的未来——消除疾病、贫困，实现科学的飞跃。关键在于我们现在做出的选择。
✨ 金句摘录
Tegmark：我们正处于宇宙历史上最重要的时刻——我们正在创造生命 3.0，一种可以重新设计自身的生命形式。
Tegmark：AI 安全不是关于防止机器人叛乱，而是关于确保 AI 追求我们真正想要的目标。
Tegmark：如果我们做对了，AI 可以带来人类历史上最美好的未来；如果我们做错了，后果可能是灾难性的。
📋 章节目录
暂无章节信息
🔑 关键词
donintelligencegoinghumanbetterintelligentuniversegoalsagiconsciousnessfuturemachinesableprocessingconsciousdoingcourseputimportantexperience
💬 精彩语录
暂无语录
🎙️ 完整对话（1780 条）
Lex Fridman (00:00.000)
As part of MIT course 6S099, Artificial General Intelligence,
作为麻省理工学院课程 6S099“通用人工智能”的一部分，
Lex Fridman (00:04.200)
I've gotten the chance to sit down with Max Tegmark.
我有机会与马克斯·泰格马克坐下来交谈。
Lex Fridman (00:06.600)
He is a professor here at MIT.
他是麻省理工学院的教授。
Lex Fridman (00:08.680)
He's a physicist, spent a large part of his career
他是一位物理学家，度过了他职业生涯的大部分时间
Lex Fridman (00:11.920)
studying the mysteries of our cosmological universe.
研究我们宇宙的奥秘。
Lex Fridman (00:16.960)
But he's also studied and delved into the beneficial
但他也研究并深入研究了有益的
Lex Fridman (00:20.680)
possibilities and the existential risks
可能性和存在风险
Max Tegmark (00:24.000)
of artificial intelligence.
人工智能的。
Lex Fridman (00:25.800)
Amongst many other things, he is the cofounder
除其他许多事情外，他是联合创始人
Max Tegmark (00:29.040)
of the Future of Life Institute, author of two books,
生命未来研究所的教授，两本书的作者，
Lex Fridman (00:33.080)
both of which I highly recommend.
我强烈推荐这两个。
Max Tegmark (00:35.160)
First, Our Mathematical Universe.
首先，我们的数学宇宙。
Lex Fridman (00:37.260)
Second is Life 3.0.
其次是生活3.0。
Max Tegmark (00:40.160)
He's truly an out of the box thinker and a fun personality,
他确实是一位打破常规的思想家和有趣的个性，
Lex Fridman (00:44.080)
so I really enjoy talking to him.
所以我真的很喜欢和他说话。
Max Tegmark (00:45.480)
If you'd like to see more of these videos in the future,
如果您以后想观看更多此类视频，
Lex Fridman (00:47.980)
please subscribe and also click the little bell icon
请订阅并点击小铃铛图标
Max Tegmark (00:50.640)
to make sure you don't miss any videos.
以确保您不会错过任何视频。
Lex Fridman (00:52.720)
Also, Twitter, LinkedIn, agi.mit.edu
另外，Twitter、LinkedIn、agi.mit.edu
Max Tegmark (00:56.840)
if you wanna watch other lectures
如果你想看其他讲座
Lex Fridman (00:59.600)
or conversations like this one.
Max Tegmark (01:01.080)
Better yet, go read Max's book, Life 3.0.
Lex Fridman (01:04.000)
Chapter seven on goals is my favorite.
Max Tegmark (01:07.940)
It's really where philosophy and engineering come together
Lex Fridman (01:10.480)
and it opens with a quote by Dostoevsky.
Max Tegmark (01:14.400)
The mystery of human existence lies not in just staying alive
Lex Fridman (01:17.940)
but in finding something to live for.
Max Tegmark (01:20.520)
Lastly, I believe that every failure rewards us
Lex Fridman (01:23.920)
with an opportunity to learn
Lex Fridman (01:26.560)
and in that sense, I've been very fortunate
Lex Fridman (01:28.360)
to fail in so many new and exciting ways
Lex Fridman (01:31.840)
and this conversation was no different.
Lex Fridman (01:34.020)
I've learned about something called
Max Tegmark (01:36.160)
radio frequency interference, RFI, look it up.
Lex Fridman (01:40.840)
Apparently, music and conversations
Max Tegmark (01:42.960)
from local radio stations can bleed into the audio
Lex Fridman (01:45.480)
that you're recording in such a way
Max Tegmark (01:47.080)
that it almost completely ruins that audio.
Lex Fridman (01:49.360)
It's an exceptionally difficult sound source to remove.
Max Tegmark (01:53.240)
So, I've gotten the opportunity to learn
Lex Fridman (01:55.520)
how to avoid RFI in the future during recording sessions.
Max Tegmark (02:00.200)
I've also gotten the opportunity to learn
Lex Fridman (02:02.680)
how to use Adobe Audition and iZotope RX 6
Max Tegmark (02:06.240)
to do some noise, some audio repair.
Lex Fridman (02:11.720)
Of course, this is an exceptionally difficult noise
Max Tegmark (02:14.380)
to remove.
Lex Fridman (02:15.220)
I am an engineer.
Max Tegmark (02:16.280)
I'm not an audio engineer.
Lex Fridman (02:18.240)
Neither is anybody else in our group
Lex Fridman (02:20.180)
but we did our best.
Lex Fridman (02:21.880)
Nevertheless, I thank you for your patience
Lex Fridman (02:25.040)
and I hope you're still able to enjoy this conversation.
Lex Fridman (02:27.960)
Do you think there's intelligent life
Lex Fridman (02:29.320)
out there in the universe?
Lex Fridman (02:31.360)
Let's open up with an easy question.
Max Tegmark (02:33.480)
I have a minority view here actually.
Lex Fridman (02:36.240)
When I give public lectures, I often ask for a show of hands
Max Tegmark (02:39.440)
who thinks there's intelligent life out there somewhere else
Lex Fridman (02:42.920)
and almost everyone put their hands up
Lex Fridman (02:45.440)
and when I ask why, they'll be like,
Lex Fridman (02:47.360)
oh, there's so many galaxies out there, there's gotta be.
Lex Fridman (02:51.840)
But I'm a numbers nerd, right?
Lex Fridman (02:54.560)
So when you look more carefully at it,
Max Tegmark (02:56.640)
it's not so clear at all.
Lex Fridman (02:59.080)
When we talk about our universe, first of all,
Max Tegmark (03:00.680)
we don't mean all of space.
Lex Fridman (03:03.040)
We actually mean, I don't know,
Max Tegmark (03:04.040)
you can throw me the universe if you want,
Lex Fridman (03:05.440)
it's behind you there.
Max Tegmark (03:07.280)
It's, we simply mean the spherical region of space
Lex Fridman (03:11.440)
from which light has a time to reach us so far
Max Tegmark (03:15.360)
during the 14.8 billion year,
Lex Fridman (03:17.040)
13.8 billion years since our Big Bang.
Max Tegmark (03:19.320)
There's more space here but this is what we call a universe
Lex Fridman (03:22.320)
because that's all we have access to.
Lex Fridman (03:24.040)
So is there intelligent life here
Lex Fridman (03:25.960)
that's gotten to the point of building telescopes
Lex Fridman (03:28.920)
and computers?
Lex Fridman (03:31.160)
My guess is no, actually.
Max Tegmark (03:34.540)
The probability of it happening on any given planet
Lex Fridman (03:39.240)
is some number we don't know what it is.
Lex Fridman (03:42.620)
And what we do know is that the number can't be super high
Lex Fridman (03:48.480)
because there's over a billion Earth like planets
Max Tegmark (03:50.300)
in the Milky Way galaxy alone,
Lex Fridman (03:52.880)
many of which are billions of years older than Earth.
Lex Fridman (03:56.280)
And aside from some UFO believers,
Lex Fridman (04:00.600)
there isn't much evidence
Max Tegmark (04:01.880)
that any superduran civilization has come here at all.
Lex Fridman (04:05.600)
And so that's the famous Fermi paradox, right?
Lex Fridman (04:08.440)
And then if you work the numbers,
Lex Fridman (04:10.180)
what you find is that if you have no clue
Lex Fridman (04:13.440)
what the probability is of getting life on a given planet,
Lex Fridman (04:16.880)
so it could be 10 to the minus 10, 10 to the minus 20,
Max Tegmark (04:19.680)
or 10 to the minus two, or any power of 10
Lex Fridman (04:22.960)
is sort of equally likely
Max Tegmark (04:23.800)
if you wanna be really open minded,
Lex Fridman (04:25.480)
that translates into it being equally likely
Max Tegmark (04:27.600)
that our nearest neighbor is 10 to the 16 meters away,
Lex Fridman (04:31.800)
10 to the 17 meters away, 10 to the 18.
Max Tegmark (04:35.400)
By the time you get much less than 10 to the 16 already,
Lex Fridman (04:41.080)
we pretty much know there is nothing else that close.
Lex Fridman (04:45.960)
And when you get beyond 10.
Lex Fridman (04:47.280)
Because they would have discovered us.
Max Tegmark (04:48.680)
Yeah, they would have been discovered as long ago,
Lex Fridman (04:50.360)
or if they're really close,
Max Tegmark (04:51.440)
we would have probably noted some engineering projects
Lex Fridman (04:53.560)
that they're doing.
Lex Fridman (04:54.640)
And if it's beyond 10 to the 26 meters,
Lex Fridman (04:57.880)
that's already outside of here.
Lex Fridman (05:00.000)
So my guess is actually that we are the only life in here
Lex Fridman (05:05.800)
that's gotten the point of building advanced tech,
Max Tegmark (05:09.040)
which I think is very,
Lex Fridman (05:12.680)
puts a lot of responsibility on our shoulders, not screw up.
Max Tegmark (05:15.360)
I think people who take for granted
Lex Fridman (05:17.240)
that it's okay for us to screw up,
Max Tegmark (05:20.120)
have an accidental nuclear war or go extinct somehow
Lex Fridman (05:22.760)
because there's a sort of Star Trek like situation out there
Max Tegmark (05:25.960)
where some other life forms are gonna come and bail us out
Lex Fridman (05:28.360)
and it doesn't matter as much.
Max Tegmark (05:30.400)
I think they're leveling us into a false sense of security.
Lex Fridman (05:33.400)
I think it's much more prudent to say,
Max Tegmark (05:35.200)
let's be really grateful
Lex Fridman (05:36.400)
for this amazing opportunity we've had
Lex Fridman (05:38.720)
and make the best of it just in case it is down to us.
Lex Fridman (05:44.080)
So from a physics perspective,
Lex Fridman (05:45.680)
do you think intelligent life,
Lex Fridman (05:48.800)
so it's unique from a sort of statistical view
Max Tegmark (05:51.360)
of the size of the universe,
Lex Fridman (05:52.560)
but from the basic matter of the universe,
Lex Fridman (05:55.840)
how difficult is it for intelligent life to come about?
Lex Fridman (05:59.040)
The kind of advanced tech building life
Max Tegmark (06:03.120)
is implied in your statement that it's really difficult
Lex Fridman (06:05.720)
to create something like a human species.
Max Tegmark (06:07.640)
Well, I think what we know is that going from no life
Lex Fridman (06:11.560)
to having life that can do a level of tech,
Max Tegmark (06:15.720)
there's some sort of two going beyond that
Lex Fridman (06:18.720)
than actually settling our whole universe with life.
Max Tegmark (06:22.200)
There's some major roadblock there,
Lex Fridman (06:26.560)
which is some great filter as it's sometimes called,
Max Tegmark (06:30.880)
which is tough to get through.
Lex Fridman (06:33.520)
It's either that roadblock is either behind us
Max Tegmark (06:37.160)
or in front of us.
Lex Fridman (06:38.720)
I'm hoping very much that it's behind us.
Max Tegmark (06:41.080)
I'm super excited every time we get a new report from NASA
Lex Fridman (06:45.960)
saying they failed to find any life on Mars.
Max Tegmark (06:48.480)
I'm like, yes, awesome.
Lex Fridman (06:50.080)
Because that suggests that the hard part,
Max Tegmark (06:51.680)
maybe it was getting the first ribosome
Lex Fridman (06:54.240)
or some very low level kind of stepping stone
Lex Fridman (06:59.520)
so that we're home free.
Lex Fridman (07:00.400)
Because if that's true,
Max Tegmark (07:01.720)
then the future is really only limited
Lex Fridman (07:03.640)
by our own imagination.
Max Tegmark (07:05.200)
It would be much suckier if it turns out
Lex Fridman (07:07.360)
that this level of life is kind of a dime a dozen,
Lex Fridman (07:11.440)
but maybe there's some other problem.
Lex Fridman (07:12.760)
Like as soon as a civilization gets advanced technology,
Max Tegmark (07:16.160)
within a hundred years,
Lex Fridman (07:17.000)
they get into some stupid fight with themselves and poof.
Max Tegmark (07:20.320)
That would be a bummer.
Lex Fridman (07:21.760)
Yeah, so you've explored the mysteries of the universe,
Max Tegmark (07:26.160)
the cosmological universe, the one that's sitting
Lex Fridman (07:29.000)
between us today.
Max Tegmark (07:31.080)
I think you've also begun to explore the other universe,
Lex Fridman (07:35.960)
which is sort of the mystery,
Max Tegmark (07:38.000)
the mysterious universe of the mind of intelligence,
Lex Fridman (07:40.960)
of intelligent life.
Lex Fridman (07:42.840)
So is there a common thread between your interest
Lex Fridman (07:45.280)
or the way you think about space and intelligence?
Max Tegmark (07:48.760)
Oh yeah, when I was a teenager,
Lex Fridman (07:53.040)
I was already very fascinated by the biggest questions.
Lex Fridman (07:57.280)
And I felt that the two biggest mysteries of all in science
Lex Fridman (08:00.560)
were our universe out there and our universe in here.
Lex Fridman (08:05.000)
So it's quite natural after having spent
Lex Fridman (08:08.120)
a quarter of a century on my career,
Max Tegmark (08:11.040)
thinking a lot about this one,
Lex Fridman (08:12.680)
that I'm now indulging in the luxury
Max Tegmark (08:14.320)
of doing research on this one.
Lex Fridman (08:15.960)
It's just so cool.
Max Tegmark (08:17.720)
I feel the time is ripe now
Lex Fridman (08:20.120)
for you trans greatly deepening our understanding of this.
Max Tegmark (08:25.120)
Just start exploring this one.
Lex Fridman (08:26.640)
Yeah, because I think a lot of people view intelligence
Max Tegmark (08:29.560)
as something mysterious that can only exist
Lex Fridman (08:33.520)
in biological organisms like us,
Lex Fridman (08:36.120)
and therefore dismiss all talk
Lex Fridman (08:37.680)
about artificial general intelligence as science fiction.
Lex Fridman (08:41.160)
But from my perspective as a physicist,
Lex Fridman (08:43.200)
I am a blob of quarks and electrons
Max Tegmark (08:46.680)
moving around in a certain pattern
Lex Fridman (08:48.360)
and processing information in certain ways.
Lex Fridman (08:50.080)
And this is also a blob of quarks and electrons.
Lex Fridman (08:53.600)
I'm not smarter than the water bottle
Max Tegmark (08:55.360)
because I'm made of different kinds of quarks.
Lex Fridman (08:57.880)
I'm made of up quarks and down quarks,
Max Tegmark (08:59.640)
exact same kind as this.
Lex Fridman (09:01.400)
There's no secret sauce, I think, in me.
Max Tegmark (09:05.080)
It's all about the pattern of the information processing.
Lex Fridman (09:08.560)
And this means that there's no law of physics
Max Tegmark (09:12.240)
saying that we can't create technology,
Lex Fridman (09:15.600)
which can help us by being incredibly intelligent
Lex Fridman (09:19.960)
and help us crack mysteries that we couldn't.
Lex Fridman (09:21.680)
In other words, I think we've really only seen
Max Tegmark (09:23.560)
the tip of the intelligence iceberg so far.
Lex Fridman (09:26.480)
Yeah, so the perceptronium.
Max Tegmark (09:29.960)
Yeah.
Lex Fridman (09:31.280)
So you coined this amazing term.
Max Tegmark (09:33.200)
It's a hypothetical state of matter,
Lex Fridman (09:35.760)
sort of thinking from a physics perspective,
Lex Fridman (09:38.360)
what is the kind of matter that can help,
Lex Fridman (09:40.080)
as you're saying, subjective experience emerge,
Max Tegmark (09:42.920)
consciousness emerge.
Lex Fridman (09:44.280)
So how do you think about consciousness
Lex Fridman (09:46.640)
from this physics perspective?
Lex Fridman (09:49.960)
Very good question.
Lex Fridman (09:50.800)
So again, I think many people have underestimated
Lex Fridman (09:55.800)
our ability to make progress on this
Max Tegmark (09:59.120)
by convincing themselves it's hopeless
Lex Fridman (10:01.320)
because somehow we're missing some ingredient that we need.
Max Tegmark (10:05.840)
There's some new consciousness particle or whatever.
Lex Fridman (10:09.560)
I happen to think that we're not missing anything
Lex Fridman (10:12.720)
and that it's not the interesting thing
Lex Fridman (10:16.320)
about consciousness that gives us
Max Tegmark (10:18.560)
this amazing subjective experience of colors
Lex Fridman (10:21.400)
and sounds and emotions.
Max Tegmark (10:23.320)
It's rather something at the higher level
Lex Fridman (10:26.320)
about the patterns of information processing.
Lex Fridman (10:28.800)
And that's why I like to think about this idea
Lex Fridman (10:33.160)
of perceptronium.
Lex Fridman (10:34.480)
What does it mean for an arbitrary physical system
Lex Fridman (10:36.920)
to be conscious in terms of what its particles are doing
Lex Fridman (10:41.920)
or its information is doing?
Lex Fridman (10:43.560)
I don't think, I hate carbon chauvinism,
Max Tegmark (10:46.080)
this attitude you have to be made of carbon atoms
Lex Fridman (10:47.960)
to be smart or conscious.
Max Tegmark (10:50.160)
There's something about the information processing
Lex Fridman (10:53.520)
that this kind of matter performs.
Max Tegmark (10:55.360)
Yeah, and you can see I have my favorite equations here
Lex Fridman (10:57.840)
describing various fundamental aspects of the world.
Max Tegmark (11:00.720)
I feel that I think one day,
Lex Fridman (11:02.560)
maybe someone who's watching this will come up
Max Tegmark (11:04.360)
with the equations that information processing
Lex Fridman (11:07.280)
has to satisfy to be conscious.
Max Tegmark (11:08.760)
I'm quite convinced there is big discovery
Lex Fridman (11:11.800)
to be made there because let's face it,
Max Tegmark (11:15.400)
we know that so many things are made up of information.
Lex Fridman (11:18.720)
We know that some information processing is conscious
Max Tegmark (11:21.960)
because we are conscious.
Lex Fridman (11:25.520)
But we also know that a lot of information processing
Max Tegmark (11:27.600)
is not conscious.
Lex Fridman (11:28.440)
Like most of the information processing happening
Max Tegmark (11:30.040)
in your brain right now is not conscious.
Lex Fridman (11:32.680)
There are like 10 megabytes per second coming in
Max Tegmark (11:36.040)
even just through your visual system.
Lex Fridman (11:38.080)
You're not conscious about your heartbeat regulation
Max Tegmark (11:40.480)
or most things.
Lex Fridman (11:42.120)
Even if I just ask you to like read what it says here,
Max Tegmark (11:45.680)
you look at it and then, oh, now you know what it said.
Lex Fridman (11:48.040)
But you're not aware of how the computation actually happened.
Max Tegmark (11:51.560)
Your consciousness is like the CEO
Lex Fridman (11:53.680)
that got an email at the end with the final answer.
Lex Fridman (11:56.680)
So what is it that makes a difference?
Lex Fridman (12:01.000)
I think that's both a great science mystery.
Max Tegmark (12:05.120)
We're actually studying it a little bit in my lab here
Lex Fridman (12:07.080)
at MIT, but I also think it's just a really urgent question
Max Tegmark (12:10.920)
to answer.
Lex Fridman (12:12.080)
For starters, I mean, if you're an emergency room doctor
Lex Fridman (12:14.880)
and you have an unresponsive patient coming in,
Lex Fridman (12:17.160)
wouldn't it be great if in addition to having
Max Tegmark (12:22.360)
a CT scanner, you had a consciousness scanner
Lex Fridman (12:25.320)
that could figure out whether this person
Max Tegmark (12:27.920)
is actually having locked in syndrome
Lex Fridman (12:30.960)
or is actually comatose.
Lex Fridman (12:33.360)
And in the future, imagine if we build robots
Lex Fridman (12:37.000)
or the machine that we can have really good conversations
Max Tegmark (12:41.480)
with, which I think is very likely to happen.
Lex Fridman (12:44.840)
Wouldn't you want to know if your home helper robot
Max Tegmark (12:47.760)
is actually experiencing anything or just like a zombie,
Lex Fridman (12:51.320)
I mean, would you prefer it?
Lex Fridman (12:53.520)
What would you prefer?
Lex Fridman (12:54.360)
Would you prefer that it's actually unconscious
Lex Fridman (12:56.200)
so that you don't have to feel guilty about switching it off
Lex Fridman (12:58.560)
or giving boring chores or what would you prefer?
Max Tegmark (13:02.120)
Well, certainly we would prefer,
Lex Fridman (13:06.520)
I would prefer the appearance of consciousness.
Lex Fridman (13:08.960)
But the question is whether the appearance of consciousness
Lex Fridman (13:11.720)
is different than consciousness itself.
Lex Fridman (13:15.040)
And sort of to ask that as a question,
Lex Fridman (13:18.200)
do you think we need to understand what consciousness is,
Max Tegmark (13:21.760)
solve the hard problem of consciousness
Lex Fridman (13:23.520)
in order to build something like an AGI system?
Max Tegmark (13:28.240)
No, I don't think that.
Lex Fridman (13:30.440)
And I think we will probably be able to build things
Max Tegmark (13:34.520)
even if we don't answer that question.
Lex Fridman (13:36.080)
But if we want to make sure that what happens
Max Tegmark (13:37.720)
is a good thing, we better solve it first.
Lex Fridman (13:40.960)
So it's a wonderful controversy you're raising there
Max Tegmark (13:44.960)
where you have basically three points of view
Lex Fridman (13:47.960)
about the hard problem.
Lex Fridman (13:48.800)
So there are two different points of view.
Lex Fridman (13:52.800)
They both conclude that the hard problem of consciousness
Max Tegmark (13:55.160)
is BS.
Lex Fridman (13:56.840)
On one hand, you have some people like Daniel Dennett
Max Tegmark (13:59.320)
who say that consciousness is just BS
Lex Fridman (14:01.480)
because consciousness is the same thing as intelligence.
Max Tegmark (14:05.000)
There's no difference.
Lex Fridman (14:06.440)
So anything which acts conscious is conscious,
Max Tegmark (14:11.080)
just like we are.
Lex Fridman (14:13.480)
And then there are also a lot of people,
Max Tegmark (14:15.960)
including many top AI researchers I know,
Lex Fridman (14:18.400)
who say, oh, consciousness is just bullshit
Max Tegmark (14:19.920)
because, of course, machines can never be conscious.
Lex Fridman (14:22.760)
They're always going to be zombies.
Max Tegmark (14:24.520)
You never have to feel guilty about how you treat them.
Lex Fridman (14:27.880)
And then there's a third group of people,
Max Tegmark (14:30.880)
including Giulio Tononi, for example,
Lex Fridman (14:34.920)
and Krzysztof Koch and a number of others.
Max Tegmark (14:37.440)
I would put myself also in this middle camp
Lex Fridman (14:39.520)
who say that actually some information processing
Max Tegmark (14:41.880)
is conscious and some is not.
Lex Fridman (14:44.160)
So let's find the equation which can be used
Max Tegmark (14:46.960)
to determine which it is.
Lex Fridman (14:49.080)
And I think we've just been a little bit lazy,
Max Tegmark (14:52.040)
kind of running away from this problem for a long time.
Lex Fridman (14:54.960)
It's been almost taboo to even mention the C word
Max Tegmark (14:57.840)
in a lot of circles because,
Lex Fridman (15:00.520)
but we should stop making excuses.
Max Tegmark (15:03.520)
This is a science question and there are ways
Lex Fridman (15:07.920)
we can even test any theory that makes predictions for this.
Lex Fridman (15:11.960)
And coming back to this helper robot,
Lex Fridman (15:13.640)
I mean, so you said you'd want your helper robot
Max Tegmark (15:16.080)
to certainly act conscious and treat you,
Lex Fridman (15:18.160)
like have conversations with you and stuff.
Max Tegmark (15:20.880)
I think so.
Lex Fridman (15:21.720)
But wouldn't you, would you feel,
Max Tegmark (15:22.560)
would you feel a little bit creeped out
Lex Fridman (15:23.920)
if you realized that it was just a glossed up tape recorder,
Lex Fridman (15:27.680)
you know, that was just zombie and was a faking emotion?
Lex Fridman (15:31.560)
Would you prefer that it actually had an experience
Max Tegmark (15:34.560)
or would you prefer that it's actually
Lex Fridman (15:37.000)
not experiencing anything so you feel,
Lex Fridman (15:39.120)
you don't have to feel guilty about what you do to it?
Lex Fridman (15:42.200)
It's such a difficult question because, you know,
Max Tegmark (15:45.040)
it's like when you're in a relationship and you say,
Lex Fridman (15:47.280)
well, I love you.
Lex Fridman (15:48.120)
And the other person said, I love you back.
Lex Fridman (15:49.760)
It's like asking, well, do they really love you back
Lex Fridman (15:52.640)
or are they just saying they love you back?
Lex Fridman (15:55.360)
Don't you really want them to actually love you?
Max Tegmark (15:58.120)
It's hard to, it's hard to really know the difference
Lex Fridman (16:03.520)
between everything seeming like there's consciousness
Max Tegmark (16:09.000)
present, there's intelligence present,
Lex Fridman (16:10.640)
there's affection, passion, love,
Lex Fridman (16:13.840)
and it actually being there.
Lex Fridman (16:16.200)
I'm not sure, do you have?
Lex Fridman (16:17.720)
But like, can I ask you a question about this?
Lex Fridman (16:19.400)
Like to make it a bit more pointed.
Lex Fridman (16:20.760)
So Mass General Hospital is right across the river, right?
Lex Fridman (16:22.920)
Yes.
Max Tegmark (16:23.760)
Suppose you're going in for a medical procedure
Lex Fridman (16:26.720)
and they're like, you know, for anesthesia,
Lex Fridman (16:29.320)
what we're going to do is we're going to give you
Lex Fridman (16:31.000)
muscle relaxants so you won't be able to move
Lex Fridman (16:33.160)
and you're going to feel excruciating pain
Lex Fridman (16:35.040)
during the whole surgery,
Lex Fridman (16:35.880)
but you won't be able to do anything about it.
Lex Fridman (16:37.600)
But then we're going to give you this drug
Max Tegmark (16:39.200)
that erases your memory of it.
Lex Fridman (16:41.960)
Would you be cool about that?
Max Tegmark (16:44.960)
What's the difference that you're conscious about it
Lex Fridman (16:48.600)
or not if there's no behavioral change, right?
Max Tegmark (16:51.640)
Right, that's a really, that's a really clear way to put it.
Lex Fridman (16:54.520)
That's, yeah, it feels like in that sense,
Max Tegmark (16:57.400)
experiencing it is a valuable quality.
Lex Fridman (17:01.080)
So actually being able to have subjective experiences,
Max Tegmark (17:05.840)
at least in that case, is valuable.
Lex Fridman (17:09.120)
And I think we humans have a little bit
Max Tegmark (17:11.240)
of a bad track record also of making
Lex Fridman (17:13.600)
these self serving arguments
Max Tegmark (17:15.480)
that other entities aren't conscious.
Lex Fridman (17:18.040)
You know, people often say,
Max Tegmark (17:19.160)
oh, these animals can't feel pain.
Lex Fridman (17:21.800)
It's okay to boil lobsters because we ask them
Max Tegmark (17:24.040)
if it hurt and they didn't say anything.
Lex Fridman (17:25.960)
And now there was just a paper out saying,
Max Tegmark (17:27.400)
lobsters do feel pain when you boil them
Lex Fridman (17:29.320)
and they're banning it in Switzerland.
Lex Fridman (17:31.040)
And we did this with slaves too often and said,
Lex Fridman (17:33.560)
oh, they don't mind.
Max Tegmark (17:36.240)
They don't maybe aren't conscious
Lex Fridman (17:39.480)
or women don't have souls or whatever.
Lex Fridman (17:41.160)
So I'm a little bit nervous when I hear people
Lex Fridman (17:43.200)
just take as an axiom that machines
Max Tegmark (17:46.360)
can't have experience ever.
Lex Fridman (17:48.960)
I think this is just a really fascinating science question
Max Tegmark (17:51.560)
is what it is.
Lex Fridman (17:52.400)
Let's research it and try to figure out
Lex Fridman (17:54.720)
what it is that makes the difference
Lex Fridman (17:56.000)
between unconscious intelligent behavior
Lex Fridman (17:58.880)
and conscious intelligent behavior.
Lex Fridman (18:01.120)
So in terms of, so if you think of a Boston Dynamics
Max Tegmark (18:04.680)
human or robot being sort of with a broom
Lex Fridman (18:07.680)
being pushed around, it starts pushing
Max Tegmark (18:11.920)
on a consciousness question.
Lex Fridman (18:13.320)
So let me ask, do you think an AGI system
Max Tegmark (18:17.040)
like a few neuroscientists believe
Lex Fridman (18:19.720)
needs to have a physical embodiment?
Lex Fridman (18:22.320)
Needs to have a body or something like a body?
Lex Fridman (18:25.720)
No, I don't think so.
Lex Fridman (18:28.280)
You mean to have a conscious experience?
Lex Fridman (18:30.560)
To have consciousness.
Max Tegmark (18:33.160)
I do think it helps a lot to have a physical embodiment
Lex Fridman (18:36.080)
to learn the kind of things about the world
Max Tegmark (18:38.440)
that are important to us humans, for sure.
Lex Fridman (18:42.560)
But I don't think the physical embodiment
Max Tegmark (18:45.600)
is necessary after you've learned it
Lex Fridman (18:47.120)
to just have the experience.
Lex Fridman (18:48.760)
Think about when you're dreaming, right?
Lex Fridman (18:51.400)
Your eyes are closed.
Max Tegmark (18:52.600)
You're not getting any sensory input.
Lex Fridman (18:54.240)
You're not behaving or moving in any way
Lex Fridman (18:55.960)
but there's still an experience there, right?
Lex Fridman (18:59.720)
And so clearly the experience that you have
Max Tegmark (19:01.400)
when you see something cool in your dreams
Lex Fridman (19:03.320)
isn't coming from your eyes.
Max Tegmark (19:04.800)
It's just the information processing itself in your brain
Lex Fridman (19:08.640)
which is that experience, right?
Lex Fridman (19:10.920)
But if I put it another way, I'll say
Lex Fridman (19:13.640)
because it comes from neuroscience
Max Tegmark (19:15.120)
is the reason you want to have a body and a physical
Lex Fridman (19:18.280)
something like a physical, you know, a physical system
Max Tegmark (19:23.920)
is because you want to be able to preserve something.
Lex Fridman (19:27.040)
In order to have a self, you could argue,
Max Tegmark (19:30.840)
would you need to have some kind of embodiment of self
Lex Fridman (19:36.400)
to want to preserve?
Max Tegmark (19:38.920)
Well, now we're getting a little bit anthropomorphic
Lex Fridman (19:42.400)
into anthropomorphizing things.
Max Tegmark (19:45.200)
Maybe talking about self preservation instincts.
Lex Fridman (19:47.280)
I mean, we are evolved organisms, right?
Lex Fridman (19:50.560)
So Darwinian evolution endowed us
Lex Fridman (19:53.520)
and other evolved organism with a self preservation instinct
Max Tegmark (19:57.120)
because those that didn't have those self preservation genes
Lex Fridman (1:00:01.560)
We can now, I think, quite convincingly answer
Max Tegmark (1:00:03.360)
that question of no, it's enough to have just one kind.
Lex Fridman (1:00:07.640)
If you look under the hood of AlphaZero,
Max Tegmark (1:00:09.920)
there's only one kind of neuron
Lex Fridman (1:00:11.080)
and it's ridiculously simple mathematical thing.
Lex Fridman (1:00:15.000)
So it's just like in physics,
Lex Fridman (1:00:17.280)
it's not, if you have a gas with waves in it,
Max Tegmark (1:00:20.320)
it's not the detailed nature of the molecule that matter,
Lex Fridman (1:00:24.240)
it's the collective behavior somehow.
Max Tegmark (1:00:26.040)
Similarly, it's this higher level structure
Lex Fridman (1:00:30.720)
of the network that matters,
Max Tegmark (1:00:31.760)
not that you have 20 kinds of neurons.
Lex Fridman (1:00:34.080)
I think our brain is such a complicated mess
Max Tegmark (1:00:37.040)
because it wasn't evolved just to be intelligent,
Lex Fridman (1:00:41.720)
it was involved to also be self assembling
Lex Fridman (1:00:47.000)
and self repairing, right?
Lex Fridman (1:00:48.760)
And evolutionarily attainable.
Lex Fridman (1:00:51.920)
And so on and so on.
Lex Fridman (1:00:53.560)
So I think it's pretty,
Max Tegmark (1:00:54.720)
my hunch is that we're going to understand
Lex Fridman (1:00:57.040)
how to build AGI before we fully understand
Lex Fridman (1:00:59.520)
how our brains work, just like we understood
Lex Fridman (1:01:02.600)
how to build flying machines long before
Max Tegmark (1:01:05.560)
we were able to build a mechanical bird.
Lex Fridman (1:01:07.800)
Yeah, that's right.
Max Tegmark (1:01:08.640)
You've given the example exactly of mechanical birds
Lex Fridman (1:01:13.280)
and airplanes and airplanes do a pretty good job
Max Tegmark (1:01:15.680)
of flying without really mimicking bird flight.
Lex Fridman (1:01:18.560)
And even now after 100 years later,
Lex Fridman (1:01:20.920)
did you see the Ted talk with this German mechanical bird?
Lex Fridman (1:01:23.880)
I heard you mention it.
Max Tegmark (1:01:25.040)
Check it out, it's amazing.
Lex Fridman (1:01:26.520)
But even after that, right,
Max Tegmark (1:01:27.760)
we still don't fly in mechanical birds
Lex Fridman (1:01:29.360)
because it turned out the way we came up with was simpler
Lex Fridman (1:01:32.720)
and it's better for our purposes.
Lex Fridman (1:01:33.840)
And I think it might be the same there.
Max Tegmark (1:01:35.280)
That's one lesson.
Lex Fridman (1:01:37.520)
And another lesson, it's more what our paper was about.
Max Tegmark (1:01:42.640)
First, as a physicist thought it was fascinating
Lex Fridman (1:01:45.800)
how there's a very close mathematical relationship
Max Tegmark (1:01:48.240)
actually between our artificial neural networks
Lex Fridman (1:01:50.800)
and a lot of things that we've studied for in physics
Max Tegmark (1:01:54.560)
go by nerdy names like the renormalization group equation
Lex Fridman (1:01:57.520)
and Hamiltonians and yada, yada, yada.
Lex Fridman (1:01:59.800)
And when you look a little more closely at this,
Lex Fridman (1:02:05.720)
you have,
Max Tegmark (1:02:10.320)
at first I was like, well, there's something crazy here
Lex Fridman (1:02:12.360)
that doesn't make sense.
Max Tegmark (1:02:13.520)
Because we know that if you even want to build
Lex Fridman (1:02:19.200)
a super simple neural network to tell apart cat pictures
Lex Fridman (1:02:22.560)
and dog pictures, right,
Lex Fridman (1:02:23.400)
that you can do that very, very well now.
Lex Fridman (1:02:25.400)
But if you think about it a little bit,
Lex Fridman (1:02:27.520)
you convince yourself it must be impossible
Max Tegmark (1:02:29.080)
because if I have one megapixel,
Lex Fridman (1:02:31.920)
even if each pixel is just black or white,
Max Tegmark (1:02:34.160)
there's two to the power of 1 million possible images,
Lex Fridman (1:02:36.960)
which is way more than there are atoms in our universe,
Max Tegmark (1:02:38.960)
right, so in order to,
Lex Fridman (1:02:42.040)
and then for each one of those,
Max Tegmark (1:02:43.200)
I have to assign a number,
Lex Fridman (1:02:44.640)
which is the probability that it's a dog.
Lex Fridman (1:02:47.080)
So an arbitrary function of images
Lex Fridman (1:02:49.440)
is a list of more numbers than there are atoms in our universe.
Lex Fridman (1:02:54.440)
So clearly I can't store that under the hood of my GPU
Lex Fridman (1:02:57.360)
or my computer, yet somehow it works.
Lex Fridman (1:03:00.640)
So what does that mean?
Lex Fridman (1:03:01.480)
Well, it means that out of all of the problems
Max Tegmark (1:03:04.960)
that you could try to solve with a neural network,
Lex Fridman (1:03:10.120)
almost all of them are impossible to solve
Max Tegmark (1:03:12.880)
with a reasonably sized one.
Lex Fridman (1:03:15.480)
But then what we showed in our paper
Max Tegmark (1:03:17.440)
was that the fraction, the kind of problems,
Lex Fridman (1:03:22.360)
the fraction of all the problems
Max Tegmark (1:03:23.800)
that you could possibly pose,
Lex Fridman (1:03:26.520)
that we actually care about given the laws of physics
Max Tegmark (1:03:29.480)
is also an infinite testimony, tiny little part.
Lex Fridman (1:03:32.480)
And amazingly, they're basically the same part.
Max Tegmark (1:03:35.440)
Yeah, it's almost like our world was created for,
Lex Fridman (1:03:37.560)
I mean, they kind of come together.
Max Tegmark (1:03:39.000)
Yeah, well, you could say maybe where the world was created
Lex Fridman (1:03:42.800)
for us, but I have a more modest interpretation,
Max Tegmark (1:03:44.960)
which is that the world was created for us,
Lex Fridman (1:03:46.680)
but I have a more modest interpretation,
Max Tegmark (1:03:48.040)
which is that instead evolution endowed us
Lex Fridman (1:03:50.360)
with neural networks precisely for that reason.
Max Tegmark (1:03:53.120)
Because this particular architecture,
Lex Fridman (1:03:54.640)
as opposed to the one in your laptop,
Max Tegmark (1:03:56.040)
is very, very well adapted to solving the kind of problems
Lex Fridman (1:04:02.480)
that nature kept presenting our ancestors with.
Lex Fridman (1:04:05.560)
So it makes sense that why do we have a brain
Lex Fridman (1:04:08.120)
in the first place?
Max Tegmark (1:04:09.280)
It's to be able to make predictions about the future
Lex Fridman (1:04:11.880)
and so on.
Lex Fridman (1:04:12.880)
So if we had a sucky system, which could never solve it,
Lex Fridman (1:04:16.440)
we wouldn't have a world.
Lex Fridman (1:04:18.280)
So this is, I think, a very beautiful fact.
Lex Fridman (1:04:23.680)
Yeah.
Max Tegmark (1:04:24.520)
We also realize that there's been earlier work
Lex Fridman (1:04:29.000)
on why deeper networks are good,
Lex Fridman (1:04:32.040)
but we were able to show an additional cool fact there,
Lex Fridman (1:04:34.680)
which is that even incredibly simple problems,
Max Tegmark (1:04:38.360)
like suppose I give you a thousand numbers
Lex Fridman (1:04:41.080)
and ask you to multiply them together,
Lex Fridman (1:04:42.720)
and you can write a few lines of code, boom, done, trivial.
Lex Fridman (1:04:46.680)
If you just try to do that with a neural network
Max Tegmark (1:04:49.520)
that has only one single hidden layer in it,
Lex Fridman (1:04:52.440)
you can do it,
Lex Fridman (1:04:54.320)
but you're going to need two to the power of a thousand
Lex Fridman (1:04:57.360)
neurons to multiply a thousand numbers,
Max Tegmark (1:05:00.920)
which is, again, more neurons than there are atoms
Lex Fridman (1:05:02.520)
in our universe.
Max Tegmark (1:05:04.600)
That's fascinating.
Lex Fridman (1:05:05.480)
But if you allow yourself to make it a deep network
Max Tegmark (1:05:09.960)
with many layers, you only need 4,000 neurons.
Lex Fridman (1:05:13.240)
It's perfectly feasible.
Max Tegmark (1:05:16.400)
That's really interesting.
Lex Fridman (1:05:17.960)
Yeah.
Lex Fridman (1:05:18.800)
So on another architecture type,
Lex Fridman (1:05:21.040)
I mean, you mentioned Schrodinger's equation,
Lex Fridman (1:05:22.720)
and what are your thoughts about quantum computing
Lex Fridman (1:05:27.240)
and the role of this kind of computational unit
Lex Fridman (1:05:32.400)
in creating an intelligence system?
Lex Fridman (1:05:34.880)
In some Hollywood movies that I will not mention by name
Max Tegmark (1:05:39.520)
because I don't want to spoil them.
Lex Fridman (1:05:41.040)
The way they get AGI is building a quantum computer.
Max Tegmark (1:05:45.480)
Because the word quantum sounds cool and so on.
Lex Fridman (1:05:47.600)
That's right.
Max Tegmark (1:05:50.040)
First of all, I think we don't need quantum computers
Lex Fridman (1:05:52.880)
to build AGI.
Max Tegmark (1:05:54.920)
I suspect your brain is not a quantum computer
Lex Fridman (1:05:59.240)
in any profound sense.
Lex Fridman (1:06:01.600)
So you don't even wrote a paper about that
Lex Fridman (1:06:03.200)
a lot many years ago.
Max Tegmark (1:06:04.560)
I calculated the so called decoherence time,
Lex Fridman (1:06:08.120)
how long it takes until the quantum computerness
Max Tegmark (1:06:10.320)
of what your neurons are doing gets erased
Lex Fridman (1:06:15.320)
by just random noise from the environment.
Lex Fridman (1:06:17.960)
And it's about 10 to the minus 21 seconds.
Lex Fridman (1:06:21.320)
So as cool as it would be to have a quantum computer
Max Tegmark (1:06:24.600)
in my head, I don't think that fast.
Lex Fridman (1:06:27.320)
On the other hand,
Max Tegmark (1:06:28.360)
there are very cool things you could do
Lex Fridman (1:06:33.040)
with quantum computers.
Max Tegmark (1:06:35.240)
Or I think we'll be able to do soon
Lex Fridman (1:06:37.480)
when we get bigger ones.
Max Tegmark (1:06:39.360)
That might actually help machine learning
Lex Fridman (1:06:40.960)
do even better than the brain.
Lex Fridman (1:06:43.160)
So for example,
Lex Fridman (1:06:47.040)
one, this is just a moonshot,
Lex Fridman (1:06:50.760)
but learning is very much same thing as search.
Lex Fridman (1:07:01.800)
If you're trying to train a neural network
Max Tegmark (1:07:03.160)
to get really learned to do something really well,
Lex Fridman (1:07:06.240)
you have some loss function,
Max Tegmark (1:07:07.280)
you have a bunch of knobs you can turn,
Lex Fridman (1:07:10.360)
represented by a bunch of numbers,
Lex Fridman (1:07:12.080)
and you're trying to tweak them
Lex Fridman (1:07:12.920)
so that it becomes as good as possible at this thing.
Lex Fridman (1:07:15.080)
So if you think of a landscape with some valley,
Lex Fridman (1:07:20.720)
where each dimension of the landscape
Max Tegmark (1:07:22.120)
corresponds to some number you can change,
Lex Fridman (1:07:24.120)
you're trying to find the minimum.
Lex Fridman (1:07:25.640)
And it's well known that
Lex Fridman (1:07:26.760)
if you have a very high dimensional landscape,
Max Tegmark (1:07:29.040)
complicated things, it's super hard to find the minimum.
Lex Fridman (1:07:31.840)
Quantum mechanics is amazingly good at this.
Max Tegmark (1:07:35.840)
Like if I want to know what's the lowest energy state
Lex Fridman (1:07:38.240)
this water can possibly have,
Max Tegmark (1:07:41.720)
incredibly hard to compute,
Lex Fridman (1:07:42.560)
but nature will happily figure this out for you
Max Tegmark (1:07:45.400)
if you just cool it down, make it very, very cold.
Lex Fridman (1:07:49.800)
If you put a ball somewhere,
Max Tegmark (1:07:50.880)
it'll roll down to its minimum.
Lex Fridman (1:07:52.240)
And this happens metaphorically
Max Tegmark (1:07:54.280)
at the energy landscape too.
Lex Fridman (1:07:56.320)
And quantum mechanics even uses some clever tricks,
Max Tegmark (1:07:59.280)
which today's machine learning systems don't.
Lex Fridman (1:08:02.520)
Like if you're trying to find the minimum
Lex Fridman (1:08:04.160)
and you get stuck in the little local minimum here,
Lex Fridman (1:08:06.960)
in quantum mechanics you can actually tunnel
Max Tegmark (1:08:08.760)
through the barrier and get unstuck again.
Lex Fridman (1:08:13.480)
That's really interesting.
Max Tegmark (1:08:14.320)
Yeah, so it may be, for example,
Lex Fridman (1:08:16.120)
that we'll one day use quantum computers
Max Tegmark (1:08:19.160)
that help train neural networks better.
Lex Fridman (1:08:22.840)
That's really interesting.
Max Tegmark (1:08:23.680)
Okay, so as a component of kind of the learning process,
Lex Fridman (1:08:27.040)
for example.
Max Tegmark (1:08:27.880)
Yeah.
Lex Fridman (1:08:29.440)
Let me ask sort of wrapping up here a little bit,
Max Tegmark (1:08:33.080)
let me return to the questions of our human nature
Lex Fridman (1:08:36.880)
and love, as I mentioned.
Lex Fridman (1:08:40.000)
So do you think,
Lex Fridman (1:08:44.280)
you mentioned sort of a helper robot,
Lex Fridman (1:08:46.000)
but you could think of also personal robots.
Lex Fridman (1:08:48.640)
Do you think the way we human beings fall in love
Lex Fridman (1:08:52.480)
and get connected to each other
Lex Fridman (1:08:54.680)
is possible to achieve in an AI system
Lex Fridman (1:08:58.040)
and human level AI intelligence system?
Lex Fridman (1:09:00.360)
Do you think we would ever see that kind of connection?
Max Tegmark (1:09:03.720)
Or, you know, in all this discussion
Lex Fridman (1:09:06.160)
about solving complex goals,
Max Tegmark (1:09:08.520)
is this kind of human social connection,
Lex Fridman (1:09:10.760)
do you think that's one of the goals
Max Tegmark (1:09:12.560)
on the peaks and valleys with the raising sea levels
Lex Fridman (1:09:16.280)
that we'll be able to achieve?
Max Tegmark (1:09:17.360)
Or do you think that's something that's ultimately,
Lex Fridman (1:09:20.040)
or at least in the short term,
Lex Fridman (1:09:21.760)
relative to the other goals is not achievable?
Lex Fridman (1:09:23.640)
I think it's all possible.
Lex Fridman (1:09:25.120)
And I mean, in recent,
Lex Fridman (1:09:27.600)
there's a very wide range of guesses, as you know,
Max Tegmark (1:09:30.840)
among AI researchers, when we're going to get AGI.
Lex Fridman (1:09:35.120)
Some people, you know, like our friend Rodney Brooks
Max Tegmark (1:09:37.640)
says it's going to be hundreds of years at least.
Lex Fridman (1:09:41.040)
And then there are many others
Max Tegmark (1:09:42.200)
who think it's going to happen much sooner.
Lex Fridman (1:09:44.040)
And recent polls,
Max Tegmark (1:09:46.840)
maybe half or so of AI researchers
Lex Fridman (1:09:48.640)
think we're going to get AGI within decades.
Lex Fridman (1:09:50.920)
So if that happens, of course,
Lex Fridman (1:09:52.720)
then I think these things are all possible.
Lex Fridman (1:09:55.040)
But in terms of whether it will happen,
Lex Fridman (1:09:56.840)
I think we shouldn't spend so much time asking
Lex Fridman (1:10:00.600)
what do we think will happen in the future?
Lex Fridman (1:10:03.240)
As if we are just some sort of pathetic,
Max Tegmark (1:10:05.160)
your passive bystanders, you know,
Lex Fridman (1:10:07.040)
waiting for the future to happen to us.
Lex Fridman (1:10:09.280)
Hey, we're the ones creating this future, right?
Lex Fridman (1:10:11.640)
So we should be proactive about it
Lex Fridman (1:10:15.520)
and ask ourselves what sort of future
Lex Fridman (1:10:16.920)
we would like to have happen.
Max Tegmark (1:10:18.240)
We're going to make it like that.
Lex Fridman (1:10:19.920)
Well, what I prefer is just some sort of incredibly boring,
Max Tegmark (1:10:22.720)
zombie like future where there's all these
Lex Fridman (1:10:24.320)
mechanical things happening and there's no passion,
Max Tegmark (1:10:26.040)
no emotion, no experience, maybe even.
Lex Fridman (1:10:29.600)
No, I would of course, much rather prefer it
Max Tegmark (1:10:32.040)
if all the things that we find that we value the most
Lex Fridman (1:10:36.240)
about humanity are our subjective experience,
Max Tegmark (1:10:40.680)
passion, inspiration, love, you know.
Lex Fridman (1:10:43.000)
If we can create a future where those things do happen,
Max Tegmark (1:10:48.000)
where those things do exist, you know,
Lex Fridman (1:10:50.840)
I think ultimately it's not our universe
Max Tegmark (1:10:54.560)
giving meaning to us, it's us giving meaning to our universe.
Lex Fridman (1:10:57.960)
And if we build more advanced intelligence,
Max Tegmark (1:11:01.840)
let's make sure we build it in such a way
Lex Fridman (1:11:03.680)
that meaning is part of it.
Max Tegmark (1:11:09.120)
A lot of people that seriously study this problem
Lex Fridman (1:11:11.400)
and think of it from different angles
Max Tegmark (1:11:13.600)
have trouble in the majority of cases,
Lex Fridman (1:11:16.880)
if they think through that happen,
Max Tegmark (1:11:19.160)
are the ones that are not beneficial to humanity.
Lex Fridman (1:11:22.520)
And so, yeah, so what are your thoughts?
Max Tegmark (1:11:25.560)
What's should people, you know,
Lex Fridman (1:11:29.400)
I really don't like people to be terrified.
Max Tegmark (1:11:33.440)
What's a way for people to think about it
Lex Fridman (1:11:35.040)
in a way we can solve it and we can make it better?
Max Tegmark (1:11:39.600)
No, I don't think panicking is going to help in any way.
Lex Fridman (1:11:42.960)
It's not going to increase chances
Max Tegmark (1:11:44.840)
of things going well either.
Lex Fridman (1:11:45.880)
Even if you are in a situation where there is a real threat,
Lex Fridman (1:11:48.400)
does it help if everybody just freaks out?
Lex Fridman (1:11:51.080)
No, of course, of course not.
Max Tegmark (1:11:53.640)
I think, yeah, there are of course ways
Lex Fridman (1:11:56.600)
in which things can go horribly wrong.
Max Tegmark (1:11:59.560)
First of all, it's important when we think about this thing,
Lex Fridman (1:12:03.680)
about the problems and risks,
Max Tegmark (1:12:05.280)
to also remember how huge the upsides can be
Lex Fridman (1:12:07.160)
if we get it right, right?
Max Tegmark (1:12:08.440)
Everything we love about society and civilization
Lex Fridman (1:12:12.360)
is a product of intelligence.
Lex Fridman (1:12:13.400)
So if we can amplify our intelligence
Lex Fridman (1:12:15.320)
with machine intelligence and not anymore lose our loved one
Max Tegmark (1:12:18.760)
to what we're told is an incurable disease
Lex Fridman (1:12:21.080)
and things like this, of course, we should aspire to that.
Lex Fridman (1:12:24.800)
So that can be a motivator, I think,
Lex Fridman (1:12:26.680)
reminding ourselves that the reason we try to solve problems
Max Tegmark (1:12:29.120)
is not just because we're trying to avoid gloom,
Lex Fridman (1:12:33.520)
but because we're trying to do something great.
Lex Fridman (1:12:35.760)
But then in terms of the risks,
Lex Fridman (1:12:37.680)
I think the really important question is to ask,
Lex Fridman (1:12:42.680)
what can we do today that will actually help
Lex Fridman (1:12:45.480)
make the outcome good, right?
Lex Fridman (1:12:47.320)
And dismissing the risk is not one of them.
Lex Fridman (1:12:51.240)
I find it quite funny often when I'm in discussion panels
Max Tegmark (1:12:54.800)
about these things,
Lex Fridman (1:12:55.960)
how the people who work for companies,
Max Tegmark (1:13:01.200)
always be like, oh, nothing to worry about,
Lex Fridman (1:13:03.120)
nothing to worry about, nothing to worry about.
Lex Fridman (1:13:04.760)
And it's only academics sometimes express concerns.
Lex Fridman (1:13:09.600)
That's not surprising at all if you think about it.
Max Tegmark (1:13:11.880)
Right.
Lex Fridman (1:13:12.880)
Upton Sinclair quipped, right,
Max Tegmark (1:13:15.200)
that it's hard to make a man believe in something
Lex Fridman (1:13:18.040)
when his income depends on not believing in it.
Lex Fridman (1:13:20.120)
And frankly, we know a lot of these people in companies
Lex Fridman (1:13:24.080)
that they're just as concerned as anyone else.
Lex Fridman (1:13:26.240)
But if you're the CEO of a company,
Lex Fridman (1:13:28.480)
that's not something you want to go on record saying
Max Tegmark (1:13:30.280)
when you have silly journalists who are gonna put a picture
Lex Fridman (1:13:33.440)
of a Terminator robot when they quote you.
Lex Fridman (1:13:35.720)
So the issues are real.
Lex Fridman (1:13:39.040)
And the way I think about what the issue is,
Max Tegmark (1:13:41.920)
is basically the real choice we have is,
Lex Fridman (1:13:48.040)
first of all, are we gonna just dismiss the risks
Lex Fridman (1:13:50.840)
and say, well, let's just go ahead and build machines
Lex Fridman (1:13:54.480)
that can do everything we can do better and cheaper.
Max Tegmark (1:13:57.560)
Let's just make ourselves obsolete as fast as possible.
Lex Fridman (1:14:00.200)
What could possibly go wrong?
Max Tegmark (1:14:01.720)
That's one attitude.
Lex Fridman (1:14:03.440)
The opposite attitude, I think, is to say,
Max Tegmark (1:14:06.400)
here's this incredible potential,
Lex Fridman (1:14:08.800)
let's think about what kind of future
Max Tegmark (1:14:11.960)
we're really, really excited about.
Lex Fridman (1:14:14.640)
What are the shared goals that we can really aspire towards?
Lex Fridman (1:14:18.480)
And then let's think really hard
Lex Fridman (1:14:19.960)
about how we can actually get there.
Lex Fridman (1:14:22.000)
So start with, don't start thinking about the risks,
Lex Fridman (1:14:24.160)
start thinking about the goals.
Lex Fridman (1:14:26.720)
And then when you do that,
Lex Fridman (1:14:28.200)
then you can think about the obstacles you want to avoid.
Max Tegmark (1:14:30.480)
I often get students coming in right here into my office
Lex Fridman (1:14:32.840)
for career advice.
Max Tegmark (1:14:34.120)
I always ask them this very question,
Lex Fridman (1:14:35.560)
where do you want to be in the future?
Max Tegmark (1:14:37.920)
If all she can say is, oh, maybe I'll have cancer,
Lex Fridman (1:14:40.640)
maybe I'll get run over by a truck.
Max Tegmark (1:14:42.480)
Yeah, focus on the obstacles instead of the goals.
Lex Fridman (1:14:44.280)
She's just going to end up a hypochondriac paranoid.
Max Tegmark (1:14:47.920)
Whereas if she comes in and fire in her eyes
Lex Fridman (1:14:49.920)
and is like, I want to be there.
Lex Fridman (1:14:51.840)
And then we can talk about the obstacles
Lex Fridman (1:14:53.960)
and see how we can circumvent them.
Max Tegmark (1:14:55.760)
That's, I think, a much, much healthier attitude.
Lex Fridman (1:14:58.880)
And I feel it's very challenging to come up with a vision
Max Tegmark (1:15:03.880)
for the future, which we are unequivocally excited about.
Lex Fridman (1:15:08.120)
I'm not just talking now in the vague terms,
Max Tegmark (1:15:10.320)
like, yeah, let's cure cancer, fine.
Lex Fridman (1:15:12.360)
I'm talking about what kind of society
Lex Fridman (1:15:14.720)
do we want to create?
Lex Fridman (1:15:15.840)
What do we want it to mean to be human in the age of AI,
Lex Fridman (1:15:20.360)
in the age of AGI?
Lex Fridman (1:15:22.840)
So if we can have this conversation,
Max Tegmark (1:15:25.360)
broad, inclusive conversation,
Lex Fridman (1:15:28.200)
and gradually start converging towards some,
Max Tegmark (1:15:31.400)
some future that with some direction, at least,
Lex Fridman (1:15:34.240)
that we want to steer towards, right,
Max Tegmark (1:15:35.400)
then we'll be much more motivated
Lex Fridman (1:15:38.160)
to constructively take on the obstacles.
Lex Fridman (1:15:39.960)
And I think if I had, if I had to,
Lex Fridman (1:15:43.560)
if I try to wrap this up in a more succinct way,
Max Tegmark (1:15:46.640)
I think we can all agree already now
Lex Fridman (1:15:51.480)
that we should aspire to build AGI
Max Tegmark (1:15:56.160)
that doesn't overpower us, but that empowers us.
Lex Fridman (1:16:05.160)
And think of the many various ways that can do that,
Max Tegmark (1:16:08.560)
whether that's from my side of the world
Lex Fridman (1:16:11.000)
of autonomous vehicles.
Max Tegmark (1:16:12.720)
I'm personally actually from the camp
Lex Fridman (1:16:14.720)
that believes this human level intelligence
Max Tegmark (1:16:16.800)
is required to achieve something like vehicles
Lex Fridman (1:16:20.480)
that would actually be something we would enjoy using
Lex Fridman (1:16:23.880)
and being part of.
Lex Fridman (1:16:25.120)
So that's one example, and certainly there's a lot
Max Tegmark (1:16:27.040)
of other types of robots and medicine and so on.
Lex Fridman (1:16:30.920)
So focusing on those and then coming up with the obstacles,
Max Tegmark (1:16:33.880)
coming up with the ways that that can go wrong
Lex Fridman (1:16:35.920)
and solving those one at a time.
Lex Fridman (1:16:38.160)
And just because you can build an autonomous vehicle,
Lex Fridman (1:16:41.520)
even if you could build one
Max Tegmark (1:16:42.800)
that would drive just fine without you,
Lex Fridman (1:16:45.080)
maybe there are some things in life
Max Tegmark (1:16:46.720)
that we would actually want to do ourselves.
Lex Fridman (1:16:48.400)
That's right.
Max Tegmark (1:16:49.240)
Right, like, for example,
Lex Fridman (1:16:51.400)
if you think of our society as a whole,
Max Tegmark (1:16:53.040)
there are some things that we find very meaningful to do.
Lex Fridman (1:16:57.200)
And that doesn't mean we have to stop doing them
Max Tegmark (1:16:59.640)
just because machines can do them better.
Lex Fridman (1:17:02.000)
I'm not gonna stop playing tennis
Max Tegmark (1:17:04.080)
just the day someone builds a tennis robot and beat me.
Lex Fridman (1:17:07.360)
People are still playing chess and even go.
Max Tegmark (1:17:09.600)
Yeah, and in the very near term even,
Lex Fridman (1:17:14.600)
some people are advocating basic income, replace jobs.
Lex Fridman (1:17:18.880)
But if the government is gonna be willing
Lex Fridman (1:17:20.840)
to just hand out cash to people for doing nothing,
Max Tegmark (1:17:24.040)
then one should also seriously consider
Lex Fridman (1:17:25.840)
whether the government should also hire
Max Tegmark (1:17:27.640)
a lot more teachers and nurses
Lex Fridman (1:17:29.480)
and the kind of jobs which people often
Lex Fridman (1:17:32.160)
find great fulfillment in doing, right?
Lex Fridman (1:17:34.440)
We get very tired of hearing politicians saying,
Max Tegmark (1:17:36.320)
oh, we can't afford hiring more teachers,
Lex Fridman (1:17:39.320)
but we're gonna maybe have basic income.
Max Tegmark (1:17:41.480)
If we can have more serious research and thought
Lex Fridman (1:17:44.000)
into what gives meaning to our lives,
Lex Fridman (1:17:46.200)
the jobs give so much more than income, right?
Lex Fridman (1:17:48.960)
Mm hmm.
Lex Fridman (1:17:50.520)
And then think about in the future,
Lex Fridman (1:17:53.320)
what are the roles that we wanna have people
Lex Fridman (1:18:00.000)
continually feeling empowered by machines?
Lex Fridman (1:18:03.040)
And I think sort of, I come from Russia,
Max Tegmark (1:18:06.120)
from the Soviet Union.
Lex Fridman (1:18:07.240)
And I think for a lot of people in the 20th century,
Max Tegmark (1:18:10.160)
going to the moon, going to space was an inspiring thing.
Lex Fridman (1:18:14.080)
I feel like the universe of the mind,
Lex Fridman (1:18:18.080)
so AI, understanding, creating intelligence
Lex Fridman (1:18:20.880)
is that for the 21st century.
Lex Fridman (1:18:23.240)
So it's really surprising.
Lex Fridman (1:18:24.400)
And I've heard you mention this.
Max Tegmark (1:18:25.640)
It's really surprising to me,
Lex Fridman (1:18:27.400)
both on the research funding side,
Max Tegmark (1:18:29.240)
that it's not funded as greatly as it could be,
Lex Fridman (1:18:31.760)
but most importantly, on the politician side,
Max Tegmark (1:18:34.760)
that it's not part of the public discourse
Lex Fridman (1:18:36.520)
except in the killer bots terminator kind of view,
Max Tegmark (1:18:40.800)
that people are not yet, I think, perhaps excited
Lex Fridman (1:18:44.880)
by the possible positive future
Max Tegmark (1:18:46.680)
that we can build together.
Lex Fridman (1:18:48.120)
So we should be, because politicians usually just focus
Lex Fridman (1:18:51.520)
on the next election cycle, right?
Lex Fridman (1:18:54.480)
The single most important thing I feel we humans have learned
Max Tegmark (1:18:57.160)
in the entire history of science
Lex Fridman (1:18:59.320)
is they were the masters of underestimation.
Max Tegmark (1:19:02.040)
We underestimated the size of our cosmos again and again,
Lex Fridman (1:19:08.480)
realizing that everything we thought existed
Lex Fridman (1:19:10.200)
was just a small part of something grander, right?
Lex Fridman (1:19:12.240)
Planet, solar system, the galaxy, clusters of galaxies.
Max Tegmark (1:19:16.640)
The universe.
Lex Fridman (1:19:18.440)
And we now know that the future has just
Lex Fridman (1:19:23.120)
so much more potential
Lex Fridman (1:19:25.160)
than our ancestors could ever have dreamt of.
Max Tegmark (1:19:27.640)
This cosmos, imagine if all of Earth
Lex Fridman (1:19:33.600)
was completely devoid of life,
Max Tegmark (1:19:36.640)
except for Cambridge, Massachusetts.
Lex Fridman (1:19:39.560)
Wouldn't it be kind of lame if all we ever aspired to
Max Tegmark (1:19:42.680)
was to stay in Cambridge, Massachusetts forever
Lex Fridman (1:19:45.560)
and then go extinct in one week,
Lex Fridman (1:19:47.160)
even though Earth was gonna continue on for longer?
Lex Fridman (1:19:49.760)
That sort of attitude I think we have now
Max Tegmark (1:19:54.200)
on the cosmic scale, life can flourish on Earth,
Lex Fridman (1:19:57.800)
not for four years, but for billions of years.
Max Tegmark (1:20:00.840)
I can even tell you about how to move it out of harm's way
Lex Fridman (1:20:02.920)
when the sun gets too hot.
Lex Fridman (1:20:04.840)
And then we have so much more resources out here,
Lex Fridman (1:20:09.520)
which today, maybe there are a lot of other planets
Max Tegmark (1:20:12.480)
with bacteria or cow like life on them,
Lex Fridman (1:20:14.960)
but most of this, all this opportunity seems,
Max Tegmark (1:20:19.880)
as far as we can tell, to be largely dead,
Lex Fridman (1:20:22.440)
like the Sahara Desert.
Lex Fridman (1:20:23.560)
And yet we have the opportunity to help life flourish
Lex Fridman (1:20:28.480)
around this for billions of years.
Lex Fridman (1:20:30.280)
So let's quit squabbling about
Lex Fridman (1:20:34.080)
whether some little border should be drawn
Max Tegmark (1:20:36.480)
one mile to the left or right,
Lex Fridman (1:20:38.440)
and look up into the skies and realize,
Max Tegmark (1:20:41.080)
hey, we can do such incredible things.
Lex Fridman (1:20:44.040)
Yeah, and that's, I think, why it's really exciting
Max Tegmark (1:20:46.640)
that you and others are connected
Lex Fridman (1:20:49.440)
with some of the work Elon Musk is doing,
Max Tegmark (1:20:51.880)
because he's literally going out into that space,
Lex Fridman (1:20:54.480)
really exploring our universe, and it's wonderful.
Lex Fridman (1:20:57.000)
That is exactly why Elon Musk is so misunderstood, right?
Lex Fridman (1:21:02.000)
Misconstrued him as some kind of pessimistic doomsayer.
Max Tegmark (1:21:05.000)
The reason he cares so much about AI safety
Lex Fridman (1:21:07.640)
is because he more than almost anyone else appreciates
Max Tegmark (1:21:12.080)
these amazing opportunities that we'll squander
Lex Fridman (1:21:14.280)
if we wipe out here on Earth.
Max Tegmark (1:21:16.640)
We're not just going to wipe out the next generation,
Lex Fridman (1:21:19.680)
all generations, and this incredible opportunity
Max Tegmark (1:21:23.320)
that's out there, and that would really be a waste.
Lex Fridman (1:21:25.400)
And AI, for people who think that it would be better
Max Tegmark (1:21:30.080)
to do without technology, let me just mention that
Lex Fridman (1:21:34.680)
if we don't improve our technology,
Max Tegmark (1:21:36.320)
the question isn't whether humanity is going to go extinct.
Lex Fridman (1:21:39.320)
The question is just whether we're going to get taken out
Max Tegmark (1:21:41.160)
by the next big asteroid or the next super volcano
Lex Fridman (1:21:44.800)
or something else dumb that we could easily prevent
Lex Fridman (1:21:48.280)
with more tech, right?
Lex Fridman (1:21:49.840)
And if we want life to flourish throughout the cosmos,
Max Tegmark (1:21:53.160)
AI is the key to it.
Lex Fridman (1:21:56.120)
As I mentioned in a lot of detail in my book right there,
Max Tegmark (1:21:59.840)
even many of the most inspired sci fi writers,
Lex Fridman (1:22:04.880)
I feel have totally underestimated the opportunities
Max Tegmark (1:22:08.120)
for space travel, especially at the other galaxies,
Lex Fridman (1:22:11.240)
because they weren't thinking about the possibility of AGI,
Max Tegmark (1:22:15.360)
which just makes it so much easier.
Lex Fridman (1:22:17.520)
Right, yeah.
Lex Fridman (1:22:18.440)
So that goes to your view of AGI that enables our progress,
Lex Fridman (1:22:24.080)
that enables a better life.
Lex Fridman (1:22:25.760)
So that's a beautiful way to put it
Lex Fridman (1:22:28.320)
and then something to strive for.
Lex Fridman (1:22:29.960)
So Max, thank you so much.
Lex Fridman (1:22:31.440)
Thank you for your time today.
Max Tegmark (1:22:32.560)
It's been awesome.
Lex Fridman (1:22:33.560)
Thank you so much.
Max Tegmark (1:22:34.400)
Thanks.
Lex Fridman (1:22:35.240)
Have a great day.
Lex Fridman (20:00.560)
got cleaned out of the gene pool, right?
Lex Fridman (20:02.960)
But if you build an artificial general intelligence
Max Tegmark (20:06.880)
the mind space that you can design is much, much larger
Lex Fridman (20:10.040)
than just a specific subset of minds that can evolve.
Lex Fridman (20:14.440)
So an AGI mind doesn't necessarily have
Lex Fridman (20:17.280)
to have any self preservation instinct.
Max Tegmark (20:19.880)
It also doesn't necessarily have to be
Lex Fridman (20:21.600)
so individualistic as us.
Max Tegmark (20:24.040)
Like, imagine if you could just, first of all,
Lex Fridman (20:26.080)
or we are also very afraid of death.
Max Tegmark (20:27.960)
You know, I suppose you could back yourself up
Lex Fridman (20:29.920)
every five minutes and then your airplane
Max Tegmark (20:32.000)
is about to crash.
Lex Fridman (20:32.840)
You're like, shucks, I'm gonna lose the last five minutes
Max Tegmark (20:36.680)
of experiences since my last cloud backup, dang.
Lex Fridman (20:39.520)
You know, it's not as big a deal.
Max Tegmark (20:41.520)
Or if we could just copy experiences between our minds
Lex Fridman (20:45.680)
easily like we, which we could easily do
Lex Fridman (20:47.640)
if we were silicon based, right?
Lex Fridman (20:50.360)
Then maybe we would feel a little bit more
Max Tegmark (20:54.040)
like a hive mind actually, that maybe it's the,
Lex Fridman (20:56.560)
so I don't think we should take for granted at all
Max Tegmark (20:59.960)
that AGI will have to have any of those sort of
Lex Fridman (21:04.880)
competitive as alpha male instincts.
Max Tegmark (21:07.360)
On the other hand, you know, this is really interesting
Lex Fridman (21:10.160)
because I think some people go too far and say,
Max Tegmark (21:13.840)
of course we don't have to have any concerns either
Lex Fridman (21:16.680)
that advanced AI will have those instincts
Max Tegmark (21:20.800)
because we can build anything we want.
Lex Fridman (21:22.680)
That there's a very nice set of arguments going back
Max Tegmark (21:26.280)
to Steve Omohundro and Nick Bostrom and others
Lex Fridman (21:28.560)
just pointing out that when we build machines,
Max Tegmark (21:32.280)
we normally build them with some kind of goal, you know,
Lex Fridman (21:34.680)
win this chess game, drive this car safely or whatever.
Lex Fridman (21:38.520)
And as soon as you put in a goal into machine,
Lex Fridman (21:40.960)
especially if it's kind of open ended goal
Lex Fridman (21:42.760)
and the machine is very intelligent,
Lex Fridman (21:44.640)
it'll break that down into a bunch of sub goals.
Lex Fridman (21:48.280)
And one of those goals will almost always
Lex Fridman (21:51.280)
be self preservation because if it breaks or dies
Lex Fridman (21:54.200)
in the process, it's not gonna accomplish the goal, right?
Lex Fridman (21:56.120)
Like suppose you just build a little,
Max Tegmark (21:58.040)
you have a little robot and you tell it to go down
Lex Fridman (22:01.000)
the store market here and get you some food,
Max Tegmark (22:04.040)
make you cook an Italian dinner, you know,
Lex Fridman (22:06.200)
and then someone mugs it and tries to break it
Max Tegmark (22:08.400)
on the way.
Lex Fridman (22:09.480)
That robot has an incentive to not get destroyed
Lex Fridman (22:12.920)
and defend itself or run away,
Lex Fridman (22:14.720)
because otherwise it's gonna fail in cooking your dinner.
Max Tegmark (22:17.720)
It's not afraid of death,
Lex Fridman (22:19.560)
but it really wants to complete the dinner cooking goal.
Lex Fridman (22:22.960)
So it will have a self preservation instinct.
Lex Fridman (22:25.040)
Continue being a functional agent somehow.
Lex Fridman (22:27.920)
And similarly, if you give any kind of more ambitious goal
Lex Fridman (22:33.720)
to an AGI, it's very likely they wanna acquire
Max Tegmark (22:37.000)
more resources so it can do that better.
Lex Fridman (22:39.840)
And it's exactly from those sort of sub goals
Max Tegmark (22:42.720)
that we might not have intended
Lex Fridman (22:43.800)
that some of the concerns about AGI safety come.
Max Tegmark (22:47.160)
You give it some goal that seems completely harmless.
Lex Fridman (22:50.600)
And then before you realize it,
Max Tegmark (22:53.360)
it's also trying to do these other things
Lex Fridman (22:55.480)
which you didn't want it to do.
Lex Fridman (22:56.920)
And it's maybe smarter than us.
Lex Fridman (22:59.160)
So it's fascinating.
Lex Fridman (23:01.000)
And let me pause just because I am in a very kind
Lex Fridman (23:05.680)
of human centric way, see fear of death
Max Tegmark (23:08.720)
as a valuable motivator.
Lex Fridman (23:11.840)
So you don't think, you think that's an artifact
Max Tegmark (23:16.440)
of evolution, so that's the kind of mind space
Lex Fridman (23:19.120)
evolution created that we're sort of almost obsessed
Max Tegmark (23:22.120)
about self preservation, some kind of genetic flow.
Lex Fridman (23:24.400)
You don't think that's necessary to be afraid of death.
Lex Fridman (23:29.480)
So not just a kind of sub goal of self preservation
Lex Fridman (23:32.920)
just so you can keep doing the thing,
Lex Fridman (23:34.920)
but more fundamentally sort of have the finite thing
Lex Fridman (23:38.720)
like this ends for you at some point.
Max Tegmark (23:43.080)
Interesting.
Lex Fridman (23:44.160)
Do I think it's necessary for what precisely?
Max Tegmark (23:47.440)
For intelligence, but also for consciousness.
Lex Fridman (23:50.920)
So for those, for both, do you think really
Lex Fridman (23:55.040)
like a finite death and the fear of it is important?
Lex Fridman (23:59.120)
So before I can answer, before we can agree
Max Tegmark (24:05.160)
on whether it's necessary for intelligence
Lex Fridman (24:06.960)
or for consciousness, we should be clear
Max Tegmark (24:08.360)
on how we define those two words.
Lex Fridman (24:09.800)
Cause a lot of really smart people define them
Max Tegmark (24:11.960)
in very different ways.
Lex Fridman (24:13.320)
I was on this panel with AI experts
Lex Fridman (24:17.080)
and they couldn't agree on how to define intelligence even.
Lex Fridman (24:20.080)
So I define intelligence simply
Max Tegmark (24:22.000)
as the ability to accomplish complex goals.
Lex Fridman (24:25.640)
I like your broad definition, because again
Max Tegmark (24:27.280)
I don't want to be a carbon chauvinist.
Lex Fridman (24:29.040)
Right.
Lex Fridman (24:30.400)
And in that case, no, certainly
Lex Fridman (24:34.600)
it doesn't require fear of death.
Max Tegmark (24:36.480)
I would say alpha go, alpha zero is quite intelligent.
Lex Fridman (24:40.120)
I don't think alpha zero has any fear of being turned off
Max Tegmark (24:43.080)
because it doesn't understand the concept of it even.
Lex Fridman (24:46.320)
And similarly consciousness.
Max Tegmark (24:48.440)
I mean, you could certainly imagine very simple
Lex Fridman (24:52.240)
kind of experience.
Max Tegmark (24:53.920)
If certain plants have any kind of experience
Lex Fridman (24:57.200)
I don't think they're very afraid of dying
Max Tegmark (24:58.560)
or there's nothing they can do about it anyway much.
Lex Fridman (25:00.920)
So there wasn't that much value in, but more seriously
Max Tegmark (25:04.560)
I think if you ask, not just about being conscious
Lex Fridman (25:09.200)
but maybe having what you would, we might call
Max Tegmark (25:14.320)
an exciting life where you feel passion
Lex Fridman (25:16.400)
and really appreciate the things.
Max Tegmark (25:21.480)
Maybe there somehow, maybe there perhaps it does help
Lex Fridman (25:24.440)
having a backdrop that, Hey, it's finite.
Max Tegmark (25:27.880)
No, let's make the most of this, let's live to the fullest.
Lex Fridman (25:31.200)
So if you knew you were going to live forever
Lex Fridman (25:34.880)
do you think you would change your?
Lex Fridman (25:37.400)
Yeah, I mean, in some perspective
Max Tegmark (25:39.560)
it would be an incredibly boring life living forever.
Lex Fridman (25:43.960)
So in the sort of loose subjective terms that you said
Max Tegmark (25:47.360)
of something exciting and something in this
Lex Fridman (25:50.480)
that other humans would understand, I think is, yeah
Max Tegmark (25:53.240)
it seems that the finiteness of it is important.
Lex Fridman (25:57.120)
Well, the good news I have for you then is
Max Tegmark (25:59.560)
based on what we understand about cosmology
Lex Fridman (26:02.120)
everything is in our universe is probably
Max Tegmark (26:05.120)
ultimately probably finite, although.
Lex Fridman (26:07.960)
Big crunch or big, what's the, the infinite expansion.
Max Tegmark (26:11.560)
Yeah, we could have a big chill or a big crunch
Lex Fridman (26:13.840)
or a big rip or that's the big snap or death bubbles.
Max Tegmark (26:18.440)
All of them are more than a billion years away.
Lex Fridman (26:20.040)
So we should, we certainly have vastly more time
Max Tegmark (26:24.600)
than our ancestors thought, but there is still
Lex Fridman (26:29.160)
it's still pretty hard to squeeze in an infinite number
Max Tegmark (26:32.360)
of compute cycles, even though there are some loopholes
Lex Fridman (26:36.560)
that just might be possible.
Lex Fridman (26:37.720)
But I think, you know, some people like to say
Lex Fridman (26:41.960)
that you should live as if you're about to
Max Tegmark (26:44.760)
you're going to die in five years or so.
Lex Fridman (26:46.720)
And that's sort of optimal.
Max Tegmark (26:47.960)
Maybe it's a good assumption.
Lex Fridman (26:50.560)
We should build our civilization as if it's all finite
Max Tegmark (26:54.680)
to be on the safe side.
Lex Fridman (26:55.680)
Right, exactly.
Lex Fridman (26:56.960)
So you mentioned defining intelligence
Lex Fridman (26:59.720)
as the ability to solve complex goals.
Max Tegmark (27:02.960)
Where would you draw a line or how would you try
Lex Fridman (27:05.440)
to define human level intelligence
Lex Fridman (27:08.200)
and superhuman level intelligence?
Lex Fridman (27:10.680)
Where is consciousness part of that definition?
Max Tegmark (27:13.280)
No, consciousness does not come into this definition.
Lex Fridman (27:16.640)
So, so I think of intelligence as it's a spectrum
Lex Fridman (27:20.280)
but there are very many different kinds of goals
Lex Fridman (27:21.960)
you can have.
Max Tegmark (27:22.800)
You can have a goal to be a good chess player
Lex Fridman (27:24.000)
a good goal player, a good car driver, a good investor
Max Tegmark (27:28.520)
good poet, et cetera.
Lex Fridman (27:31.160)
So intelligence that by its very nature
Max Tegmark (27:34.320)
isn't something you can measure by this one number
Lex Fridman (27:36.680)
or some overall goodness.
Max Tegmark (27:37.960)
No, no.
Lex Fridman (27:38.800)
There are some people who are more better at this.
Max Tegmark (27:40.320)
Some people are better than that.
Lex Fridman (27:42.360)
Right now we have machines that are much better than us
Max Tegmark (27:45.440)
at some very narrow tasks like multiplying large numbers
Lex Fridman (27:49.040)
fast, memorizing large databases, playing chess
Max Tegmark (27:53.200)
playing go and soon driving cars.
Lex Fridman (27:57.480)
But there's still no machine that can match
Max Tegmark (28:00.080)
a human child in general intelligence
Lex Fridman (28:02.720)
but artificial general intelligence, AGI
Max Tegmark (28:05.720)
the name of your course, of course
Lex Fridman (28:07.880)
that is by its very definition, the quest
Max Tegmark (28:13.400)
to build a machine that can do everything
Lex Fridman (28:16.000)
as well as we can.
Lex Fridman (28:17.800)
So the old Holy grail of AI from back to its inception
Lex Fridman (28:21.960)
in the sixties, if that ever happens, of course
Max Tegmark (28:25.560)
I think it's going to be the biggest transition
Lex Fridman (28:27.320)
in the history of life on earth
Lex Fridman (28:29.040)
but it doesn't necessarily have to wait the big impact
Lex Fridman (28:33.200)
until machines are better than us at knitting
Max Tegmark (28:35.400)
that the really big change doesn't come exactly
Lex Fridman (28:39.160)
at the moment they're better than us at everything.
Max Tegmark (28:41.800)
The really big change comes first
Lex Fridman (28:44.120)
there are big changes when they start becoming better
Max Tegmark (28:45.840)
at us at doing most of the jobs that we do
Lex Fridman (28:48.800)
because that takes away much of the demand
Max Tegmark (28:51.160)
for human labor.
Lex Fridman (28:53.200)
And then the really whopping change comes
Lex Fridman (28:55.640)
when they become better than us at AI research, right?
Lex Fridman (29:01.040)
Because right now the timescale of AI research
Max Tegmark (29:03.760)
is limited by the human research and development cycle
Lex Fridman (29:08.400)
of years typically, you know
Lex Fridman (29:10.160)
how long does it take from one release of some software
Lex Fridman (29:13.480)
or iPhone or whatever to the next?
Lex Fridman (29:15.720)
But once Google can replace 40,000 engineers
Lex Fridman (29:20.920)
by 40,000 equivalent pieces of software or whatever
Lex Fridman (29:26.400)
but then there's no reason that has to be years
Lex Fridman (29:29.680)
it can be in principle much faster
Lex Fridman (29:31.840)
and the timescale of future progress in AI
Lex Fridman (29:36.040)
and all of science and technology will be driven
Max Tegmark (29:39.320)
by machines, not humans.
Lex Fridman (29:40.960)
So it's this simple point which gives right
Max Tegmark (29:46.520)
this incredibly fun controversy
Lex Fridman (29:48.720)
about whether there can be intelligence explosion
Lex Fridman (29:51.880)
so called singularity as Werner Vinge called it.
Lex Fridman (29:54.400)
Now the idea is articulated by I.J. Good
Max Tegmark (29:57.040)
is obviously way back fifties
Lex Fridman (29:59.480)
but you can see Alan Turing
Lex Fridman (30:01.040)
and others thought about it even earlier.
Lex Fridman (30:06.920)
So you asked me what exactly would I define
Max Tegmark (30:10.080)
human level intelligence, yeah.
Lex Fridman (30:12.800)
So the glib answer is to say something
Max Tegmark (30:15.680)
which is better than us at all cognitive tasks
Lex Fridman (30:18.520)
with a better than any human at all cognitive tasks
Lex Fridman (30:21.800)
but the really interesting bar
Lex Fridman (30:23.080)
I think goes a little bit lower than that actually.
Max Tegmark (30:25.760)
It's when they can, when they're better than us
Lex Fridman (30:27.920)
at AI programming and general learning
Lex Fridman (30:31.760)
so that they can if they want to get better
Lex Fridman (30:35.360)
than us at anything by just studying.
Lex Fridman (30:37.240)
So they're better is a key word and better is towards
Lex Fridman (30:40.560)
this kind of spectrum of the complexity of goals
Max Tegmark (30:44.120)
it's able to accomplish.
Lex Fridman (30:45.680)
So another way to, and that's certainly
Max Tegmark (30:50.360)
a very clear definition of human love.
Lex Fridman (30:53.040)
So there's, it's almost like a sea that's rising
Max Tegmark (30:55.240)
you can do more and more and more things
Lex Fridman (30:56.800)
it's a geographic that you show
Max Tegmark (30:58.640)
it's really nice way to put it.
Lex Fridman (30:59.880)
So there's some peaks that
Lex Fridman (31:01.560)
and there's an ocean level elevating
Lex Fridman (31:03.280)
and you solve more and more problems
Lex Fridman (31:04.800)
but just kind of to take a pause
Lex Fridman (31:07.720)
and we took a bunch of questions
Lex Fridman (31:09.000)
and a lot of social networks
Lex Fridman (31:10.240)
and a bunch of people asked
Max Tegmark (31:11.720)
a sort of a slightly different direction
Lex Fridman (31:14.480)
on creativity and things that perhaps aren't a peak.
Max Tegmark (31:23.560)
Human beings are flawed
Lex Fridman (31:24.720)
and perhaps better means having contradiction
Max Tegmark (31:28.720)
being flawed in some way.
Lex Fridman (31:30.200)
So let me sort of start easy, first of all.
Lex Fridman (31:34.960)
So you have a lot of cool equations.
Lex Fridman (31:36.600)
Let me ask, what's your favorite equation, first of all?
Max Tegmark (31:39.760)
I know they're all like your children, but like
Lex Fridman (31:42.760)
which one is that?
Max Tegmark (31:43.680)
This is the shirt in your equation.
Lex Fridman (31:45.560)
It's the master key of quantum mechanics
Max Tegmark (31:48.640)
of the micro world.
Lex Fridman (31:49.880)
So this equation will protect everything
Max Tegmark (31:52.800)
to do with atoms, molecules and all the way up.
Lex Fridman (31:55.840)
Right?
Max Tegmark (31:58.560)
Yeah, so, okay.
Lex Fridman (31:59.760)
So quantum mechanics is certainly a beautiful
Max Tegmark (32:02.080)
mysterious formulation of our world.
Lex Fridman (32:05.160)
So I'd like to sort of ask you, just as an example
Max Tegmark (32:08.760)
it perhaps doesn't have the same beauty as physics does
Lex Fridman (32:12.160)
but in mathematics abstract, the Andrew Wiles
Max Tegmark (32:16.960)
who proved the Fermat's last theorem.
Lex Fridman (32:19.360)
So he just saw this recently
Lex Fridman (32:22.040)
and it kind of caught my eye a little bit.
Lex Fridman (32:24.160)
This is 358 years after it was conjectured.
Lex Fridman (32:27.960)
So this is very simple formulation.
Lex Fridman (32:29.960)
Everybody tried to prove it, everybody failed.
Lex Fridman (32:32.640)
And so here's this guy comes along
Lex Fridman (32:34.800)
and eventually proves it and then fails to prove it
Lex Fridman (32:38.640)
and then proves it again in 94.
Lex Fridman (32:41.320)
And he said like the moment when everything connected
Max Tegmark (32:43.480)
into place in an interview said
Lex Fridman (32:46.040)
it was so indescribably beautiful.
Max Tegmark (32:47.880)
That moment when you finally realize the connecting piece
Lex Fridman (32:51.040)
of two conjectures.
Max Tegmark (32:52.800)
He said, it was so indescribably beautiful.
Lex Fridman (32:55.280)
It was so simple and so elegant.
Max Tegmark (32:57.040)
I couldn't understand how I'd missed it.
Lex Fridman (32:58.760)
And I just stared at it in disbelief for 20 minutes.
Max Tegmark (33:02.080)
Then during the day, I walked around the department
Lex Fridman (33:05.240)
and I keep coming back to my desk
Max Tegmark (33:07.880)
looking to see if it was still there.
Lex Fridman (33:09.840)
It was still there.
Max Tegmark (33:10.680)
I couldn't contain myself.
Lex Fridman (33:11.760)
I was so excited.
Max Tegmark (33:12.880)
It was the most important moment on my working life.
Lex Fridman (33:15.880)
Nothing I ever do again will mean as much.
Lex Fridman (33:18.960)
So that particular moment.
Lex Fridman (33:20.800)
And it kind of made me think of what would it take?
Lex Fridman (33:24.640)
And I think we have all been there at small levels.
Lex Fridman (33:29.480)
Maybe let me ask, have you had a moment like that
Lex Fridman (33:32.240)
in your life where you just had an idea?
Lex Fridman (33:34.880)
It's like, wow, yes.
Max Tegmark (33:40.000)
I wouldn't mention myself in the same breath
Lex Fridman (33:42.480)
as Andrew Wiles, but I've certainly had a number
Max Tegmark (33:44.760)
of aha moments when I realized something very cool
Lex Fridman (33:52.200)
about physics, which has completely made my head explode.
Max Tegmark (33:56.000)
In fact, some of my favorite discoveries I made later,
Lex Fridman (33:58.320)
I later realized that they had been discovered earlier
Max Tegmark (34:01.080)
by someone who sometimes got quite famous for it.
Lex Fridman (34:03.240)
So it's too late for me to even publish it,
Lex Fridman (34:05.480)
but that doesn't diminish in any way.
Lex Fridman (34:07.440)
The emotional experience you have when you realize it,
Max Tegmark (34:09.760)
like, wow.
Lex Fridman (34:11.320)
Yeah, so what would it take in that moment, that wow,
Lex Fridman (34:15.520)
that was yours in that moment?
Lex Fridman (34:17.320)
So what do you think it takes for an intelligence system,
Lex Fridman (34:21.440)
an AGI system, an AI system to have a moment like that?
Lex Fridman (34:25.640)
That's a tricky question
Lex Fridman (34:26.760)
because there are actually two parts to it, right?
Lex Fridman (34:29.200)
One of them is, can it accomplish that proof?
Max Tegmark (34:33.920)
Can it prove that you can never write A to the N
Lex Fridman (34:37.640)
plus B to the N equals three to that equal Z to the N
Max Tegmark (34:42.760)
for all integers, et cetera, et cetera,
Lex Fridman (34:45.320)
when N is bigger than two?
Max Tegmark (34:48.720)
That's simply a question about intelligence.
Lex Fridman (34:51.360)
Can you build machines that are that intelligent?
Lex Fridman (34:54.120)
And I think by the time we get a machine
Lex Fridman (34:57.280)
that can independently come up with that level of proofs,
Max Tegmark (35:00.840)
probably quite close to AGI.
Lex Fridman (35:03.360)
The second question is a question about consciousness.
Max Tegmark (35:07.240)
When will we, how likely is it that such a machine
Lex Fridman (35:11.760)
will actually have any experience at all,
Lex Fridman (35:14.240)
as opposed to just being like a zombie?
Lex Fridman (35:16.160)
And would we expect it to have some sort of emotional response
Max Tegmark (35:20.560)
to this or anything at all akin to human emotion
Lex Fridman (35:24.640)
where when it accomplishes its machine goal,
Max Tegmark (35:28.320)
it views it as somehow something very positive
Lex Fridman (35:31.920)
and sublime and deeply meaningful?
Max Tegmark (35:39.160)
I would certainly hope that if in the future
Lex Fridman (35:41.440)
we do create machines that are our peers
Max Tegmark (35:45.120)
or even our descendants, that I would certainly
Lex Fridman (35:50.160)
hope that they do have this sublime appreciation of life.
Max Tegmark (35:55.480)
In a way, my absolutely worst nightmare
Lex Fridman (35:58.840)
would be that at some point in the future,
Max Tegmark (36:05.760)
the distant future, maybe our cosmos
Lex Fridman (36:07.400)
is teeming with all this post biological life doing
Max Tegmark (36:10.600)
all the seemingly cool stuff.
Lex Fridman (36:12.880)
And maybe the last humans, by the time
Max Tegmark (36:16.480)
our species eventually fizzles out,
Lex Fridman (36:20.120)
will be like, well, that's OK because we're
Lex Fridman (36:21.920)
so proud of our descendants here.
Lex Fridman (36:23.600)
And look what all the, my worst nightmare
Max Tegmark (36:26.680)
is that we haven't solved the consciousness problem.
Lex Fridman (36:30.360)
And we haven't realized that these are all the zombies.
Max Tegmark (36:32.880)
They're not aware of anything any more than a tape recorder
Lex Fridman (36:36.200)
has any kind of experience.
Lex Fridman (36:37.840)
So the whole thing has just become
Lex Fridman (36:40.040)
a play for empty benches.
Max Tegmark (36:41.520)
That would be the ultimate zombie apocalypse.
Lex Fridman (36:44.640)
So I would much rather, in that case,
Max Tegmark (36:47.200)
that we have these beings which can really
Lex Fridman (36:52.240)
appreciate how amazing it is.
Lex Fridman (36:57.000)
And in that picture, what would be the role of creativity?
Lex Fridman (37:01.080)
A few people ask about creativity.
Max Tegmark (37:04.960)
When you think about intelligence,
Lex Fridman (37:07.080)
certainly the story you told at the beginning of your book
Max Tegmark (37:09.840)
involved creating movies and so on, making money.
Lex Fridman (37:15.200)
You can make a lot of money in our modern world
Max Tegmark (37:17.240)
with music and movies.
Lex Fridman (37:18.600)
So if you are an intelligent system,
Max Tegmark (37:20.880)
you may want to get good at that.
Lex Fridman (37:22.960)
But that's not necessarily what I mean by creativity.
Max Tegmark (37:26.280)
Is it important on that complex goals
Lex Fridman (37:29.640)
where the sea is rising for there
Lex Fridman (37:31.600)
to be something creative?
Lex Fridman (37:33.800)
Or am I being very human centric and thinking creativity
Lex Fridman (37:37.400)
somehow special relative to intelligence?
Lex Fridman (37:41.880)
My hunch is that we should think of creativity simply
Max Tegmark (37:47.240)
as an aspect of intelligence.
Lex Fridman (37:50.760)
And we have to be very careful with human vanity.
Max Tegmark (37:57.840)
We have this tendency to very often want
Lex Fridman (37:59.520)
to say, as soon as machines can do something,
Max Tegmark (38:01.560)
we try to diminish it and say, oh, but that's
Lex Fridman (38:03.560)
not real intelligence.
Lex Fridman (38:05.920)
Isn't it creative or this or that?
Lex Fridman (38:08.400)
The other thing, if we ask ourselves
Max Tegmark (38:12.200)
to write down a definition of what we actually mean
Lex Fridman (38:14.320)
by being creative, what we mean by Andrew Wiles, what he did
Max Tegmark (38:18.840)
there, for example, don't we often mean that someone takes
Lex Fridman (38:21.880)
a very unexpected leap?
Max Tegmark (38:26.000)
It's not like taking 573 and multiplying it
Lex Fridman (38:29.680)
by 224 by just a step of straightforward cookbook
Lex Fridman (38:33.840)
like rules, right?
Lex Fridman (38:36.520)
You can maybe make a connection between two things
Max Tegmark (38:39.680)
that people had never thought was connected or something
Lex Fridman (38:42.640)
like that.
Max Tegmark (38:44.480)
I think this is an aspect of intelligence.
Lex Fridman (38:47.720)
And this is actually one of the most important aspects of it.
Max Tegmark (38:53.000)
Maybe the reason we humans tend to be better at it
Lex Fridman (38:55.520)
than traditional computers is because it's
Max Tegmark (38:57.840)
something that comes more naturally if you're
Lex Fridman (38:59.640)
a neural network than if you're a traditional logic gate
Max Tegmark (39:04.120)
based computer machine.
Lex Fridman (39:05.720)
We physically have all these connections.
Lex Fridman (39:08.640)
And you activate here, activate here, activate here.
Lex Fridman (39:13.800)
Bing.
Max Tegmark (39:16.560)
My hunch is that if we ever build a machine where you could
Lex Fridman (39:21.040)
just give it the task, hey, you say, hey, I just realized
Max Tegmark (39:29.200)
I want to travel around the world instead this month.
Lex Fridman (39:32.320)
Can you teach my AGI course for me?
Lex Fridman (39:34.600)
And it's like, OK, I'll do it.
Lex Fridman (39:35.960)
And it does everything that you would have done
Lex Fridman (39:37.920)
and improvises and stuff.
Lex Fridman (39:39.760)
That would, in my mind, involve a lot of creativity.
Max Tegmark (39:43.360)
Yeah, so it's actually a beautiful way to put it.
Lex Fridman (39:45.680)
I think we do try to grasp at the definition of intelligence
Max Tegmark (39:52.640)
is everything we don't understand how to build.
Lex Fridman (39:56.360)
So we as humans try to find things
Max Tegmark (39:59.360)
that we have and machines don't have.
Lex Fridman (40:01.240)
And maybe creativity is just one of the things, one
Max Tegmark (40:03.800)
of the words we use to describe that.
Lex Fridman (40:05.480)
That's a really interesting way to put it.
Max Tegmark (40:07.200)
I don't think we need to be that defensive.
Lex Fridman (40:09.520)
I don't think anything good comes out of saying,
Lex Fridman (40:11.560)
well, we're somehow special, you know?
Lex Fridman (40:18.080)
Contrary wise, there are many examples in history
Max Tegmark (40:21.040)
of where trying to pretend that we're somehow superior
Lex Fridman (40:27.840)
to all other intelligent beings has led to pretty bad results,
Lex Fridman (40:33.120)
right?
Lex Fridman (40:35.960)
Nazi Germany, they said that they were somehow superior
Max Tegmark (40:38.440)
to other people.
Lex Fridman (40:40.080)
Today, we still do a lot of cruelty to animals
Max Tegmark (40:42.440)
by saying that we're so superior somehow,
Lex Fridman (40:44.440)
and they can't feel pain.
Max Tegmark (40:46.440)
Slavery was justified by the same kind
Lex Fridman (40:48.480)
of just really weak arguments.
Lex Fridman (40:52.200)
And I don't think if we actually go ahead and build
Lex Fridman (40:57.120)
artificial general intelligence, it
Max Tegmark (40:59.440)
can do things better than us, I don't
Lex Fridman (41:01.360)
think we should try to found our self worth on some sort
Max Tegmark (41:04.080)
of bogus claims of superiority in terms
Lex Fridman (41:09.760)
of our intelligence.
Max Tegmark (41:12.120)
I think we should instead find our calling
Lex Fridman (41:18.080)
and the meaning of life from the experiences that we have.
Max Tegmark (41:23.360)
I can have very meaningful experiences
Lex Fridman (41:28.720)
even if there are other people who are smarter than me.
Max Tegmark (41:32.920)
When I go to a faculty meeting here,
Lex Fridman (41:34.400)
and we talk about something, and then I certainly realize,
Max Tegmark (41:36.520)
oh, boy, he has an old prize, he has an old prize,
Lex Fridman (41:39.080)
he has an old prize, I don't have one.
Max Tegmark (41:40.800)
Does that make me enjoy life any less
Lex Fridman (41:43.760)
or enjoy talking to those people less?
Max Tegmark (41:47.560)
Of course not.
Lex Fridman (41:49.560)
And the contrary, I feel very honored and privileged
Max Tegmark (41:54.160)
to get to interact with other very intelligent beings that
Lex Fridman (41:58.760)
are better than me at a lot of stuff.
Lex Fridman (42:00.680)
So I don't think there's any reason why
Lex Fridman (42:02.840)
we can't have the same approach with intelligent machines.
Max Tegmark (42:06.080)
That's a really interesting.
Lex Fridman (42:07.320)
So people don't often think about that.
Max Tegmark (42:08.920)
They think about when there's going,
Lex Fridman (42:10.600)
if there's machines that are more intelligent,
Max Tegmark (42:13.320)
you naturally think that that's not
Lex Fridman (42:15.080)
going to be a beneficial type of intelligence.
Max Tegmark (42:19.080)
You don't realize it could be like peers with Nobel prizes
Lex Fridman (42:23.000)
that would be just fun to talk with,
Lex Fridman (42:25.120)
and they might be clever about certain topics,
Lex Fridman (42:27.560)
and you can have fun having a few drinks with them.
Max Tegmark (42:32.240)
Well, also, another example we can all
Lex Fridman (42:35.880)
relate to of why it doesn't have to be a terrible thing
Max Tegmark (42:39.320)
to be in the presence of people who are even smarter than us
Lex Fridman (42:42.560)
all around is when you and I were both two years old,
Max Tegmark (42:45.600)
I mean, our parents were much more intelligent than us,
Lex Fridman (42:48.360)
right?
Max Tegmark (42:49.040)
Worked out OK, because their goals
Lex Fridman (42:51.960)
were aligned with our goals.
Lex Fridman (42:53.960)
And that, I think, is really the number one key issue
Lex Fridman (42:58.680)
we have to solve if we value align the value alignment
Max Tegmark (43:02.280)
problem, exactly.
Lex Fridman (43:03.080)
Because people who see too many Hollywood movies
Max Tegmark (43:06.520)
with lousy science fiction plot lines,
Lex Fridman (43:10.000)
they worry about the wrong thing, right?
Max Tegmark (43:12.200)
They worry about some machine suddenly turning evil.
Lex Fridman (43:16.320)
It's not malice that is the concern.
Max Tegmark (43:21.480)
It's competence.
Lex Fridman (43:22.880)
By definition, intelligent makes you very competent.
Max Tegmark (43:27.440)
If you have a more intelligent goal playing,
Lex Fridman (43:31.920)
computer playing is a less intelligent one.
Lex Fridman (43:33.680)
And when we define intelligence as the ability
Lex Fridman (43:36.120)
to accomplish goal winning, it's going
Max Tegmark (43:38.600)
to be the more intelligent one that wins.
Lex Fridman (43:40.560)
And if you have a human and then you
Max Tegmark (43:43.560)
have an AGI that's more intelligent in all ways
Lex Fridman (43:47.720)
and they have different goals, guess who's
Lex Fridman (43:49.520)
going to get their way, right?
Lex Fridman (43:50.720)
So I was just reading about this particular rhinoceros species
Max Tegmark (43:57.120)
that was driven extinct just a few years ago.
Lex Fridman (43:59.200)
Ellen Bummer is looking at this cute picture of a mommy
Max Tegmark (44:02.280)
rhinoceros with its child.
Lex Fridman (44:05.080)
And why did we humans drive it to extinction?
Max Tegmark (44:09.320)
It wasn't because we were evil rhino haters as a whole.
Lex Fridman (44:12.800)
It was just because our goals weren't aligned
Max Tegmark (44:14.920)
with those of the rhinoceros.
Lex Fridman (44:16.000)
And it didn't work out so well for the rhinoceros
Lex Fridman (44:17.680)
because we were more intelligent, right?
Lex Fridman (44:19.560)
So I think it's just so important
Max Tegmark (44:21.240)
that if we ever do build AGI, before we unleash anything,
Lex Fridman (44:27.120)
we have to make sure that it learns
Max Tegmark (44:31.840)
to understand our goals, that it adopts our goals,
Lex Fridman (44:36.000)
and that it retains those goals.
Lex Fridman (44:37.920)
So the cool, interesting problem there
Lex Fridman (44:40.520)
is us as human beings trying to formulate our values.
Lex Fridman (44:47.040)
So you could think of the United States Constitution as a way
Lex Fridman (44:51.360)
that people sat down, at the time a bunch of white men,
Max Tegmark (44:56.680)
which is a good example, I should say.
Lex Fridman (44:59.680)
They formulated the goals for this country.
Lex Fridman (45:01.480)
And a lot of people agree that those goals actually
Lex Fridman (45:03.760)
held up pretty well.
Max Tegmark (45:05.360)
That's an interesting formulation of values
Lex Fridman (45:07.160)
and failed miserably in other ways.
Lex Fridman (45:09.440)
So for the value alignment problem and the solution to it,
Lex Fridman (45:13.320)
we have to be able to put on paper or in a program
Max Tegmark (45:19.560)
human values.
Lex Fridman (45:20.400)
How difficult do you think that is?
Max Tegmark (45:22.400)
Very.
Lex Fridman (45:24.040)
But it's so important.
Max Tegmark (45:25.880)
We really have to give it our best.
Lex Fridman (45:28.000)
And it's difficult for two separate reasons.
Max Tegmark (45:30.120)
There's the technical value alignment problem
Lex Fridman (45:33.440)
of figuring out just how to make machines understand our goals,
Max Tegmark (45:39.120)
adopt them, and retain them.
Lex Fridman (45:40.440)
And then there's the separate part of it,
Max Tegmark (45:43.200)
the philosophical part.
Lex Fridman (45:44.200)
Whose values anyway?
Lex Fridman (45:45.920)
And since it's not like we have any great consensus
Lex Fridman (45:48.320)
on this planet on values, what mechanism should we
Max Tegmark (45:52.040)
create then to aggregate and decide, OK,
Lex Fridman (45:54.120)
what's a good compromise?
Max Tegmark (45:56.520)
That second discussion can't just
Lex Fridman (45:58.440)
be left to tech nerds like myself.
Lex Fridman (46:01.560)
And if we refuse to talk about it and then AGI gets built,
Lex Fridman (46:05.720)
who's going to be actually making
Lex Fridman (46:07.160)
the decision about whose values?
Lex Fridman (46:08.480)
It's going to be a bunch of dudes in some tech company.
Lex Fridman (46:12.080)
And are they necessarily so representative of all
Lex Fridman (46:17.240)
of humankind that we want to just entrust it to them?
Max Tegmark (46:19.400)
Are they even uniquely qualified to speak
Lex Fridman (46:23.000)
to future human happiness just because they're
Lex Fridman (46:25.240)
good at programming AI?
Lex Fridman (46:26.480)
I'd much rather have this be a really inclusive conversation.
Lex Fridman (46:30.200)
But do you think it's possible?
Lex Fridman (46:32.560)
So you create a beautiful vision that includes the diversity,
Max Tegmark (46:37.560)
cultural diversity, and various perspectives on discussing
Lex Fridman (46:40.960)
rights, freedoms, human dignity.
Lex Fridman (46:43.600)
But how hard is it to come to that consensus?
Lex Fridman (46:46.520)
Do you think it's certainly a really important thing
Lex Fridman (46:50.400)
that we should all try to do?
Lex Fridman (46:51.880)
But do you think it's feasible?
Max Tegmark (46:54.240)
I think there's no better way to guarantee failure than to
Lex Fridman (47:00.160)
refuse to talk about it or refuse to try.
Lex Fridman (47:02.840)
And I also think it's a really bad strategy
Lex Fridman (47:05.320)
to say, OK, let's first have a discussion for a long time.
Lex Fridman (47:08.560)
And then once we reach complete consensus,
Lex Fridman (47:11.040)
then we'll try to load it into some machine.
Max Tegmark (47:13.360)
No, we shouldn't let perfect be the enemy of good.
Lex Fridman (47:16.560)
Instead, we should start with the kindergarten ethics
Max Tegmark (47:20.600)
that pretty much everybody agrees on
Lex Fridman (47:22.120)
and put that into machines now.
Max Tegmark (47:24.360)
We're not doing that even.
Lex Fridman (47:25.880)
Look at anyone who builds this passenger aircraft,
Max Tegmark (47:31.000)
wants it to never under any circumstances
Lex Fridman (47:33.000)
fly into a building or a mountain.
Max Tegmark (47:35.600)
Yet the September 11 hijackers were able to do that.
Lex Fridman (47:38.480)
And even more embarrassingly, Andreas Lubitz,
Max Tegmark (47:41.800)
this depressed Germanwings pilot,
Lex Fridman (47:43.960)
when he flew his passenger jet into the Alps killing over 100
Max Tegmark (47:47.360)
people, he just told the autopilot to do it.
Lex Fridman (47:50.640)
He told the freaking computer to change the altitude
Max Tegmark (47:53.200)
to 100 meters.
Lex Fridman (47:55.040)
And even though it had the GPS maps, everything,
Max Tegmark (47:58.160)
the computer was like, OK.
Lex Fridman (48:00.640)
So we should take those very basic values,
Max Tegmark (48:05.320)
where the problem is not that we don't agree.
Lex Fridman (48:08.400)
The problem is just we've been too lazy
Max Tegmark (48:10.120)
to try to put it into our machines
Lex Fridman (48:11.480)
and make sure that from now on, airplanes will just,
Max Tegmark (48:15.520)
which all have computers in them,
Lex Fridman (48:16.920)
but will just refuse to do something like that.
Max Tegmark (48:19.720)
Go into safe mode, maybe lock the cockpit door,
Lex Fridman (48:22.160)
go over to the nearest airport.
Lex Fridman (48:24.480)
And there's so much other technology in our world
Lex Fridman (48:28.080)
as well now, where it's really becoming quite timely
Max Tegmark (48:31.320)
to put in some sort of very basic values like this.
Lex Fridman (48:34.120)
Even in cars, we've had enough vehicle terrorism attacks
Max Tegmark (48:39.240)
by now, where people have driven trucks and vans
Lex Fridman (48:42.040)
into pedestrians, that it's not at all a crazy idea
Max Tegmark (48:45.480)
to just have that hardwired into the car.
Lex Fridman (48:48.680)
Because yeah, there are a lot of,
Max Tegmark (48:50.280)
there's always going to be people who for some reason
Lex Fridman (48:52.240)
want to harm others, but most of those people
Max Tegmark (48:54.800)
don't have the technical expertise to figure out
Lex Fridman (48:56.760)
how to work around something like that.
Lex Fridman (48:58.520)
So if the car just won't do it, it helps.
Lex Fridman (49:01.760)
So let's start there.
Lex Fridman (49:02.840)
So there's a lot of, that's a great point.
Lex Fridman (49:04.960)
So not chasing perfect.
Max Tegmark (49:06.800)
There's a lot of things that most of the world agrees on.
Lex Fridman (49:10.840)
Yeah, let's start there.
Max Tegmark (49:11.840)
Let's start there.
Lex Fridman (49:12.680)
And then once we start there,
Max Tegmark (49:14.560)
we'll also get into the habit of having
Lex Fridman (49:17.240)
these kind of conversations about, okay,
Lex Fridman (49:18.520)
what else should we put in here and have these discussions?
Lex Fridman (49:21.760)
This should be a gradual process then.
Max Tegmark (49:23.920)
Great, so, but that also means describing these things
Lex Fridman (49:28.600)
and describing it to a machine.
Lex Fridman (49:31.240)
So one thing, we had a few conversations
Lex Fridman (49:34.200)
with Stephen Wolfram.
Max Tegmark (49:35.640)
I'm not sure if you're familiar with Stephen.
Lex Fridman (49:37.080)
Oh yeah, I know him quite well.
Lex Fridman (49:38.360)
So he is, he works with a bunch of things,
Lex Fridman (49:42.040)
but cellular automata, these simple computable things,
Max Tegmark (49:46.560)
these computation systems.
Lex Fridman (49:47.960)
And he kind of mentioned that,
Max Tegmark (49:49.880)
we probably have already within these systems
Lex Fridman (49:52.480)
already something that's AGI,
Max Tegmark (49:56.120)
meaning like we just don't know it
Lex Fridman (49:58.720)
because we can't talk to it.
Lex Fridman (50:00.400)
So if you give me this chance to try to at least
Lex Fridman (50:04.800)
form a question out of this is,
Max Tegmark (50:07.600)
I think it's an interesting idea to think
Lex Fridman (50:10.880)
that we can have intelligent systems,
Lex Fridman (50:12.680)
but we don't know how to describe something to them
Lex Fridman (50:15.600)
and they can't communicate with us.
Max Tegmark (50:17.360)
I know you're doing a little bit of work in explainable AI,
Lex Fridman (50:19.840)
trying to get AI to explain itself.
Lex Fridman (50:22.040)
So what are your thoughts of natural language processing
Lex Fridman (50:25.520)
or some kind of other communication?
Lex Fridman (50:27.640)
How does the AI explain something to us?
Lex Fridman (50:30.120)
How do we explain something to it, to machines?
Lex Fridman (50:33.640)
Or you think of it differently?
Lex Fridman (50:35.320)
So there are two separate parts to your question there.
Max Tegmark (50:39.960)
One of them has to do with communication,
Lex Fridman (50:42.440)
which is super interesting, I'll get to that in a sec.
Max Tegmark (50:44.440)
The other is whether we already have AGI
Lex Fridman (50:47.280)
but we just haven't noticed it there.
Max Tegmark (50:49.240)
Right.
Lex Fridman (50:51.800)
There I beg to differ.
Max Tegmark (50:54.280)
I don't think there's anything in any cellular automaton
Lex Fridman (50:56.480)
or anything or the internet itself or whatever
Max Tegmark (50:59.040)
that has artificial general intelligence
Lex Fridman (51:03.560)
and that it can really do exactly everything
Max Tegmark (51:05.520)
we humans can do better.
Lex Fridman (51:07.000)
I think the day that happens, when that happens,
Max Tegmark (51:11.600)
we will very soon notice, we'll probably notice even before
Lex Fridman (51:15.600)
because in a very, very big way.
Lex Fridman (51:17.440)
But for the second part, though.
Lex Fridman (51:18.840)
Wait, can I ask, sorry.
Max Tegmark (51:20.720)
So, because you have this beautiful way
Lex Fridman (51:24.400)
to formulating consciousness as information processing,
Lex Fridman (51:30.360)
and you can think of intelligence
Lex Fridman (51:31.360)
as information processing,
Lex Fridman (51:32.280)
and you can think of the entire universe
Lex Fridman (51:34.320)
as these particles and these systems roaming around
Max Tegmark (51:38.720)
that have this information processing power.
Lex Fridman (51:41.360)
You don't think there is something with the power
Max Tegmark (51:44.840)
to process information in the way that we human beings do
Lex Fridman (51:49.040)
that's out there that needs to be sort of connected to.
Max Tegmark (51:55.400)
It seems a little bit philosophical, perhaps,
Lex Fridman (51:57.880)
but there's something compelling to the idea
Max Tegmark (52:00.080)
that the power is already there,
Lex Fridman (52:01.920)
which the focus should be more on being able
Max Tegmark (52:05.440)
to communicate with it.
Lex Fridman (52:07.360)
Well, I agree that in a certain sense,
Max Tegmark (52:11.960)
the hardware processing power is already out there
Lex Fridman (52:15.360)
because our universe itself can think of it
Lex Fridman (52:19.000)
as being a computer already, right?
Lex Fridman (52:21.000)
It's constantly computing what water waves,
Lex Fridman (52:23.800)
how it devolved the water waves in the River Charles
Lex Fridman (52:26.120)
and how to move the air molecules around.
Max Tegmark (52:28.440)
Seth Lloyd has pointed out, my colleague here,
Lex Fridman (52:30.480)
that you can even in a very rigorous way
Max Tegmark (52:32.920)
think of our entire universe as being a quantum computer.
Lex Fridman (52:35.480)
It's pretty clear that our universe
Max Tegmark (52:37.680)
supports this amazing processing power
Lex Fridman (52:40.320)
because you can even,
Lex Fridman (52:42.160)
within this physics computer that we live in, right?
Lex Fridman (52:44.920)
We can even build actual laptops and stuff,
Lex Fridman (52:47.040)
so clearly the power is there.
Lex Fridman (52:49.000)
It's just that most of the compute power that nature has,
Max Tegmark (52:52.040)
it's, in my opinion, kind of wasting on boring stuff
Lex Fridman (52:54.240)
like simulating yet another ocean wave somewhere
Lex Fridman (52:56.520)
where no one is even looking, right?
Lex Fridman (52:58.040)
So in a sense, what life does, what we are doing
Max Tegmark (53:00.880)
when we build computers is we're rechanneling
Lex Fridman (53:03.880)
all this compute that nature is doing anyway
Max Tegmark (53:07.200)
into doing things that are more interesting
Lex Fridman (53:09.360)
than just yet another ocean wave,
Lex Fridman (53:11.440)
and let's do something cool here.
Lex Fridman (53:14.080)
So the raw hardware power is there, for sure,
Lex Fridman (53:17.080)
but then even just computing what's going to happen
Lex Fridman (53:21.080)
for the next five seconds in this water bottle,
Max Tegmark (53:23.520)
takes a ridiculous amount of compute
Lex Fridman (53:26.000)
if you do it on a human computer.
Max Tegmark (53:27.920)
This water bottle just did it.
Lex Fridman (53:29.920)
But that does not mean that this water bottle has AGI
Max Tegmark (53:34.760)
because AGI means it should also be able to,
Lex Fridman (53:37.040)
like I've written my book, done this interview.
Lex Fridman (53:40.160)
And I don't think it's just communication problems.
Lex Fridman (53:42.080)
I don't really think it can do it.
Max Tegmark (53:46.760)
Although Buddhists say when they watch the water
Lex Fridman (53:49.280)
and that there is some beauty,
Max Tegmark (53:51.240)
that there's some depth and beauty in nature
Lex Fridman (53:53.720)
that they can communicate with.
Max Tegmark (53:54.840)
Communication is also very important though
Lex Fridman (53:56.480)
because I mean, look, part of my job is being a teacher.
Lex Fridman (54:01.200)
And I know some very intelligent professors even
Lex Fridman (54:06.200)
who just have a bit of hard time communicating.
Max Tegmark (54:09.800)
They come up with all these brilliant ideas,
Lex Fridman (54:12.640)
but to communicate with somebody else,
Max Tegmark (54:14.520)
you have to also be able to simulate their own mind.
Lex Fridman (54:16.920)
Yes, empathy.
Max Tegmark (54:18.360)
Build well enough and understand model of their mind
Lex Fridman (54:20.640)
that you can say things that they will understand.
Lex Fridman (54:24.400)
And that's quite difficult.
Lex Fridman (54:26.480)
And that's why today it's so frustrating
Max Tegmark (54:28.280)
if you have a computer that makes some cancer diagnosis
Lex Fridman (54:32.600)
and you ask it, well, why are you saying
Lex Fridman (54:34.120)
I should have this surgery?
Lex Fridman (54:36.120)
And if it can only reply,
Max Tegmark (54:37.960)
I was trained on five terabytes of data
Lex Fridman (54:40.800)
and this is my diagnosis, boop, boop, beep, beep.
Lex Fridman (54:45.080)
It doesn't really instill a lot of confidence, right?
Lex Fridman (54:49.120)
So I think we have a lot of work to do
Max Tegmark (54:51.120)
on communication there.
Lex Fridman (54:54.320)
So what kind of, I think you're doing a little bit of work
Max Tegmark (54:58.040)
in explainable AI.
Lex Fridman (54:59.320)
What do you think are the most promising avenues?
Max Tegmark (55:01.320)
Is it mostly about sort of the Alexa problem
Lex Fridman (55:05.240)
of natural language processing of being able
Max Tegmark (55:07.200)
to actually use human interpretable methods
Lex Fridman (55:11.600)
of communication?
Lex Fridman (55:13.160)
So being able to talk to a system and it talk back to you,
Lex Fridman (55:16.000)
or is there some more fundamental problems to be solved?
Max Tegmark (55:18.640)
I think it's all of the above.
Lex Fridman (55:21.160)
The natural language processing is obviously important,
Lex Fridman (55:23.520)
but there are also more nerdy fundamental problems.
Lex Fridman (55:27.600)
Like if you take, you play chess?
Max Tegmark (55:31.640)
Of course, I'm Russian.
Lex Fridman (55:33.040)
I have to.
Lex Fridman (55:33.880)
You speak Russian?
Lex Fridman (55:34.720)
Yes, I speak Russian.
Max Tegmark (55:35.560)
Excellent, I didn't know.
Lex Fridman (55:38.040)
When did you learn Russian?
Max Tegmark (55:39.160)
I speak very bad Russian, I'm only an autodidact,
Lex Fridman (55:41.800)
but I bought a book, Teach Yourself Russian,
Max Tegmark (55:44.560)
read a lot, but it was very difficult.
Lex Fridman (55:47.720)
Wow.
Max Tegmark (55:48.560)
That's why I speak so bad.
Lex Fridman (55:49.960)
How many languages do you know?
Max Tegmark (55:51.960)
Wow, that's really impressive.
Lex Fridman (55:53.840)
I don't know, my wife has some calculation,
Lex Fridman (55:56.320)
but my point was, if you play chess,
Lex Fridman (55:58.400)
have you looked at the AlphaZero games?
Max Tegmark (56:01.040)
The actual games, no.
Lex Fridman (56:02.600)
Check it out, some of them are just mind blowing,
Max Tegmark (56:06.320)
really beautiful.
Lex Fridman (56:07.720)
And if you ask, how did it do that?
Max Tegmark (56:13.760)
You go talk to Demis Hassabis,
Lex Fridman (56:16.520)
I know others from DeepMind,
Max Tegmark (56:19.120)
all they'll ultimately be able to give you
Lex Fridman (56:20.600)
is big tables of numbers, matrices,
Max Tegmark (56:23.920)
that define the neural network.
Lex Fridman (56:25.720)
And you can stare at these tables of numbers
Max Tegmark (56:28.080)
till your face turn blue,
Lex Fridman (56:29.600)
and you're not gonna understand much
Max Tegmark (56:32.520)
about why it made that move.
Lex Fridman (56:34.520)
And even if you have natural language processing
Max Tegmark (56:37.640)
that can tell you in human language about,
Lex Fridman (56:40.280)
oh, five, seven, points, two, eight,
Max Tegmark (56:42.520)
still not gonna really help.
Lex Fridman (56:43.560)
So I think there's a whole spectrum of fun challenges
Max Tegmark (56:47.480)
that are involved in taking a computation
Lex Fridman (56:50.520)
that does intelligent things
Lex Fridman (56:52.240)
and transforming it into something equally good,
Lex Fridman (56:57.760)
equally intelligent, but that's more understandable.
Lex Fridman (57:01.840)
And I think that's really valuable
Lex Fridman (57:03.240)
because I think as we put machines in charge
Max Tegmark (57:07.440)
of ever more infrastructure in our world,
Lex Fridman (57:09.760)
the power grid, the trading on the stock market,
Max Tegmark (57:12.680)
weapon systems and so on,
Lex Fridman (57:14.320)
it's absolutely crucial that we can trust
Max Tegmark (57:17.760)
these AIs to do all we want.
Lex Fridman (57:19.400)
And trust really comes from understanding
Max Tegmark (57:22.520)
in a very fundamental way.
Lex Fridman (57:24.400)
And that's why I'm working on this,
Max Tegmark (57:27.560)
because I think the more,
Lex Fridman (57:29.160)
if we're gonna have some hope of ensuring
Max Tegmark (57:31.840)
that machines have adopted our goals
Lex Fridman (57:33.520)
and that they're gonna retain them,
Max Tegmark (57:35.800)
that kind of trust, I think,
Lex Fridman (57:38.800)
needs to be based on things you can actually understand,
Max Tegmark (57:41.200)
preferably even improve theorems on.
Lex Fridman (57:44.240)
Even with a self driving car, right?
Max Tegmark (57:47.040)
If someone just tells you it's been trained
Lex Fridman (57:48.680)
on tons of data and it never crashed,
Max Tegmark (57:50.640)
it's less reassuring than if someone actually has a proof.
Lex Fridman (57:54.200)
Maybe it's a computer verified proof,
Lex Fridman (57:55.960)
but still it says that under no circumstances
Lex Fridman (57:58.800)
is this car just gonna swerve into oncoming traffic.
Lex Fridman (58:02.320)
And that kind of information helps to build trust
Lex Fridman (58:04.640)
and helps build the alignment of goals,
Max Tegmark (58:09.400)
at least awareness that your goals, your values are aligned.
Lex Fridman (58:12.200)
And I think even in the very short term,
Lex Fridman (58:13.840)
if you look at how, you know, today, right?
Lex Fridman (58:16.360)
This absolutely pathetic state of cybersecurity
Lex Fridman (58:19.320)
that we have, where is it?
Lex Fridman (58:21.720)
Three billion Yahoo accounts we can't pack,
Max Tegmark (58:27.200)
almost every American's credit card and so on.
Lex Fridman (58:32.800)
Why is this happening?
Max Tegmark (58:34.120)
It's ultimately happening because we have software
Lex Fridman (58:37.960)
that nobody fully understood how it worked.
Lex Fridman (58:41.200)
That's why the bugs hadn't been found, right?
Lex Fridman (58:44.800)
And I think AI can be used very effectively
Max Tegmark (58:47.480)
for offense, for hacking,
Lex Fridman (58:49.640)
but it can also be used for defense.
Max Tegmark (58:52.320)
Hopefully automating verifiability
Lex Fridman (58:55.360)
and creating systems that are built in different ways
Lex Fridman (59:00.680)
so you can actually prove things about them.
Lex Fridman (59:02.920)
And it's important.
Lex Fridman (59:05.240)
So speaking of software that nobody understands
Lex Fridman (59:07.680)
how it works, of course, a bunch of people ask
Max Tegmark (59:10.640)
about your paper, about your thoughts
Lex Fridman (59:12.160)
of why does deep and cheap learning work so well?
Max Tegmark (59:14.680)
That's the paper.
Lex Fridman (59:15.520)
But what are your thoughts on deep learning?
Max Tegmark (59:18.320)
These kind of simplified models of our own brains
Lex Fridman (59:21.880)
have been able to do some successful perception work,
Max Tegmark (59:26.440)
pattern recognition work, and now with AlphaZero and so on,
Lex Fridman (59:29.560)
do some clever things.
Lex Fridman (59:30.880)
What are your thoughts about the promise limitations
Lex Fridman (59:33.880)
of this piece?
Max Tegmark (59:35.680)
Great, I think there are a number of very important insights,
Lex Fridman (59:43.080)
very important lessons we can always draw
Max Tegmark (59:44.640)
from these kinds of successes.
Lex Fridman (59:47.120)
One of them is when you look at the human brain,
Max Tegmark (59:48.960)
you see it's very complicated, 10th of 11 neurons,
Lex Fridman (59:51.480)
and there are all these different kinds of neurons
Lex Fridman (59:53.320)
and yada, yada, and there's been this long debate
Lex Fridman (59:55.040)
about whether the fact that we have dozens
Max Tegmark (59:57.200)
of different kinds is actually necessary for intelligence.
🔗 相关节目