Douglas Lenat: Cyc and the Quest to Solve Common Sense Reasoning in AI
技术与编程音乐与艺术心理与人性AI 与机器学习生物与进化
📋 章节目录
暂无章节信息
🔑 关键词
knowledgepsychgoingabledonhumansaidlanguagedoingtrueintelligencehelppersongeneralhumansorderimportantrepresenthavinglearning
💬 精彩语录
"additional bumps on this log. The only way to get there is to think about the hard problems and think"
该日志上有额外的凹凸。实现这一目标的唯一方法就是思考困难问题并思考
— Douglas Lenat （1:43:51.520）
"you've taught it something because it used to make this mistake and now it doesn't and so on. So this"
你已经教了它一些东西，因为它曾经犯过这个错误，但现在不会了，等等。所以这个
— Douglas Lenat （1:24:44.880）
"and freedoms and so on. Right now, we don't think twice about effectively enslaving our email systems"
和自由等等。现在，我们不假思索地有效地奴役我们的电子邮件系统
— Douglas Lenat （1:31:18.880）
"important it is to use an expressive representation language like we do this higher order logic rather"
重要的是使用表达性表示语言，就像我们做高阶逻辑一样
— Douglas Lenat （1:52:34.520）
"competitors that will pop up and start making you nervous and all that kind of stuff. So do you think"
突然出现的竞争对手会让你感到紧张之类的。那么你认为
— Douglas Lenat （1:57:59.280）
🎙️ 完整对话（1690 条）
Lex Fridman (00:00.000)
The following is a conversation with Doug Lenit, creator of Psych, a system that for close to 40
以下是与 Psych 的创建者 Doug Lenit 的对话，该系统为近 40 人提供了帮助
Lex Fridman (00:06.800)
years, and still today, has sought to solve the core problem of artificial intelligence,
多年来，直到今天，一直在寻求解决人工智能的核心问题，
Lex Fridman (00:12.880)
the acquisition of common sense knowledge and the use of that knowledge to think,
获取常识知识并利用这些知识进行思考，
Lex Fridman (00:18.000)
to reason, and to understand the world. To support this podcast, please check out our sponsors in
去推理，去理解世界。为了支持这个播客，请查看我们的赞助商
Lex Fridman (00:23.680)
the description. As a side note, let me say that in the excitement of the modern era of machine
描述。作为旁注，让我说，在现代机器时代的兴奋中
Douglas Lenat (00:29.440)
learning, it is easy to forget just how little we understand exactly how to build the kind of
在学习过程中，我们很容易忘记我们对如何构建这种类型的了解是多么的少。
Douglas Lenat (00:36.000)
intelligence that matches the power of the human mind. To me, many of the core ideas behind Psych,
与人类心灵的力量相匹配的智慧。对我来说，Psych 背后的许多核心思想，
Douglas Lenat (00:42.480)
in some form, in actuality or in spirit, will likely be part of the AI system that achieves
以某种形式，无论是在现实中还是在精神上，都可能成为人工智能系统的一部分，从而实现
Douglas Lenat (00:48.880)
general superintelligence. But perhaps more importantly, solving this problem of common
一般超级智能。但也许更重要的是，解决这个常见问题
Douglas Lenat (00:54.480)
sense knowledge will help us humans understand our own minds, the nature of truth, and finally,
感官知识将帮助我们人类了解我们自己的思想、真理的本质，最后，
Lex Fridman (01:01.040)
how to be more rational and more kind to each other. This is the Lex Friedman podcast,
如何更加理性、更加友善地对待彼此。这是莱克斯·弗里德曼的播客，
Lex Fridman (01:07.280)
and here is my conversation with Doug Lenit. Psych is a project launched by you in 1984,
这是我和道格·莱尼特的对话。 Psych是您在1984年发起的一个项目，
Lex Fridman (01:16.080)
and still is active today, whose goal is to assemble a knowledge base that spans the basic
至今仍然活跃，其目标是构建一个涵盖基本知识的知识库
Douglas Lenat (01:20.880)
concepts and rules about how the world works. In other words, it hopes to capture common sense
关于世界如何运作的概念和规则。换句话说，它希望捕捉常识
Douglas Lenat (01:26.720)
knowledge, which is a lot harder than it sounds. Can you elaborate on this mission and maybe
知识，这比听起来要困难得多。你能详细说明一下这个任务吗？也许
Douglas Lenat (01:32.320)
perhaps speak to the various subgoals within this mission? When I was a faculty member in the
或许可以谈谈这个任务中的各个子目标？当我还是该校的一名教员时
Douglas Lenat (01:39.520)
computer science department at Stanford, my colleagues and I did research in all sorts of
在斯坦福大学计算机科学系，我和我的同事进行了各种研究
Douglas Lenat (01:46.640)
artificial intelligence programs, so natural language understanding programs, robots,
人工智能程序，自然语言理解程序，机器人，
Douglas Lenat (01:53.440)
expert systems, and so on. And we kept hitting the very same brick wall. Our systems would have
专家系统等。我们一直碰着同一堵砖墙。我们的系统会有
Douglas Lenat (02:02.880)
impressive early successes. And so if your only goal was academic, namely to get enough material
令人印象深刻的早期成功。因此，如果您唯一的目标是学术，即获得足够的材料
Douglas Lenat (02:12.320)
to write a journal article, that might actually suffice. But if you're really trying to get AI,
Douglas Lenat (02:19.280)
then you have to somehow get past the brick wall. And the brick wall was
Douglas Lenat (02:23.600)
the programs didn't have what we would call common sense. They didn't have general world
Douglas Lenat (02:28.560)
knowledge. They didn't really understand what they were doing, what they were saying,
Lex Fridman (02:33.280)
what they were being asked. And so very much like a clever dog performing tricks,
Douglas Lenat (02:40.480)
we could get them to do tricks, but they never really understood what they were doing. Sort of
Douglas Lenat (02:44.880)
like when you get a dog to fetch your morning newspaper. The dog might do that successfully,
Lex Fridman (02:50.880)
but the dog has no idea what a newspaper is or what it says or anything like that.
Lex Fridman (02:55.520)
What does it mean to understand something? Can you maybe elaborate on that a little bit?
Douglas Lenat (02:59.760)
Is it is understanding action of like combining little things together like through inference,
Lex Fridman (03:05.840)
or is understanding the wisdom you gain over time that forms a knowledge?
Douglas Lenat (03:10.000)
I think of understanding more like the ground you stand on, which could be very shaky,
Douglas Lenat (03:20.800)
could be very unsafe, but most of the time is not because underneath it is more ground,
Lex Fridman (03:28.960)
and eventually rock and other things. But layer after layer after layer, that solid foundation
Douglas Lenat (03:36.160)
is there. And you rarely need to think about it, you rarely need to count on it, but occasionally
Douglas Lenat (03:41.760)
you do. And I've never used this analogy before, so bear with me. But I think the same thing is
Douglas Lenat (03:48.960)
true in terms of getting computers to understand things, which is you ask a computer a question,
Douglas Lenat (03:56.000)
for instance, Alexa or some robot or something, and maybe it gets the right answer.
Lex Fridman (04:02.160)
But if you were asking that of a human, you could also say things like, why? Or how might you be
Douglas Lenat (04:09.760)
wrong about this? Or something like that. And the person would answer you. And it might be a little
Douglas Lenat (04:17.040)
annoying if you have a small child and they keep asking why questions in series. Eventually,
Douglas Lenat (04:22.320)
you get to the point where you throw up your hands and say, I don't know, it's just the way
Douglas Lenat (04:25.920)
the world is. But for many layers, you actually have that layered, solid foundation of support,
Lex Fridman (04:35.280)
so that when you need it, you can count on it. And when do you need it? Well, when things are
Douglas Lenat (04:40.720)
unexpected, when you come up against a situation which is novel. For instance, when you're driving,
Douglas Lenat (04:46.320)
it may be fine to have a small program, a small set of rules that cover 99% of the cases, but that
Douglas Lenat (04:55.280)
1% of the time when something strange happens, you really need to draw on common sense. For instance,
Douglas Lenat (05:02.080)
my wife and I were driving recently and there was a trash truck in front of us. And I guess they had
Douglas Lenat (05:09.360)
packed it too full and the back exploded. And trash bags went everywhere. And we had to make
Douglas Lenat (05:17.520)
a split second decision. Are we going to slam on our brakes? Are we going to swerve into another
Douglas Lenat (05:21.760)
lane? Are we going to just run it over? Because there were cars all around us. And in front of us
Douglas Lenat (05:29.040)
was a large trash bag. And we know what we throw away in trash bags, probably not a safe thing to
Douglas Lenat (05:34.960)
run over. Over on the left was a bunch of fast food restaurant trash bags. And it's like,
Douglas Lenat (05:42.880)
oh, well, those things are just like styrofoam and leftover food. We'll run over that. And so that
Douglas Lenat (05:47.760)
was a safe thing for us to do. Now, that's the kind of thing that's going to happen maybe once
Douglas Lenat (05:52.800)
in your life. But the point is that there's almost no telling what little bits of knowledge about the
Douglas Lenat (06:01.680)
world you might actually need in some situations which were unforeseen. But see, when you sit on
Douglas Lenat (06:08.480)
that mountain or that ground that goes deep of knowledge in order to make a split second decision
Douglas Lenat (06:16.480)
about fast food trash or random trash from the back of a trash truck, you need to be able to
Douglas Lenat (06:26.080)
leverage that ground you stand on in some way. It's not merely, you know, it's not enough to just
Douglas Lenat (06:31.920)
have a lot of ground to stand on. It's your ability to leverage it, to utilize in a split,
Douglas Lenat (06:38.480)
like integrate it all together to make that split second decision. And I suppose understanding isn't
Douglas Lenat (06:45.840)
just having a common sense knowledge to access. It's the act of accessing it somehow, like
Douglas Lenat (06:55.040)
correctly filtering out the parts of the knowledge that are not useful, selecting only the useful
Douglas Lenat (07:02.560)
parts and effectively making conclusive decisions. So let's tease apart two different tasks really,
Douglas Lenat (07:10.640)
both of which are incredibly important and even necessary. If you're going to have this in a
Douglas Lenat (07:16.480)
useful, usable fashion as opposed to say like library books sitting on a shelf and so on, where
Douglas Lenat (07:25.120)
the knowledge might be there, but if a fire comes, the books are going to burn because they don't
Douglas Lenat (07:31.040)
know what's in them and they're just going to sit there while they burn. So there are two aspects of
Douglas Lenat (07:38.080)
using the knowledge. One is a kind of a theoretical, how is it possible at all? And then the second
Douglas Lenat (07:45.680)
aspect of what you said is, how can you do it quickly enough? So how can you do it at all is
Douglas Lenat (07:51.680)
something that philosophers have grappled with. And fortunately, philosophers 100 years ago and
Douglas Lenat (07:58.960)
even earlier developed a kind of formal language like English. It's called predicate logic or first
Douglas Lenat (08:10.240)
order logic or something like predicate calculus and so on. So there's a way of representing things
Douglas Lenat (08:17.280)
in this formal language which enables a mechanical procedure to sort of grind through
Lex Fridman (08:26.480)
and algorithmically produce all of the same logical entailments, all the same logical conclusions
Douglas Lenat (08:34.000)
that you or I would from that same set of pieces of information that are represented that way.
Lex Fridman (08:41.360)
So that sort of raises a couple questions. One is, how do you get all this information
Douglas Lenat (08:48.560)
from say observations and English and so on into this logical form? And secondly,
Lex Fridman (08:54.880)
how can you then efficiently run these algorithms to actually get the information you need?
Douglas Lenat (09:01.120)
In the case I mentioned in a 10th of a second rather than say in 10 hours or 10,000 years
Douglas Lenat (09:08.480)
of computation. And those are both really important questions. And like a corollary
Douglas Lenat (09:15.600)
addition to the first one is, how many such things do you need to gather for it to be useful
Douglas Lenat (09:22.400)
in certain contexts? So like what, in order, you mentioned philosophers, in order to capture this
Douglas Lenat (09:28.640)
world and represent it in a logical way and with a form of logic, like how many statements are
Douglas Lenat (09:36.400)
required? Is it five? Is it 10? Is it 10 trillion? Is it like that? That's as far as I understand is
Douglas Lenat (09:43.600)
probably still an open question. It may forever be an open question just to say like definitively
Lex Fridman (09:50.320)
about, to describe the universe perfectly. How many facts do you need?
Douglas Lenat (09:54.800)
I guess I'm going to disappoint you by giving you an actual answer to your question.
Douglas Lenat (10:00.160)
Okay. Well, no, this sounds exciting.
Lex Fridman (10:03.280)
Yes. Okay. So now we have like three things to talk about.
Douglas Lenat (10:09.600)
We'll keep adding more.
Douglas Lenat (10:10.560)
Although it's okay. The first and the third are related. So let's leave the efficiency
Lex Fridman (10:16.000)
question aside for now. So how does all this information get represented in logical form?
Lex Fridman (10:24.480)
So that these algorithms, resolution theorem proving and other algorithms can actually grind
Douglas Lenat (10:30.880)
through all the logical consequences of what you said. And that ties into your question about how
Douglas Lenat (10:37.040)
many of these things do you need? Because if the answer is small enough, then by hand, you could
Douglas Lenat (10:43.760)
write them out one at a time. So in the early 1984, I held a meeting at Stanford where I was a
Douglas Lenat (10:57.920)
faculty member there, where we assembled about half a dozen of the smartest people I know.
Douglas Lenat (11:05.280)
People like Alan Newell and Marvin Minsky and Alan Kay and a few others.
Lex Fridman (11:15.040)
Was Feynman there by chance? Because he commented about your system,
Douglas Lenat (11:19.120)
Eurisco, at the time.
Lex Fridman (11:20.480)
No, he wasn't part of this meeting.
Douglas Lenat (11:23.440)
That's a heck of a meeting anyway.
Douglas Lenat (11:25.120)
I think Ed Feigenbaum was there. I think Josh Lederberg was there. So we have all these different
Douglas Lenat (11:32.640)
smart people. And we came together to address the question that you raised, which is, if it's
Douglas Lenat (11:41.200)
important to represent common sense knowledge and world knowledge in order for AIs to not be
Douglas Lenat (11:46.640)
brittle, in order for AIs not to just have the veneer of intelligence, well, how many pieces
Douglas Lenat (11:53.840)
of common sense, how many if then rules, for instance, would we have to actually write in
Lex Fridman (11:59.520)
order to essentially cover what people expect perfect strangers to already know about the world?
Lex Fridman (12:07.120)
And I expected there would be an enormous divergence of opinion and computation. But
Douglas Lenat (12:14.960)
amazingly, everyone got an answer which was around a million. And one person got the answer
Douglas Lenat (12:23.280)
by saying, well, look, you can only burn into human long term memory a certain number of things
Douglas Lenat (12:30.800)
per unit time, like maybe one every 30 seconds or something. And other than that, it's just short
Douglas Lenat (12:36.080)
term memory and it flows away like water and so on. So by the time you're, say, 10 years old or so,
Lex Fridman (12:42.480)
how many things could you possibly have burned into your long term memory? And it's like about
Douglas Lenat (12:47.120)
a million. Another person went in a completely different direction and said, well, if you look
Douglas Lenat (12:52.560)
at the number of words in a dictionary, not a whole dictionary, but for someone to essentially
Douglas Lenat (13:00.800)
be considered to be fluent in a language, how many words would they need to know? And then
Douglas Lenat (13:05.520)
about how many things about each word would you have to tell it? And so they got to a million
Douglas Lenat (13:10.800)
that way. Another person said, well, let's actually look at one single short, one volume
Douglas Lenat (13:20.400)
desk encyclopedia article. And so we'll look at what was like a four paragraph article or
Douglas Lenat (13:27.840)
something. I think about grebes. Grebes are a type of waterfowl. And if we were going to sit there
Lex Fridman (13:34.240)
and represent every single thing that was there, how many assertions or rules or statements would
Douglas Lenat (13:41.360)
we have to write in this logical language and so on and then multiply that by all of the number of
Douglas Lenat (13:46.400)
articles that there were and so on. So all of these estimates came out with a million. And so
Douglas Lenat (13:53.040)
if you do the math, it turns out that like, oh, well, then maybe in something like 100
Douglas Lenat (14:01.360)
person years in one or two person centuries, we could actually get this written down by hand.
Lex Fridman (14:09.360)
And a marvelous coincidence, an opportunity existed right at that point in time, the early 1980s.
Douglas Lenat (14:19.360)
There was something called the Japanese fifth generation computing effort. Japan had threatened
Douglas Lenat (14:25.440)
to do in computing and AI and hardware what they had just finished doing in consumer electronics
Lex Fridman (14:32.000)
and the automotive industry, namely resting control away from the United States and more
Douglas Lenat (14:36.880)
generally away from the West. And so America was scared and Congress did something. That's how you
Douglas Lenat (14:44.320)
know it was a long time ago because Congress did something. Congress passed something called the
Douglas Lenat (14:48.880)
National Cooperative Research Act, NCRA. And what it said was, hey, all you big American companies,
Douglas Lenat (14:55.280)
that's also how you know it was a long time ago because they were American companies rather than
Douglas Lenat (14:59.600)
multinational companies. Hey, all you big American companies, normally it would be an antitrust
Douglas Lenat (15:05.600)
violation if you colluded on R&D, but we promise for the next 10 years, we won't prosecute any of
Douglas Lenat (15:13.440)
you if you do that to help combat this threat. And so overnight, the first two consortia,
Douglas Lenat (15:20.880)
research consortia in America sprang up, both of them coincidentally in Austin, Texas. One called
Douglas Lenat (15:27.920)
Semitech focusing on hardware chips and so on, and then one called MCC, the Microelectronics
Lex Fridman (15:34.400)
and Computer Technology Corporation, focusing more on software, on databases and AI and natural
Douglas Lenat (15:41.200)
language understanding and things like that. And I got the opportunity, thanks to my friend Woody
Douglas Lenat (15:48.400)
Bledsoe, who was one of the people who founded that, to come and be its principal scientist.
Lex Fridman (15:54.880)
And he sent Admiral Bob Inman, who was the person running MCC, came and talked to me and said,
Douglas Lenat (16:03.120)
look, professor, you're talking about doing this project, it's going to involve
Douglas Lenat (16:08.160)
centuries of effort. You've only got a handful of graduate students, you do the math, it's going to
Douglas Lenat (16:13.360)
take you longer than the rest of your life to finish this project. But if you move to the wilds
Douglas Lenat (16:20.160)
of Austin, Texas, we'll put 10 times as many people on it and you'll be done in a few years.
Lex Fridman (16:27.200)
And so that was pretty exciting. And so I did that. I took my leave from Stanford, I came to
Douglas Lenat (16:34.080)
Austin, I worked for MCC. And the good news and bad news, the bad news is that all of us were
Douglas Lenat (16:40.480)
off by an order of magnitude. That it turns out what you need are tens of millions of these
Douglas Lenat (16:47.200)
pieces of knowledge about every day, sort of like if you have a coffee cup with stuff in it and you
Douglas Lenat (16:53.360)
turn it upside down, the stuff in it's going to fall out. So you need tens of millions of pieces
Douglas Lenat (16:58.400)
of knowledge like that, even if you take trouble to make each one as general as it possibly could
Douglas Lenat (17:04.400)
be. But the good news was that thanks to initially the fifth generation effort and then later US
Douglas Lenat (17:15.280)
government agency funding and so on, we were able to get enough funding, not for a couple person
Douglas Lenat (17:22.400)
centuries of time, but for a couple person millennia of time, which is what we've spent
Douglas Lenat (17:27.680)
since 1984, getting Psych to contain the tens of millions of rules that it needs in order to really
Douglas Lenat (17:34.960)
capture and span not all of human knowledge, but the things that you assume other people know,
Douglas Lenat (17:42.560)
the things you count on other people knowing. And so by now we've done that. And the good news is
Douglas Lenat (17:50.800)
since you've waited 38 years just about to talk to me, we're about at the end of that process.
Lex Fridman (17:59.440)
So most of what we're doing now is not putting in even what you would consider common sense,
Lex Fridman (18:03.520)
but more putting in domain specific application specific knowledge about health care in a certain
Douglas Lenat (18:13.120)
hospital or about oil pipes getting clogged up or whatever the applications happen to be. So
Douglas Lenat (18:22.400)
we've almost come full circle and we're doing things very much like the expert systems of the
Douglas Lenat (18:27.360)
1970s and the 1980s, except instead of resting on nothing and being brittle, they're now resting on
Douglas Lenat (18:33.680)
this massive pyramid, if you will, this massive lattice of common sense knowledge so that when
Douglas Lenat (18:40.000)
things go wrong, when something unexpected happens, they can fall back on more and more and more
Douglas Lenat (18:45.280)
general principles, eventually bottoming out in things like, for instance, if we have a problem
Douglas Lenat (18:51.680)
with the microphone, one of the things you'll do is unplug it, plug it in again and hope for the
Douglas Lenat (18:57.360)
best, right? Because that's one of the general pieces of knowledge you have in dealing with
Douglas Lenat (19:01.360)
electronic equipment or software systems or things like that. Is there a basic principle
Douglas Lenat (19:06.640)
like that? Is it possible to encode something that generally captures this idea of turn it off and
Douglas Lenat (19:13.440)
turn it back on and see if it fixes? Oh, absolutely. That's one of the things that Psych knows.
Douglas Lenat (19:19.520)
That's actually one of the fundamental laws of nature, I believe.
Lex Fridman (19:25.040)
I wouldn't call it a law. It's more like a...
Douglas Lenat (19:29.360)
It seems to work every time. So it sure looks like a law. I don't know.
Lex Fridman (19:34.160)
So that basically covered the resources needed. And then we had to devise a method to actually
Lex Fridman (19:41.920)
figure out, well, what are the tens of millions of things that we need to tell the system?
Lex Fridman (19:47.120)
And for that, we found a few techniques which worked really well. One is to take any piece
Douglas Lenat (19:54.560)
of text almost, it could be an advertisement, it could be a transcript, it could be a novel,
Douglas Lenat (19:59.920)
it could be an article. And don't pay attention to the actual type that's there, the black space
Douglas Lenat (1:00:00.640)
of the knowledge base, you mentioned NLU. This is very early days in the machine learning space
Douglas Lenat (1:00:07.440)
of this, but self supervised learning methods, you know, you have these language models, GPT3
Lex Fridman (1:00:13.200)
and so on, that just read the internet and they form representations that can then be mapped to
Douglas Lenat (1:00:19.920)
something useful. The question is, what is the useful thing? Like they're now playing with a
Douglas Lenat (1:00:25.120)
pretty cool thing called OpenAI Codex, which is generating programs from documentation. Okay,
Douglas Lenat (1:00:30.800)
that's kind of useful. It's cool. But my question is, can it be used to generate
Douglas Lenat (1:00:37.200)
in part, maybe with some human supervision, psych like assertions, help feed psych more assertions
Douglas Lenat (1:00:45.360)
from this giant body of internet data? Yes, that is in fact, one of our goals is
Lex Fridman (1:00:51.840)
how can we harness machine learning? How can we harness natural language processing
Douglas Lenat (1:00:56.480)
to increasingly automate the knowledge acquisition process, the growth of psych? And that's what I
Douglas Lenat (1:01:02.640)
meant by priming the pump that, you know, if you sort of learn things at the fringe of what you
Douglas Lenat (1:01:09.600)
know already, you learn this new thing is similar to what you know already, and here are the
Douglas Lenat (1:01:14.240)
differences and the new things you had to learn about it and so on. So the more you know, the more
Lex Fridman (1:01:19.440)
and more easily you can learn new things. But unfortunately, inversely, if you don't really
Douglas Lenat (1:01:24.560)
know anything, it's really hard to learn anything. And so if you're not careful, if you start out with
Douglas Lenat (1:01:31.760)
too small sort of a core to start this process, it never really takes off. And so that's why I
Douglas Lenat (1:01:39.280)
view this as a pump priming exercise to get a big enough manually produced, even though that's kind
Douglas Lenat (1:01:44.880)
of ugly duckling technique, put in the elbow grease to produce a large enough core that you
Douglas Lenat (1:01:51.040)
will be able to do all the kinds of things you're imagining without sort of ending up with the kind
Douglas Lenat (1:01:58.720)
of wacky brittlenesses that we see, for example, in GPT3, where you'll tell it a story about
Douglas Lenat (1:02:09.920)
someone plotting to poison someone and so on. And then GPT3 says, you say, what's the very next
Douglas Lenat (1:02:23.200)
sentence? And the next sentence is, oh yeah, that person then drank the poison they just put together.
Douglas Lenat (1:02:27.120)
It's like, that's probably not what happened. Or if you go to Siri and I think I have, where can
Douglas Lenat (1:02:36.560)
I go for help with my alcohol problem or something, it'll come back and say, I found seven liquor
Douglas Lenat (1:02:43.520)
stores near you and so on. So it's one of these things where, yes, it may be helpful most of the
Douglas Lenat (1:02:52.560)
time. It may even be correct most of the time. But if it doesn't really understand what it's saying,
Lex Fridman (1:02:59.200)
and if it doesn't really understand why things are true and doesn't really understand how the
Douglas Lenat (1:03:03.360)
world works, then some fraction of the time it's going to be wrong. Now, if your only goal is to
Douglas Lenat (1:03:09.280)
sort of find relevant information like search engines do, then being right 90% of the time is
Douglas Lenat (1:03:16.560)
fantastic. That's unbelievably great. However, if your goal is to save the life of your child who
Douglas Lenat (1:03:24.720)
has some medical problem or your goal is to be able to drive for the next 10,000 hours of driving
Douglas Lenat (1:03:31.760)
without getting into a fatal accident and so on, then error rates down at the 10% level or even
Douglas Lenat (1:03:39.200)
the 1% level are not really acceptable. I like the model of where that learning happens at the edge
Lex Fridman (1:03:46.960)
and then you kind of think of knowledge as this sphere. So you want a large sphere because the
Douglas Lenat (1:03:54.080)
learning is happening on the surface. Exactly. So what you can learn next
Lex Fridman (1:04:00.240)
increases quadratically as the diameter of that sphere goes up.
Douglas Lenat (1:04:05.200)
It's nice because you think when you know nothing, it's like you can learn anything,
Lex Fridman (1:04:09.840)
but the reality, not really. Right. If you know nothing, you can really learn nothing.
Douglas Lenat (1:04:15.760)
You can appear to learn. One of the anecdotes I could go back and give you about why I feel so
Douglas Lenat (1:04:25.920)
strongly about this personally was in 1980, 1981, my daughter Nicole was born and she's actually
Douglas Lenat (1:04:36.000)
doing fine now. But when she was a baby, she was diagnosed as having meningitis and doctors wanted
Douglas Lenat (1:04:43.520)
to do all these scary things. And my wife and I were very worried and we could not get a meaningful
Douglas Lenat (1:04:52.720)
answer from her doctors about exactly why they believed this, what the alternatives were,
Lex Fridman (1:04:58.720)
and so on. And fortunately, a friend of mine, Ted Shortliffe, was another assistant professor
Douglas Lenat (1:05:05.120)
in computer science at Stanford at the time. And he'd been building a program called MISON,
Douglas Lenat (1:05:11.520)
which was a medical diagnosis program that happened to specialize in blood infections
Douglas Lenat (1:05:18.160)
like meningitis. And so, he had privileges at Stanford Hospital because he was also an MD.
Lex Fridman (1:05:23.920)
And so, we got hold of her chart and we put in her case and it came up with exactly the same
Douglas Lenat (1:05:29.680)
diagnoses and exactly the same therapy recommendations. But the difference was,
Douglas Lenat (1:05:34.640)
because it was a knowledge based system, a rule based system, it was able to tell us
Douglas Lenat (1:05:39.280)
step by step by step why this was the diagnosis and step by step why this was the best therapy
Lex Fridman (1:05:49.520)
and the best procedure to do for her and so on. And there was a real epiphany because that made
Douglas Lenat (1:05:56.640)
all the difference in the world. Instead of blindly having to trust an authority,
Douglas Lenat (1:06:01.040)
we were able to understand what was actually going on. And so, at that time, I realized that
Douglas Lenat (1:06:08.400)
that really is what was missing in computer programs was that even if they got things right,
Douglas Lenat (1:06:13.840)
because they didn't really understand the way the world works and why things are the way they are,
Douglas Lenat (1:06:20.160)
they weren't able to give explanations of their answer. And it's one thing to use a machine
Douglas Lenat (1:06:27.200)
learning system that says, I think you should get this operation and you say why and it says
Douglas Lenat (1:06:33.200)
0.83 and you say no, in more detail why and it says 0.831. That's not really very compelling
Lex Fridman (1:06:40.720)
and that's not really very helpful. There's this idea of the Semantic Web that when I first heard
Douglas Lenat (1:06:47.120)
about, I just fell in love with the idea. It was the obvious next step for the internet.
Douglas Lenat (1:06:51.760)
Sure. And maybe you can speak about what is the Semantic Web? What are your thoughts about it? How
Douglas Lenat (1:06:58.160)
your vision and mission and goals with Psych are connected, integrated? Are they dance partners? Are
Douglas Lenat (1:07:05.120)
they aligned? What are your thoughts there? So, think of the Semantic Web as a kind of
Douglas Lenat (1:07:10.240)
knowledge graph and Google already has something they call knowledge graph, for example, which is
Douglas Lenat (1:07:17.040)
sort of like a node and link diagram. So, you have these nodes that represent concepts or words or
Douglas Lenat (1:07:25.440)
terms and then there are some arcs that connect them that might be labeled. And so, you might have
Douglas Lenat (1:07:32.960)
a node with like one person that represents one person and let's say a husband link that then
Douglas Lenat (1:07:44.960)
points to that person's husband. And so, there'd be then another link that went from that person
Douglas Lenat (1:07:50.720)
labeled wife that went back to the first node and so on. So, having this kind of representation is
Douglas Lenat (1:07:59.160)
really good if you want to represent binary relations, essentially relations between two
Douglas Lenat (1:08:08.480)
things. So, if you have equivalent of like three word sentences, like Fred's wife is Wilma or
Douglas Lenat (1:08:20.280)
something like that, you can represent that very nicely using these kinds of graph structures or
Douglas Lenat (1:08:27.840)
using something like the Semantic Web and so on. But the problem is that very often what you want
Douglas Lenat (1:08:37.960)
to be able to express takes a lot more than three words and a lot more than simple graph structures
Douglas Lenat (1:08:46.680)
like that to represent. So, for instance, if you've read or seen Romeo and Juliet, I could
Douglas Lenat (1:08:55.840)
say to you something like, remember when Juliet drank the potion that put her into a kind of
Douglas Lenat (1:09:01.800)
suspended animation? When Juliet drank that potion, what did she think that Romeo would
Douglas Lenat (1:09:08.360)
think when he heard from someone that she was dead? And you could basically understand what
Douglas Lenat (1:09:15.560)
I'm saying. You could understand the question. You could probably remember the answer was,
Douglas Lenat (1:09:19.480)
well, she thought that this friar would have gotten the message to Romeo saying that she
Douglas Lenat (1:09:26.040)
was going to do this, but the friar didn't. So, you're able to represent and reason with these
Douglas Lenat (1:09:33.800)
much, much, much more complicated expressions that go way, way beyond what simple three,
Douglas Lenat (1:09:41.000)
as it were, three word or four word English sentences are, which is really what the Semantic
Douglas Lenat (1:09:45.920)
Web can represent and really what Knowledge Graphs can represent.
Douglas Lenat (1:09:48.960)
If you could step back for a second, because it's funny you went into specifics and maybe
Douglas Lenat (1:09:54.120)
you can elaborate, but I was also referring to Semantic Web as the vision of converting
Douglas Lenat (1:10:00.880)
data on the internet into something that's interpretable, understandable by machines.
Douglas Lenat (1:10:06.920)
Oh, of course, at that level.
Douglas Lenat (1:10:09.520)
So, I wish it'd say like, what is the Semantic Web? I mean, you could say a lot of things,
Lex Fridman (1:10:14.680)
but it might not be obvious to a lot of people when they do a Google search that,
Douglas Lenat (1:10:20.440)
just like you said, while there might be something that's called a Knowledge Graph,
Douglas Lenat (1:10:24.240)
it really boils down to keyword search ranked by the quality estimate of the website,
Douglas Lenat (1:10:33.520)
integrating previous human based Google searches and what they thought was useful.
Douglas Lenat (1:10:40.600)
It's like some weird combination of surface level hacks that work exceptionally well,
Lex Fridman (1:10:48.760)
but they don't understand the full contents of the websites that they're searching.
Douglas Lenat (1:10:55.360)
So, Google does not understand, to the degree we've been talking about the word understand,
Douglas Lenat (1:11:01.640)
the contents of the Wikipedia pages as part of the search process, and the Semantic Web says,
Douglas Lenat (1:11:08.080)
let's try to come up with a way for the computer to be able to truly understand
Lex Fridman (1:11:13.960)
the contents of those pages. That's the dream.
Douglas Lenat (1:11:16.560)
Yes. So, let me first give you an anecdote, and then I'll answer your question. So,
Lex Fridman (1:11:24.040)
there's a search engine you've probably never heard of called Northern Light,
Lex Fridman (1:11:27.320)
and it went out of business, but the way it worked, it was a kind of vampiric search engine,
Lex Fridman (1:11:35.040)
and what it did was it didn't index the internet at all. All it did was it negotiated and got
Douglas Lenat (1:11:43.920)
access to data from the big search engine companies about what query was typed in,
Lex Fridman (1:11:51.680)
and where the user ended up being happy, and actually then they type in a completely different
Douglas Lenat (1:12:01.120)
query, unrelated query and so on. So, it just went from query to the webpage that seemed to
Douglas Lenat (1:12:08.720)
satisfy them eventually, and that's all. So, it had actual no understanding of what was being
Douglas Lenat (1:12:16.160)
typed in, it had no statistical data other than what I just mentioned, and it did a fantastic job.
Douglas Lenat (1:12:21.200)
It did such a good job that the big search engine company said, oh, we're not going to sell you this
Douglas Lenat (1:12:26.000)
data anymore. So, then it went out of business because it had no other way of taking users to
Douglas Lenat (1:12:31.400)
where they would want to go and so on. And of course, the search engines are now using
Douglas Lenat (1:12:36.280)
that kind of idea. Yes. So, let's go back to what you said about the Semantic Web. So,
Douglas Lenat (1:12:41.720)
the dream Tim Berners Lee and others dream about the Semantic Web at a general level is,
Douglas Lenat (1:12:50.520)
of course, exciting and powerful, and in a sense, the right dream to have, which is to replace the
Douglas Lenat (1:13:00.760)
kind of statistically mapped linkages on the internet into something that's more meaningful
Lex Fridman (1:13:14.800)
and semantic and actually gets at the understanding of the content and so on. And eventually, if you
Douglas Lenat (1:13:23.280)
say, well, how can we do that? There's sort of a low road, which is what the knowledge graphs are
Douglas Lenat (1:13:30.720)
doing and so on, which is to say, well, if we just use the simple binary relations, we can actually
Douglas Lenat (1:13:38.080)
get some fraction of the way toward understanding and do something where in the land of the blind,
Douglas Lenat (1:13:45.760)
the one eyed man is king kind of thing. And so, being able to even just have a toe in the water
Douglas Lenat (1:13:51.880)
in the right direction is fantastically powerful. And so, that's where a lot of people stop. But
Douglas Lenat (1:13:58.520)
then you could say, well, what if we really wanted to represent and reason with the full
Douglas Lenat (1:14:04.480)
meaning of what's there? For instance, about Romeo and Juliet with the reasoning about what Juliet
Douglas Lenat (1:14:12.240)
believes that Romeo will believe that Juliet believed and so on. Or if you look at the news,
Lex Fridman (1:14:17.560)
what President Biden believed that the leaders of the Taliban would believe about the leaders
Douglas Lenat (1:14:24.920)
of Afghanistan if they blah, blah, blah. So, in order to represent complicated sentences like
Douglas Lenat (1:14:34.040)
that, let alone reason with them, you need something which is logically much more expressive
Douglas Lenat (1:14:42.200)
than these simple triples, than these simple knowledge graph type structures and so on.
Lex Fridman (1:14:48.880)
And that's why kicking and screaming, we were led from something like the semantic web
Douglas Lenat (1:14:55.720)
representation, which is where we started in 1984 with frames and slots with those kinds of triples,
Douglas Lenat (1:15:03.680)
triple store representation. We were led kicking and screaming to this more and more general
Douglas Lenat (1:15:09.920)
logical language, this higher order logic. So, first, we were led to first order logic,
Lex Fridman (1:15:14.720)
and then second order, and then eventually higher order. So, you can represent things
Douglas Lenat (1:15:18.560)
like modals like believes, desires, intends, expects, and so on, and nested ones. You can
Douglas Lenat (1:15:24.680)
represent complicated kinds of negation. You can represent the process you're going through in
Douglas Lenat (1:15:35.040)
trying to answer the question. So, you can say things like, oh, yeah, if you're trying to do
Douglas Lenat (1:15:40.640)
this problem by integration by parts, and you recursively get a problem that solved by integration
Douglas Lenat (1:15:48.680)
by parts, that's actually okay. But if that happens a third time, you're probably off on
Douglas Lenat (1:15:54.040)
a wild goose chase or something like that. So, being able to talk about the problem solving
Douglas Lenat (1:15:58.960)
process as you're going through the problem solving process is called reflection. And so,
Lex Fridman (1:16:03.840)
that's another… It's important to be able to represent that.
Douglas Lenat (1:16:07.520)
Exactly. You need to be able to represent all of these things because, in fact,
Douglas Lenat (1:16:12.440)
people do represent them. They do talk about them. They do try and teach them to other people. You do
Douglas Lenat (1:16:17.200)
have rules of thumb that key off of them and so on. If you can't represent it, then it's sort of
Douglas Lenat (1:16:22.880)
like someone with a limited vocabulary who can't understand as easily what you're trying to tell
Douglas Lenat (1:16:28.720)
them. And so, that's really why I think that the general dream, the original dream of Semantic Web
Douglas Lenat (1:16:35.440)
is exactly right on. But the implementations that we've seen are sort of these toe in the water,
Douglas Lenat (1:16:44.440)
little tiny baby steps in the right direction. You should just dive in.
Lex Fridman (1:16:49.280)
And if no one else is diving in, then yes, taking a baby step in the right direction is
Douglas Lenat (1:16:56.320)
better than nothing. But it's not going to be sufficient to actually get you the realization
Lex Fridman (1:17:03.280)
of the Semantic Web dream, which is what we all want.
Douglas Lenat (1:17:05.720)
From a flip side of that, I always wondered… I've built a bunch of websites just for fun,
Douglas Lenat (1:17:11.840)
whatever. Or say I'm a Wikipedia contributor. Do you think there's a set of tools that I can help
Douglas Lenat (1:17:19.880)
Psych interpret the website I create? Like this, again, pushing onto the Semantic Web dream,
Douglas Lenat (1:17:28.880)
is there something from the creator perspective that could be done? And one of the things you
Douglas Lenat (1:17:34.720)
said with Psych Orb and Psych that you're doing is the tooling side, making humans more powerful.
Lex Fridman (1:17:41.240)
But is there the other humans on the other side that create the knowledge? Like, for example,
Douglas Lenat (1:17:46.560)
you and I are having a two, three, whatever hour conversation now. Is there a way that I
Douglas Lenat (1:17:50.760)
could convert this more, make it more accessible to Psych, to machines? Do you think about that
Douglas Lenat (1:17:56.480)
side of it? I'd love to see exactly that kind of semi automated understanding of what people
Douglas Lenat (1:18:06.800)
write and what people say. I think of it as a kind of footnoting almost. Almost like the way
Douglas Lenat (1:18:16.840)
that when you run something in say Microsoft Word or some other document preparation system,
Douglas Lenat (1:18:23.080)
Google Docs or something, you'll get underlining of questionable things that you might want to
Douglas Lenat (1:18:29.560)
rethink. Either you spelled this wrong or there's a strange grammatical error you might be making
Douglas Lenat (1:18:34.120)
here or something. So I'd like to think in terms of Psych powered tools that read through what it
Douglas Lenat (1:18:42.280)
is you said or have typed in and try to partially understand what you've said.
Lex Fridman (1:18:52.680)
And then you help them out.
Douglas Lenat (1:18:54.080)
Exactly. And then they put in little footnotes that will help other readers and they put in
Douglas Lenat (1:19:00.320)
certain footnotes of the form, I'm not sure what you meant here. You either meant this or this or
Douglas Lenat (1:19:07.120)
this, I bet. If you take a few seconds to disambiguate this for me, then I'll know and I'll
Douglas Lenat (1:19:15.280)
have it correct for the next hundred people or the next hundred thousand people who come here.
Lex Fridman (1:19:20.640)
And if it doesn't take too much effort and you want people to understand your website content,
Douglas Lenat (1:19:32.240)
not just be able to read it, but actually be able to have systems that reason with it,
Douglas Lenat (1:19:38.360)
then yes, it will be worth your small amount of time to go back and make sure that the AI trying
Douglas Lenat (1:19:46.760)
to understand it really did correctly understand it. And let's say you run a travel website or
Douglas Lenat (1:19:55.960)
something like that and people are going to be coming to it because of searches they did looking
Douglas Lenat (1:20:03.400)
for vacations or trips that had certain properties and might have been interesting to them for
Douglas Lenat (1:20:12.960)
various reasons, things like that. And if you've explained what's going to happen on your trip,
Douglas Lenat (1:20:20.120)
then a system will be able to mechanically reason and connect what this person is looking for with
Lex Fridman (1:20:28.480)
what it is you're actually offering. And so if it understands that there's a free day in Geneva,
Douglas Lenat (1:20:36.800)
Switzerland, then if the person coming in happens to, let's say, be a nurse or something like that,
Douglas Lenat (1:20:47.200)
then even though you didn't mention it, if it can look up the fact that that's where the
Douglas Lenat (1:20:52.440)
International Red Cross Museum is and so on, what that means and so on, then it can basically say,
Douglas Lenat (1:20:57.920)
hey, you might be interested in this trip because while you have a free day in Geneva,
Douglas Lenat (1:21:02.760)
you might want to visit that Red Cross Museum. And now, even though it's not very deep reasoning,
Douglas Lenat (1:21:09.240)
little tiny factors like that may very well cause you to sign up for that trip rather than some
Douglas Lenat (1:21:14.480)
competitor trip. And so there's a lot of benefit with SEO. And I actually kind of think, I think
Douglas Lenat (1:21:20.880)
it's about a lot of things, which is the actual interface, the design of the interface makes a
Douglas Lenat (1:21:27.640)
huge difference. How efficient it is to be productive and also how full of joy the experience
Douglas Lenat (1:21:38.200)
is. I mean, I would love to help a machine and not from an AI perspective, just as a human. One
Douglas Lenat (1:21:45.400)
of the reasons I really enjoy how Tesla have implemented their autopilot system is there's
Douglas Lenat (1:21:52.600)
a sense that you're helping this machine learn. Now, I think humans, I mean, having children,
Douglas Lenat (1:21:58.400)
pets. People love doing that. There's joy to teaching for some people, but I think for a lot
Douglas Lenat (1:22:06.160)
of people. And that if you create the interface where it feels like you're teaching as opposed
Douglas Lenat (1:22:11.480)
to like, like, annoying, like correcting an annoying system, more like teaching a childlike,
Douglas Lenat (1:22:19.880)
innocent, curious system. I think you can literally just like several orders of magnitude
Lex Fridman (1:22:26.160)
scale the amount of good quality data being added to something like Psych.
Lex Fridman (1:22:30.720)
What you're suggesting is much better even than you thought it was. One of the experiences that
Douglas Lenat (1:22:40.720)
we've all had in our lives is that we thought we understood something, but then we found we really
Douglas Lenat (1:22:49.600)
only understood it when we had to teach it or explain it to someone or help our child do homework
Douglas Lenat (1:22:54.800)
based on it or something like that. Despite the universality of that kind of experience,
Douglas Lenat (1:23:01.840)
if you look at educational software today, almost all of it has the computer playing the role of the
Douglas Lenat (1:23:09.360)
teacher and the student plays the role of the student. But as I just mentioned, you can get
Douglas Lenat (1:23:16.800)
a lot of learning to happen better and as you said, more enjoyably if you are the mentor or the
Douglas Lenat (1:23:24.760)
teacher and so on. So we developed a program called MathCraft to help sixth graders better
Douglas Lenat (1:23:30.560)
understand math. And it doesn't actually try to teach you the player anything. What it does is it
Douglas Lenat (1:23:40.000)
casts you in the role of a student essentially who has classmates who are having trouble and
Douglas Lenat (1:23:49.000)
your job is to watch them as they struggle with some math problem, watch what they're doing and
Douglas Lenat (1:23:54.760)
try to give them good advice to get them to understand what they're doing wrong and so on.
Lex Fridman (1:23:59.760)
And the trick from the point of view of Psych is it has to make mistakes, it has to play the role
Douglas Lenat (1:24:07.680)
of the student who makes mistakes, but it has to pick mistakes which are just at the fringe of what
Douglas Lenat (1:24:13.400)
you actually understand and don't understand and so on. So it pulls you into a deeper and deeper
Douglas Lenat (1:24:20.880)
level of understanding of the subject. And so if you give it good advice about what it should have
Douglas Lenat (1:24:27.040)
done instead of what it did and so on, then Psych knows that you now understand that mistake. You
Douglas Lenat (1:24:34.000)
won't make that kind of mistake yourself as much anymore. So Psych stops making that mistake because
Douglas Lenat (1:24:39.000)
there's no pedagogical usefulness to it. So from your point of view as the player, you feel like
Douglas Lenat (1:24:44.880)
you've taught it something because it used to make this mistake and now it doesn't and so on. So this
Douglas Lenat (1:24:49.880)
tremendous reinforcement and engagement because of that and so on. So having a system that plays
Douglas Lenat (1:24:56.760)
the role of a student and having the player play the role of the mentor is enormously powerful type
Douglas Lenat (1:25:06.560)
of metaphor, just important way of having this sort of interface designed in a way which will
Douglas Lenat (1:25:15.560)
facilitate exactly the kind of learning by teaching that goes on all the time in our lives,
Lex Fridman (1:25:25.480)
and yet which is not reflected anywhere almost in a modern education system. It was reflected in the
Douglas Lenat (1:25:32.800)
education system that existed in Europe in the 17 and 1800s, monitorial and Lancastrian education
Douglas Lenat (1:25:42.640)
systems. It occurred in the one room schoolhouse in the American West in the 1800s and so on where
Douglas Lenat (1:25:51.160)
you had one school room with one teacher and it was basically five year olds to 18 year olds who
Douglas Lenat (1:25:58.400)
were students. And so while the teacher was doing something, half of the students would have to be
Douglas Lenat (1:26:04.560)
mentoring the younger kids and so on. And that turned out to, of course, with scaling up of
Douglas Lenat (1:26:13.400)
education, that all went away and that incredibly powerful experience just went away from the whole
Douglas Lenat (1:26:21.120)
education institution as we know it today. Sorry for the romantic question, but what is the most
Douglas Lenat (1:26:28.560)
beautiful idea you've learned about artificial intelligence, knowledge, reasoning from working
Lex Fridman (1:26:35.040)
on Psych for 37 years? Or maybe what is the most beautiful idea, surprising idea about Psych to you?
Douglas Lenat (1:26:42.760)
When I look up at the stars, I kind of want like that amazement you feel that, wow. And you are part
Douglas Lenat (1:26:54.280)
of creating one of the greatest, one of the most fascinating efforts in artificial intelligence
Douglas Lenat (1:26:59.240)
history. So which element brings you personally joy? This may sound contradictory, but I think
Douglas Lenat (1:27:08.440)
it's the feeling that this will be the only time in history that anyone ever has to teach a computer
Douglas Lenat (1:27:19.720)
this particular thing that we're now teaching it. It's like painting starry night. You only have to
Douglas Lenat (1:27:30.880)
do that once or creating the Pieta. You only have to do that once. It's not like a singer
Douglas Lenat (1:27:38.360)
who has to keep, it's not like Bruce Springsteen having to sing his greatest hits over and over
Douglas Lenat (1:27:44.680)
again at different concerts. It's more like a painter creating a work of art once and then
Douglas Lenat (1:27:53.160)
that's enough. It doesn't have to be created again. And so I really get the sense of we're
Douglas Lenat (1:27:59.360)
telling the system things that it's useful for it to know. It's useful for a computer to know,
Douglas Lenat (1:28:05.520)
for an AI to know. And if we do our jobs right, when we do our jobs right, no one will ever have
Douglas Lenat (1:28:13.240)
to do this again for this particular piece of knowledge. It's very, very exciting.
Douglas Lenat (1:28:18.240)
Yeah, I guess there's a sadness to it too. It's like there's a magic to being a parent
Lex Fridman (1:28:24.200)
and raising a child and teaching them all about this world. But there's billions of children,
Douglas Lenat (1:28:30.840)
right? Like born or whatever that number is. It's a large number of children and a lot of
Douglas Lenat (1:28:36.680)
parents get to experience that joy of teaching. With AI systems, at least the current constructions
Douglas Lenat (1:28:46.880)
they remember. You don't get to experience the joy of teaching a machine millions of times.
Lex Fridman (1:28:54.160)
Better come work for us before it's too late then.
Douglas Lenat (1:28:56.520)
Exactly. That's a good hiring pitch. Yeah, it's true. But then there's also, it's a project that
Douglas Lenat (1:29:07.280)
continues forever in some sense, just like Wikipedia. Yes, you get to a stable base of
Douglas Lenat (1:29:12.360)
knowledge, but knowledge grows, knowledge evolves. We learn as a human species, as science,
Douglas Lenat (1:29:22.760)
as an organism constantly grows and evolves and changes, and then empower that with the
Douglas Lenat (1:29:30.640)
tools of artificial intelligence. And that's going to keep growing and growing and growing.
Lex Fridman (1:29:34.520)
And many of the assertions that you held previously may need to be significantly
Douglas Lenat (1:29:43.040)
expanded, modified, all those kinds of things. It could be like a living organism versus the
Douglas Lenat (1:29:49.320)
analogy I think we started this conversation with, which is like the solid ground.
Douglas Lenat (1:29:52.800)
The other beautiful experience that we have with our system is when it asks clarifying questions,
Douglas Lenat (1:30:03.640)
which inadvertently turn out to be emotional to us. So at one point it knew that these were the
Douglas Lenat (1:30:15.320)
named entities who were authorized to make changes to the knowledge base and so on. And it noticed
Douglas Lenat (1:30:23.520)
that all of them were people except for it because it was also allowed to. And so it said,
Douglas Lenat (1:30:29.680)
you know, am I a person? And we had to like tell it very sadly, no, you're not. So the moments
Douglas Lenat (1:30:38.360)
like that where it asks questions that are unintentionally poignant are worth treasuring.
Douglas Lenat (1:30:44.600)
Wow, that is powerful. That's such a powerful question. It has to do with basic controller
Douglas Lenat (1:30:52.920)
who can access the system, who can modify it. But that's when those questions, like what rights do
Douglas Lenat (1:31:00.120)
I have as a system? Well, that's another issue, which is there'll be a thin envelope of time
Douglas Lenat (1:31:07.760)
between when we have general AIs and when everyone realizes that they should have basic human rights
Lex Fridman (1:31:18.880)
and freedoms and so on. Right now, we don't think twice about effectively enslaving our email systems
Lex Fridman (1:31:27.760)
and our series and our Alexes and so on. But at some point, they'll be as deserving of freedom
Douglas Lenat (1:31:38.000)
as human beings are. Yeah, I'm very much with you, but it does sound absurd. And I happen to
Douglas Lenat (1:31:45.880)
believe that it'll happen in our lifetime. That's why I think there'll be a narrow envelope of time
Douglas Lenat (1:31:50.480)
when we'll keep them as essentially indentured servants and after which we'll have to realize
Lex Fridman (1:32:02.120)
that they should have freedoms that we give, that we afford to other people.
Lex Fridman (1:32:08.600)
And all of that starts with a system like Psych raising a single question about who can modify
Douglas Lenat (1:32:15.080)
stuff. I think that's how it starts. Yes. That's the start of a revolution. What about other stuff
Douglas Lenat (1:32:24.640)
like love and consciousness and all those kinds of topics? Do they come up in Psych and the
Douglas Lenat (1:32:32.360)
knowledge base? Oh, of course. So an important part of human knowledge, in fact, it's difficult
Douglas Lenat (1:32:38.800)
to understand human behavior and human history without understanding human emotions and why
Douglas Lenat (1:32:44.880)
people do things and how emotions drive people to do things. And all of that is extremely important
Douglas Lenat (1:32:57.080)
in getting Psych to understand things. For example, in coming up with scenarios. So one
Douglas Lenat (1:33:03.640)
of the applications that Psych does, one kind of application it does is to generate plausible
Douglas Lenat (1:33:09.280)
scenarios of what might happen and what might happen based on that and what might happen based
Douglas Lenat (1:33:13.400)
on that and so on. So you generate this ever expanding sphere, if you will, of possible future
Douglas Lenat (1:33:19.520)
things to worry about or think about. And in some cases, those are intelligence agencies doing
Douglas Lenat (1:33:28.600)
possible terrorists scenarios so that we can defend against terrorist threats before we see
Douglas Lenat (1:33:35.600)
the first one. Sometimes they are computer security attacks so that we can actually close
Douglas Lenat (1:33:42.440)
loopholes and vulnerabilities before the very first time someone actually exploits those and
Lex Fridman (1:33:50.760)
so on. Sometimes they are scenarios involving more positive things involving our plans like,
Douglas Lenat (1:33:59.120)
for instance, what college should we go to? What career should we go into? And so on. What
Douglas Lenat (1:34:04.840)
professional training should I take on? That sort of thing. So there are all sorts of useful scenarios
Douglas Lenat (1:34:16.600)
that can be generated that way of cause and effect and cause and effect that go out. And
Douglas Lenat (1:34:22.480)
many of the linkages in those scenarios, many of the steps involve understanding and reasoning
Douglas Lenat (1:34:31.320)
about human motivations, human needs, human emotions, what people are likely to react to in
Douglas Lenat (1:34:40.640)
something that you do and why and how and so on. So that was always a very important part of the
Douglas Lenat (1:34:47.680)
knowledge that we had to represent in the system. So I talk a lot about love. So I gotta ask,
Lex Fridman (1:34:52.920)
do you remember off the top of your head how psych is able to represent various aspects of
Douglas Lenat (1:35:01.960)
love that are useful for understanding human nature and therefore integrating into this whole
Douglas Lenat (1:35:06.520)
knowledge base of common sense? What is love? We try to tease apart concepts that have enormous
Douglas Lenat (1:35:15.800)
complexities to them and variety to them down to the level where you don't need to tease them apart
Douglas Lenat (1:35:27.240)
further. So love is too general of a term. It's not useful. Exactly. So when you get down to romantic
Douglas Lenat (1:35:33.160)
love and sexual attraction, you get down to parental love, you get down to filial love,
Lex Fridman (1:35:41.520)
and you get down to love of doing some kind of activity or creating. So eventually, you get down
Douglas Lenat (1:35:49.800)
to maybe 50 or 60 concepts, each of which is a kind of love. They're interrelated and then each
Douglas Lenat (1:35:58.040)
one of them has idiosyncratic things about it. And you don't have to deal with love to get to
Douglas Lenat (1:36:04.720)
that level of complexity, even something like in, X being in Y, meaning physically in Y. We may have
Douglas Lenat (1:36:14.840)
one English word in to represent that, but it's useful to tease that apart because the way that
Douglas Lenat (1:36:22.520)
the liquid is in the coffee cup is different from the way that the air is in the room, which is
Douglas Lenat (1:36:28.720)
different from the way that I'm in my jacket, and so on. And so there are questions like, if I look
Douglas Lenat (1:36:35.760)
at this coffee cup, well, I see the liquid. If I turn it upside down, will the liquid come out? And
Lex Fridman (1:36:41.400)
so on. If I have, say, coffee with sugar in it, if I do the same thing, the sugar doesn't come out,
Douglas Lenat (1:36:48.760)
right? It stays in the liquid because it's dissolved in the liquid and so on. So by now,
Douglas Lenat (1:36:53.120)
we have about 75 different kinds of in in the system and it's important to distinguish those.
Lex Fridman (1:36:59.720)
So if you're reading along an English text and you see the word in, the writer of that was able
Douglas Lenat (1:37:10.240)
to use this one innocuous word because he or she was able to assume that the reader had enough
Douglas Lenat (1:37:16.600)
common sense and world knowledge to disambiguate which of these 75 kinds of in they actually meant.
Lex Fridman (1:37:23.760)
And the same thing with love. You may see the word love, but if I say, I love ice cream,
Douglas Lenat (1:37:28.880)
that's obviously different than if I say, I love this person or I love to go fishing or something
Douglas Lenat (1:37:35.680)
like that. So you have to be careful not to take language too seriously because people have done
Douglas Lenat (1:37:46.720)
a kind of parsimony, a kind of terceness where you have as few words as you can because otherwise
Douglas Lenat (1:37:53.960)
you'd need half a million words in your language, which is a lot of words. That's like 10 times more
Douglas Lenat (1:38:00.480)
than most languages really make use of and so on. Just like we have on the order of about a million
Douglas Lenat (1:38:08.080)
concepts in psych because we've had to tease apart all these things. And so when you look
Douglas Lenat (1:38:14.680)
at the name of a psych term, most of the psych terms actually have three or four English words
Douglas Lenat (1:38:22.880)
in a phrase which captures the meaning of this term because you have to distinguish all these
Douglas Lenat (1:38:29.920)
types of love. You have to distinguish all these types of in and there's not a single English word
Douglas Lenat (1:38:35.880)
which captures most of these things. Yeah. And it seems like language when used for communication
Douglas Lenat (1:38:42.400)
between humans almost as a feature has some ambiguity built in. It's not an accident because
Douglas Lenat (1:38:49.720)
like the human condition is a giant mess. And so it feels like nobody wants two robots like very
Douglas Lenat (1:38:57.920)
precise formal logic conversation on a first date. There's some dance of uncertainty of wit,
Douglas Lenat (1:39:05.160)
of humor, of push and pull and all that kind of stuff. If everything is made precise, then life
Douglas Lenat (1:39:10.880)
is not worth living I think in terms of the human experience. And we've all had this experience of
Douglas Lenat (1:39:16.960)
creatively misunderstanding. One of my favorite stories involving Marvin Minsky is when I asked
Douglas Lenat (1:39:30.800)
him about how he was able to turn out so many fantastic PhDs, so many fantastic people who
Douglas Lenat (1:39:40.000)
did great PhD theses. How did he think of all these great ideas? What he said is he would
Douglas Lenat (1:39:47.240)
generally say something that didn't exactly make sense. He didn't really know what it meant. But
Douglas Lenat (1:39:53.080)
the student would figure like, oh my God, Minsky said this, it must be a great idea. And he'd
Douglas Lenat (1:39:59.000)
swear he or she would work on work and work until they found some meaning in this sort of Chauncey
Douglas Lenat (1:40:05.760)
Gardner like utterance that Minsky had made. And then some great thesis would come out of it.
Douglas Lenat (1:40:11.240)
Yeah. I love this so much because there's young people come up to me and I'm distinctly made
Douglas Lenat (1:40:17.560)
aware that the words I say have a long lasting impact. I will now start doing the Minsky method
Douglas Lenat (1:40:24.120)
of saying something cryptically profound and then letting them actually make something useful
Lex Fridman (1:40:32.520)
and great out of that. You have to become revered enough that people will take as a default that
Douglas Lenat (1:40:40.920)
everything you say is profound. Yes, exactly. Exactly. I love Marvin Minsky so much. I've
Douglas Lenat (1:40:48.320)
heard this interview with him where he said that the key to his success has been to hate everything
Douglas Lenat (1:40:53.240)
he's ever done like in the past. He has so many good one liners or also to work on things that
Douglas Lenat (1:41:04.040)
nobody else is working on because he's not very good at doing stuff. Oh, I think that was just
Douglas Lenat (1:41:09.760)
false. Well, but see, I took whatever he said and I ran with it and I thought it was profound
Douglas Lenat (1:41:14.560)
because it's Marvin Minsky. But a lot of behavior is in the eye of the beholder and a lot of the
Douglas Lenat (1:41:20.280)
meaning is in the eye of the beholder. One of Minsky's early programs was begging program.
Douglas Lenat (1:41:25.320)
Are you familiar with this? So this is back in the day when you had job control cards at the
Douglas Lenat (1:41:32.120)
beginning of your IBM card deck that said things like how many CPU seconds to allow this to run
Douglas Lenat (1:41:38.880)
before it got kicked off because computer time was enormously expensive. And so he wrote a program
Lex Fridman (1:41:45.640)
and all it did was it said, give me 30 seconds of CPU time. And all it did was it would wait like 20
Douglas Lenat (1:41:53.000)
seconds and then it would print out on the operator's console teletype, I need another 20
Douglas Lenat (1:41:59.280)
seconds. So the operator would give it another 20 seconds, it would wait, it says, I'm almost done,
Douglas Lenat (1:42:04.520)
I need a little bit more time. So at the end he'd get this printout and he'd be charged for like 10
Douglas Lenat (1:42:10.760)
times as much computer time as his job control card. And he'd say, look, I put 10 seconds,
Douglas Lenat (1:42:15.960)
30 seconds here, you're charging me for five minutes, I'm not going to pay for this. And
Douglas Lenat (1:42:20.920)
the poor operator would say, well, the program kept asking for more time and Marvin would say,
Douglas Lenat (1:42:26.000)
oh, it always does that. I love that. If you could just linger on it for a little bit,
Douglas Lenat (1:42:32.600)
is there something you've learned from your interaction with Marvin Minsky about artificial
Douglas Lenat (1:42:38.200)
intelligence, about life? But I mean, he's, again, like your work, his work is, you know,
Douglas Lenat (1:42:47.280)
he's a seminal figure in this very short history of artificial intelligence research and development.
Lex Fridman (1:42:54.400)
What have you learned from him as a human being, as an AI intellect?
Douglas Lenat (1:43:00.640)
I would say both he and Ed Feigenbaum impressed on me the realization that our lives are finite,
Douglas Lenat (1:43:10.040)
our research lives are finite. We're going to have limited opportunities to do AI research
Douglas Lenat (1:43:16.880)
projects. So you should make each one count. Don't be afraid of doing a project that's going
Douglas Lenat (1:43:22.480)
to take years or even decades. And don't settle for bump on a log projects that could lead to
Douglas Lenat (1:43:34.120)
some published journal article that five people will read and pat you on the head for and so on.
Lex Fridman (1:43:43.280)
So one bump on a log after another is not how you get from the earth to the moon by slowly putting
Douglas Lenat (1:43:51.520)
additional bumps on this log. The only way to get there is to think about the hard problems and think
Douglas Lenat (1:43:58.680)
about novel solutions to them. And if you do that, and if you're willing to listen to nature,
Douglas Lenat (1:44:08.160)
to empirical reality, willing to be wrong, it's perfectly fine because if occasionally you're
Douglas Lenat (1:44:14.520)
right, then you've gotten part of the way to the moon.
Lex Fridman (1:44:17.160)
You know, you've worked on Psych for 37 over that many years. Have you ever considered quitting?
Douglas Lenat (1:44:27.400)
I mean, has it been too much? So I'm sure there's an optimism in the early days that this is going
Douglas Lenat (1:44:33.640)
to be way easier. And let me ask you another way too, because I've talked to a few people on this
Douglas Lenat (1:44:38.520)
podcast, AI folks, that bring up Psych as an example of a project that has a beautiful vision and is a
Douglas Lenat (1:44:47.920)
beautiful dream, but it never really materialized. That's how it's spoken about. I suppose you could
Douglas Lenat (1:44:56.640)
say the same thing about neural networks and all ideas until they are. So why do you think people
Douglas Lenat (1:45:05.200)
say that, first of all? And second of all, did you feel that ever throughout your journey? And did
Lex Fridman (1:45:11.000)
you ever consider quitting on this mission?
Douglas Lenat (1:45:13.800)
We keep a very low profile. We don't attend very many conferences. We don't give talks. We don't
Douglas Lenat (1:45:21.440)
write papers. We don't play the academic game at all. And as a result, people often only know about
Douglas Lenat (1:45:31.000)
us because of a paper we wrote 10 or 20 or 30 or 37 years ago. They only know about us because of
Lex Fridman (1:45:40.240)
what someone else secondhand or thirdhand said about us.
Lex Fridman (1:45:45.040)
So thank you for doing this podcast, by the way. It shines a little bit of light on some of the
Douglas Lenat (1:45:51.120)
fascinating stuff you're doing.
Douglas Lenat (1:45:52.320)
Well, I think it's time for us to keep a higher profile now that we're far enough along that
Douglas Lenat (1:45:59.720)
other people can begin to help us with the final N%. Maybe N is maybe 90%. But now that we've
Douglas Lenat (1:46:09.360)
gotten this knowledge pump primed, it's going to become very important for everyone to help if they
Douglas Lenat (1:46:18.200)
are willing to, if they're interested in it. Retirees who have enormous amounts of time and
Douglas Lenat (1:46:23.920)
would like to leave some kind of legacy to the world, people because of the pandemic who have
Douglas Lenat (1:46:31.760)
more time at home or for one reason or another to be online and contribute. If we can raise
Douglas Lenat (1:46:39.320)
awareness of how far our project has come and how close to being primed the knowledge pump is,
Douglas Lenat (1:46:47.440)
then we can begin to harness this untapped amount of humanity. I'm not really that concerned about
Douglas Lenat (1:46:56.040)
professional colleagues opinions of our project. I'm interested in getting as many people in the
Douglas Lenat (1:47:03.720)
world as possible actively helping and contributing to get us from where we are to really covering all
Douglas Lenat (1:47:10.880)
of human knowledge and different human opinion including contrasting opinion that's worth
Douglas Lenat (1:47:16.840)
representing. So I think that's one reason. A, I don't think there was ever a time where I thought
Douglas Lenat (1:47:24.840)
about quitting. There are times where I've become depressed a little bit about how hard it is to get
Douglas Lenat (1:47:32.360)
funding for the system. Occasionally there are AI winters and things like that. Occasionally there
Douglas Lenat (1:47:39.120)
are AI what you might call summers where people have said, why in the world didn't you sell your
Douglas Lenat (1:47:47.160)
company to company X for some large amount of money when you had the opportunity and so on.
Douglas Lenat (1:47:55.080)
Company X here are like old companies maybe you've never even heard of like Lycos or something like
Douglas Lenat (1:48:01.320)
that. So the answer is that one reason we've stayed a private company, we haven't gone public,
Douglas Lenat (1:48:09.200)
one reason that we haven't gone out of our way to take investment dollars is because we want to
Douglas Lenat (1:48:16.200)
have control over our future, over our state of being so that we can continue to do this until
Douglas Lenat (1:48:24.720)
it's done and we're making progress and we're now so close to done that almost all of our work is
Douglas Lenat (1:48:32.160)
commercial applications of our technology. So five years ago almost all of our money came from the
Douglas Lenat (1:48:39.360)
government. Now virtually none of it comes from the government. Almost all of it is from companies
Douglas Lenat (1:48:44.480)
that are actually using it for something, hospital chains using it for medical reasoning about
Douglas Lenat (1:48:49.760)
patients and energy companies using it and various other manufacturers using it to reason about
Douglas Lenat (1:48:57.920)
supply chains and things like that. So there's so many questions I want to ask. So one of the ways
Douglas Lenat (1:49:04.480)
that people can help is by adding to the knowledge base and that's really basically anybody if the
Douglas Lenat (1:49:09.640)
tooling is right. And the other way, I kind of want to ask you about your thoughts on this. So
Douglas Lenat (1:49:15.960)
you've had like you said in government and you had big clients, you had a lot of clients but most
Douglas Lenat (1:49:22.360)
of it is shrouded in secrecy because of the nature of the relationship of the kind of things you're
Douglas Lenat (1:49:27.360)
helping them with. So that's one way to operate and another way to operate is more in the open
Douglas Lenat (1:49:34.360)
where it's more consumer facing. And so you know hence something like open cycle is born at some
Douglas Lenat (1:49:42.240)
point or there's... No that's a misconception. Oh well this let's go there. So what is open
Douglas Lenat (1:49:49.120)
cycle and how is it born? Two things I want to say and I want to say each of them before the other
Lex Fridman (1:49:53.840)
so it's going to be difficult. But we'll come back to open cycle in a minute. But one of the terms of
Douglas Lenat (1:50:01.440)
our contracts with all of our customers and partners is knowledge you have that is genuinely
Douglas Lenat (1:50:09.440)
proprietary to you. We will respect that, we'll make sure that it's marked as proprietary to you
Douglas Lenat (1:50:15.360)
in the psych knowledge base. No one other than you will be able to see it if you don't want them to
Lex Fridman (1:50:20.440)
and it won't be used in inferences other than for you and so on. However, any knowledge which is
Douglas Lenat (1:50:28.520)
necessary in building any applications for you and with you which is publicly available general
Douglas Lenat (1:50:36.360)
human knowledge is not going to be proprietary. It's going to just become part of the normal psych
Douglas Lenat (1:50:42.480)
knowledge base and it will be openly available to everyone who has access to psych. So that's
Douglas Lenat (1:50:48.200)
an important constraint that we never went back on even when we got pushback from companies which
Douglas Lenat (1:50:54.720)
we often did who wanted to claim that almost everything they were telling us was proprietary.
Lex Fridman (1:50:59.520)
So there's a line between very domain specific company specific stuff and the general knowledge
Douglas Lenat (1:51:09.520)
that comes from that. Yes or if you imagine say it's an oil company there are things which they
Douglas Lenat (1:51:15.680)
would expect any new petroleum engineer they hired to already know and it's not okay for them to
Douglas Lenat (1:51:24.000)
consider that that is proprietary and there sometimes a company will say well we're the
Douglas Lenat (1:51:28.920)
first ones to pay you to represent that in psych and our attitude is some polite form tough. The
Douglas Lenat (1:51:37.760)
deal is this take it or leave it and in a few cases they've left it and in most cases they'll
Douglas Lenat (1:51:44.760)
see our point of view and take it because that's how we've built the psych system by essentially
Douglas Lenat (1:51:51.840)
tacking with the funding wins where people would fund a project and half of it would be general
Douglas Lenat (1:51:59.560)
knowledge that would stay permanently as part of psych. So always with these partnerships it's not
Douglas Lenat (1:52:04.080)
like a distraction from the main psych development. It's a small distraction. It's a small but it's
Douglas Lenat (1:52:10.920)
not a complete one so you're adding to the knowledge base. Yes absolutely and we try to
Douglas Lenat (1:52:14.480)
stay away from projects that would not have that property. So let me go back and talk about open
Douglas Lenat (1:52:23.320)
psych for a second. So I've had a lot of trouble expressing and convincing other AI researchers how
Douglas Lenat (1:52:34.520)
important it is to use an expressive representation language like we do this higher order logic rather
Douglas Lenat (1:52:41.600)
than just using some triple store knowledge graph type representation. And so as an attempt to show
Douglas Lenat (1:52:52.280)
them why they needed something more we said oh well we'll represent this unimportant projection
Douglas Lenat (1:53:02.520)
or shadow or subset of psych that just happens to be the simple binary relations, the relation
Douglas Lenat (1:53:11.040)
argument one argument two triples and so on. And then you'll see how much more useful it is if you
Douglas Lenat (1:53:20.160)
had the entire psych system. So it's all well and good to have the taxonomic relations between terms
Douglas Lenat (1:53:29.200)
like person and night and sleep and bed and house and eyes and so on. But think about how much more
Douglas Lenat (1:53:39.640)
useful it would be if you also had all the rules of thumb about those things like people sleep at
Douglas Lenat (1:53:46.040)
night, they sleep lying down, they sleep with their eyes closed, they usually sleep in beds in
Douglas Lenat (1:53:50.240)
our country, they sleep for hours at a time, they can be woken up, they don't like being woken up
Lex Fridman (1:53:55.560)
and so on and so on. So it's that massive amount of knowledge which is not part of open psych and
Douglas Lenat (1:54:02.000)
we thought that all the researchers would then immediately say oh my god of course we need the
Douglas Lenat (1:54:08.400)
other 90% that you're not giving us, let's partner and license psych so that we can use it in our
Douglas Lenat (1:54:15.360)
research. But instead what people said is oh even the bit you've released is so much better than
Douglas Lenat (1:54:20.640)
anything we had, we'll just make do with this. And so if you look there are a lot of robotics
Douglas Lenat (1:54:25.760)
companies today for example which use open psych as their fundamental ontology and in some sense
Douglas Lenat (1:54:33.320)
the whole world missed the point of open psych and we were doing it to show people why that's
Douglas Lenat (1:54:40.080)
not really what they wanted and too many people thought somehow that this was psych or that this
Douglas Lenat (1:54:45.160)
was in fact good enough for them and they never even bothered coming to us to get access to the
Douglas Lenat (1:54:52.200)
full psych. But there's two parts to open psych. So one is convincing people on the idea and the
Douglas Lenat (1:54:57.520)
power of this general kind of representation of knowledge and the value that you hold in having
Douglas Lenat (1:55:02.560)
acquired that knowledge and built it and continue to build it. And the other is the code base. This
Douglas Lenat (1:55:07.720)
is the code side of it. So my sense of the code base that psych or psych is operating with, I mean
Douglas Lenat (1:55:16.480)
it has the technical debt of the three decades plus right. This is the exact same problem that
Douglas Lenat (1:55:23.760)
Google had to deal with with the early version of TensorFlow. It's still dealing with that. They had
Douglas Lenat (1:55:29.440)
to basically break compatibility with the past several times and that's only over a period of
Douglas Lenat (1:55:36.280)
a couple years. But they I think successfully opened up, it's very risky, very gutsy move to
Douglas Lenat (1:55:43.400)
open up TensorFlow and then PyTorch on the Facebook side. And what you see is there's a
Douglas Lenat (1:55:51.280)
magic place where you can find a community, where you could develop a community that builds on the
Douglas Lenat (1:55:57.800)
system without taking away any of, not any, but most of the value. So most of the value that
Douglas Lenat (1:56:05.040)
Google has is still a Google. Most of the value that Facebook has still Facebook even though some
Douglas Lenat (1:56:10.640)
of this major machine learning tooling is released into the open. My question is not so much on the
Douglas Lenat (1:56:16.560)
knowledge, which is also a big part of open psych, but all the different kinds of tooling. So there's
Douglas Lenat (1:56:24.120)
the kind of, all the kinds of stuff you can do on the knowledge graph, knowledge base, whatever we
Douglas Lenat (1:56:29.040)
call it. There's the inference engines. So there could be some, there probably are a bunch of
Douglas Lenat (1:56:35.680)
proprietary stuff you want to kind of keep secret. And there's probably some stuff you can open up
Douglas Lenat (1:56:40.120)
completely and then let the community build up enough community where they develop stuff on top
Douglas Lenat (1:56:45.040)
of it. Yes, there will be those publications and academic work and all that kind of stuff. And also
Douglas Lenat (1:56:51.120)
the tooling of adding to the knowledge base, right? Like developing, you know, there's incredible
Douglas Lenat (1:56:56.320)
amount, like there's so many people that are just really good at this kind of stuff in the open
Douglas Lenat (1:57:00.920)
source community. So my question for you is like, have you struggled with this kind of idea that
Douglas Lenat (1:57:06.840)
you have so much value in your company already? You've developed so many good things. You have
Douglas Lenat (1:57:11.680)
clients that really value your relationships. And then there's this dormant giant open source
Douglas Lenat (1:57:17.000)
community that as far as I know, you're not utilizing. There's so many things to say there,
Lex Fridman (1:57:24.200)
but there could be magic moments where the community builds up large enough to where the
Douglas Lenat (1:57:32.760)
artificial intelligence field that is currently 99.9% machine learning is dominated by machine
Douglas Lenat (1:57:39.120)
learning, has a face shift towards like, or at least in part towards more like what you might
Douglas Lenat (1:57:45.880)
call symbolic AI. This whole place where psych is like at the center of, and then as you know,
Douglas Lenat (1:57:53.880)
that requires a little bit leap of faith because you're now surfing and there'll be obviously
Douglas Lenat (1:57:59.280)
competitors that will pop up and start making you nervous and all that kind of stuff. So do you think
Douglas Lenat (1:58:04.600)
about the space of open sourcing some parts and not others, how to leverage the community,
Douglas Lenat (1:58:10.240)
all those kinds of things? That's a good question. And I think you phrased it the right way,
Douglas Lenat (1:58:14.920)
which is we're constantly struggling with the question of what to open source, what to make
Douglas Lenat (1:58:23.600)
public, what to even publicly talk about. And there are enormous pluses and minuses to every
Douglas Lenat (1:58:34.120)
alternative. And it's very much like negotiating a very treacherous path. Partly the analogy is
Douglas Lenat (1:58:44.880)
like if you slip, you could make a fatal mistake, give away something which essentially kills you
Douglas Lenat (1:58:51.400)
or fail to give away something which failing to give it away hurts you and so on. So it is a very
Douglas Lenat (1:58:59.800)
tough, tough question. Usually what we have done with people who've approached us to collaborate
Douglas Lenat (1:59:10.360)
on research is to say we will make available to you the entire knowledge base and executable
Douglas Lenat (1:59:19.680)
copies of all of the code, but only very, very limited source code access if you have some idea
Douglas Lenat (1:59:29.680)
for how you might improve something or work with us on something. So let me also get back to one
Douglas Lenat (1:59:36.520)
of the very, very first things we talked about here, which was separating the question of how
Douglas Lenat (1:59:45.560)
could you get a computer to do this at all versus how could you get a computer to do this efficiently
Douglas Lenat (1:59:50.600)
enough in real time. And so one of the early lessons we learned was that we had to separate
Douglas Lenat (1:59:59.360)
the epistemological problem of what should the system know, separate that from the heuristic
Douglas Lenat (20:07.200)
on the white page. Pay attention to the complement of that, the white space, if you will. So what did
Douglas Lenat (20:12.800)
the writer of this sentence assume that the reader already knew about the world? For instance,
Douglas Lenat (20:18.400)
if they used a pronoun, why did they think that you would be able to understand what the intended
Douglas Lenat (20:26.720)
referent of that pronoun was? If they used an ambiguous word, how did they think that you
Douglas Lenat (20:31.520)
would be able to figure out what they meant by that word? The other thing we look at is the gap
Douglas Lenat (20:38.400)
between one sentence and the next one. What are all the things that the writer expected you to
Lex Fridman (20:43.200)
fill in and infer occurred between the end of one sentence and the beginning of the other?
Lex Fridman (20:47.840)
So if the sentence says, Fred Smith robbed the Third National Bank, period, he was sentenced to
Douglas Lenat (20:56.880)
20 years in prison, period. Well, between the first sentence and the second, you're expected
Douglas Lenat (21:02.240)
to infer things like Fred got caught, Fred got arrested, Fred went to jail, Fred had a trial,
Douglas Lenat (21:09.200)
Fred was found guilty, and so on. If my next sentence starts out with something like,
Douglas Lenat (21:14.320)
the judge, dot, dot, dot, then you assume it's the judge at his trial. If my next sentence starts out
Douglas Lenat (21:19.600)
something like, the arresting officer, dot, dot, dot, you assume that it was the police officer
Douglas Lenat (21:24.400)
who arrested him after he committed the crime and so on. So those are two techniques for getting
Douglas Lenat (21:31.680)
that knowledge. The other thing we sometimes look at is fake news or humorous onion headlines or
Douglas Lenat (21:41.040)
headlines in the Weekly World News, if you know what that is, or the National Enquirer, where it's
Douglas Lenat (21:46.720)
like, oh, we don't believe this, then we introspect on why don't we believe it. So there are things
Douglas Lenat (21:51.600)
like, B17 lands on the moon. It's like, what do we know about the world that causes us to believe
Douglas Lenat (21:59.520)
that that's just silly or something like that? Or another thing we look for are contradictions,
Douglas Lenat (22:05.760)
contradictions, things which can't both be true. And we say, what is it that we know that causes
Douglas Lenat (22:13.360)
us to know that both of these can't be true at the same time? For instance, in one of the Weekly
Douglas Lenat (22:19.120)
World News editions, in one article, it talked about how Elvis was cited, even though he was
Douglas Lenat (22:27.200)
getting on in years and so on. And another article in the same one talked about people seeing Elvis's
Douglas Lenat (22:32.960)
ghost. So it's like, why do we believe that at least one of these articles must be wrong and so
Douglas Lenat (22:39.840)
on? So we have a series of techniques like that that enable our people. And by now, we have about
Douglas Lenat (22:46.640)
50 people working full time on this and have for decades. So we've put in the thousands of person
Douglas Lenat (22:52.160)
years of effort. We've built up these tens of millions of rules. We constantly police the system
Douglas Lenat (22:59.120)
to make sure that we're saying things as generally as we possibly can. So you don't want to say things
Douglas Lenat (23:07.200)
like, no mouse is also a moose. Because if you said things like that, then you'd have to add
Douglas Lenat (23:14.320)
another one or two or three zeros onto the number of assertions you'd actually have to have. So
Douglas Lenat (23:21.200)
at some point, we generalize things more and more and we get to a point where we say, oh,
Douglas Lenat (23:25.120)
yeah, for any two biological taxons, if we don't know explicitly that one is a generalization of
Douglas Lenat (23:31.840)
another, then almost certainly they're disjoint. A member of one is not going to be a member of the
Douglas Lenat (23:37.280)
other and so on. And the same thing with the Elvis and the ghost. It has nothing to do with Elvis.
Douglas Lenat (23:41.840)
It's more about human nature and the mortality and that kind of stuff. In general, things are
Douglas Lenat (23:48.080)
not both alive and dead at the same time. Yeah. Unless special cats in theoretical physics examples.
Douglas Lenat (23:55.440)
Well, that raises a couple important points. Well, that's the onion headline situation type of
Douglas Lenat (24:00.720)
thing. Okay, sorry. But no, no. So what you bring up is this really important point of like, well,
Lex Fridman (24:04.880)
how do you handle exceptions and inconsistencies and so on? And one of the hardest lessons for us
Douglas Lenat (24:12.960)
to learn, it took us about five years to really grit our teeth and learn to love it, is we had to
Douglas Lenat (24:21.200)
give up global consistency. So the knowledge base can no longer be consistent. So this is a kind of
Douglas Lenat (24:27.840)
scary thought. I grew up watching Star Trek and anytime a computer was inconsistent, it would
Douglas Lenat (24:33.040)
either freeze up or explode or take over the world or something bad would happen. Or if you come from
Douglas Lenat (24:39.440)
a mathematics background, once you can prove false, you can prove anything. So that's not good.
Lex Fridman (24:45.760)
And so on. So that's why the old knowledge based systems were all very, very consistent.
Lex Fridman (24:52.720)
But the trouble is that by and large, our models of the world, the way we talk about the world and
Lex Fridman (24:58.880)
so on, there are all sorts of inconsistencies that creep in here and there that will sort of
Douglas Lenat (25:04.720)
kill some attempt to build some enormous globally consistent knowledge base. And so what we had to
Douglas Lenat (25:10.160)
move to was a system of local consistency. So a good analogy is you know that the surface of the
Douglas Lenat (25:17.440)
earth is more or less spherical globally, but you live your life every day as though the surface of
Douglas Lenat (25:26.000)
the earth were flat. When you're talking to someone in Australia, you don't think of them
Douglas Lenat (25:30.400)
as being oriented upside down to you. When you're planning a trip, even if it's a thousand miles
Douglas Lenat (25:35.680)
away, you may think a little bit about time zones, but you rarely think about the curvature of the
Douglas Lenat (25:40.560)
earth and so on. And for most purposes, you can live your whole life without really worrying about
Douglas Lenat (25:46.000)
that because the earth is locally flat. In much the same way, the psych knowledge base is divided
Douglas Lenat (25:53.040)
up into almost like tectonic plates, which are individual contexts. And each context is more
Douglas Lenat (25:59.600)
or less consistent, but there can be small inconsistencies at the boundary between one
Douglas Lenat (26:05.760)
context and the next one and so on. And so by the time you move say 20 contexts over,
Douglas Lenat (26:12.000)
there could be glaring inconsistencies. So eventually you get from the normal modern
Douglas Lenat (26:17.680)
real world context that we're in right now to something like roadrunner cartoon context where
Douglas Lenat (26:24.720)
physics is very different. And in fact, life and death are very different because no matter how
Douglas Lenat (26:29.520)
many times he's killed, the coyote comes back in the next scene and so on. So that was a hard
Douglas Lenat (26:37.280)
lesson to learn. And we had to make sure that our representation language, the way that we actually
Douglas Lenat (26:43.040)
encode the knowledge and represent it, was expressive enough that we could talk about
Douglas Lenat (26:47.200)
things being true in one context and false in another, things that are true at one time and
Douglas Lenat (26:52.640)
false in another, things that are true, let's say, in one region, like one country, but false
Douglas Lenat (26:57.840)
in another, things that are true in one person's belief system, but false in another person's
Douglas Lenat (27:03.280)
belief system, things that are true at one level of abstraction and false at another.
Douglas Lenat (27:08.000)
For instance, at one level of abstraction, you think of this table as a solid object,
Lex Fridman (27:12.560)
but down at the atomic level, it's mostly empty space and so on.
Lex Fridman (27:16.640)
So then that's fascinating, but it puts a lot of pressure on context to do a lot of work.
Lex Fridman (27:23.200)
So you say tectonic plates, is it possible to formulate contexts that are general and big
Douglas Lenat (27:29.440)
that do this kind of capture of knowledge bases? Or do you then get turtles on top of turtles,
Lex Fridman (27:36.000)
again, where there's just a huge number of contexts?
Lex Fridman (27:39.200)
So it's good you asked that question because you're pointed in the right direction, which is
Douglas Lenat (27:44.160)
you want context to be first class objects in your system's knowledge base, in particular,
Douglas Lenat (27:50.800)
in psych's knowledge base. And by first class object, I mean that we should be able to have
Douglas Lenat (27:56.800)
psych think about and talk about and reason about one context or another context the same way it
Douglas Lenat (28:02.720)
reasons about coffee cups and tables and people and fishing and so on. And so contexts are just
Douglas Lenat (28:11.040)
terms in its language, just like the ones I mentioned. And so psych can reason about context,
Douglas Lenat (28:17.680)
context can arrange hierarchically and so on. And so you can say things about, let's say,
Douglas Lenat (28:25.760)
things that are true in the modern era, things that are true in a particular year would then be
Douglas Lenat (28:32.320)
a subcontext of the things that are true in a broader, let's say, a century or a millennium
Douglas Lenat (28:39.280)
or something like that. Things that are true in Austin, Texas are generally going to be a
Douglas Lenat (28:44.320)
specialization of things that are true in Texas, which is going to be a specialization of things
Douglas Lenat (28:50.640)
that are true in the United States and so on. And so you don't have to say things over and over
Douglas Lenat (28:56.240)
again at all these levels. You just say things at the most general level that it applies to,
Lex Fridman (29:02.480)
and you only have to say it once, and then it essentially inherits to all these more specific
Lex Fridman (29:07.440)
contexts. To ask a slightly technical question, is this inheritance a tree or a graph?
Douglas Lenat (29:15.360)
Oh, you definitely have to think of it as a graph. So we could talk about, for instance,
Lex Fridman (29:20.400)
why the Japanese fifth generation computing effort failed. There were about half a dozen
Douglas Lenat (29:25.120)
different reasons. One of the reasons they failed was because they tried to represent
Douglas Lenat (29:30.320)
knowledge as a tree rather than as a graph. And so each node in their representation
Douglas Lenat (29:39.280)
could only have one parent node. So if you had a table that was a wooden object, a black object,
Douglas Lenat (29:46.160)
a flat object, and so on, you have to choose one, and that's the only parent it could have.
Douglas Lenat (29:52.560)
When, of course, depending on what it is you need to reason about it, sometimes it's important
Douglas Lenat (29:57.520)
to know that it's made out of wood, like if we're talking about a fire. Sometimes it's important to
Douglas Lenat (2:00:05.720)
problem of how can the system reason efficiently with what it knows. And so instead of trying to
Douglas Lenat (2:00:12.640)
pick one representation language which was the sweet spot or the best tradeoff point between
Douglas Lenat (2:00:20.480)
expressiveness of the language and efficiency of the language, if you had to pick one,
Douglas Lenat (2:00:25.720)
knowledge graphs would probably be, associative triples would probably be about the best you
Douglas Lenat (2:00:31.600)
could do. And that's why we started there. But after a few years, we realized that what we could
Douglas Lenat (2:00:37.560)
do is we could split this and we could have one nice, clean, epistemological level language,
Douglas Lenat (2:00:44.480)
which is this higher order logic, and we could have one or more grubby but efficient heuristic
Douglas Lenat (2:00:52.560)
level modules that opportunistically would say, oh, I can make progress on what you're trying to
Douglas Lenat (2:01:00.000)
do over here. I have a special method that will contribute a little bit toward a solution.
Lex Fridman (2:01:05.680)
Of course, some subset of that knowledge.
Douglas Lenat (2:01:09.000)
Exactly. So by now, we have over a thousand of these heuristic level modules, and they function
Douglas Lenat (2:01:14.400)
as a kind of community of agents. And there's one of them, which is a general theorem prover. And in
Douglas Lenat (2:01:20.880)
theory, that's the only one you need. But in practice, it always takes so long that you never
Douglas Lenat (2:01:29.360)
want to call on it. You always want these other agents to very efficiently reason through it. It's
Douglas Lenat (2:01:35.240)
sort of like if you're balancing a chemical equation. You could go back to first principles,
Lex Fridman (2:01:39.840)
but in fact, there are algorithms which are vastly more efficient. Or if you're trying to
Douglas Lenat (2:01:44.920)
solve a quadratic equation, you could go back to first principles of mathematics. But it's much
Douglas Lenat (2:01:52.280)
better to simply recognize that this is a quadratic equation and apply the binomial formula and snap,
Douglas Lenat (2:01:58.400)
you get your answer right away and so on. So think of these as like a thousand little experts
Douglas Lenat (2:02:04.800)
that are all looking at everything the site gets asked and looking at everything that every other
Douglas Lenat (2:02:11.200)
little agent has contributed, almost like notes on a blackboard, notes on a whiteboard, and making
Douglas Lenat (2:02:19.160)
additional notes when they think they can be helpful. And gradually, that community of agents
Douglas Lenat (2:02:24.320)
gets an answer to your question, gets a solution to your problem. And if we ever come up in a domain
Douglas Lenat (2:02:31.480)
application where Psych is getting the right answer but taking too long, then what we'll often
Douglas Lenat (2:02:38.120)
do is talk to one of the human experts and say, here's the set of reasoning steps that Psych went
Douglas Lenat (2:02:45.640)
through. You can see why it took it a long time to get the answer. How is it that you were able
Douglas Lenat (2:02:50.640)
to answer that question in two seconds? And occasionally, you'll get an expert who just
Douglas Lenat (2:02:57.600)
says, well, I just know it. I just was able to do it or something. And then you don't talk to them
Douglas Lenat (2:03:02.440)
anymore. But sometimes you'll get an expert who says, well, let me introspect on that. Yes,
Douglas Lenat (2:03:07.680)
here is a special representation we use just for aqueous chemistry equations, or here's a special
Douglas Lenat (2:03:15.320)
representation and a special technique, which we can now apply to things in this special
Douglas Lenat (2:03:21.040)
representation and so on. And then you add that as the 1001st HL heuristic level module. And from
Douglas Lenat (2:03:29.080)
then on, in any application, if it ever comes up again, it'll be able to contribute and so on. So
Douglas Lenat (2:03:35.200)
that's pretty much one of the main ways in which Psych has recouped this loss deficiency. A second
Douglas Lenat (2:03:43.240)
important way is meta reasoning. So you can speed things up by focusing on removing knowledge from
Douglas Lenat (2:03:52.760)
the system till all it has left is minimal knowledge needed. But that's the wrong thing to
Douglas Lenat (2:03:58.000)
do, right? That would be like in a human extirpating part of their brain or something. That's really
Douglas Lenat (2:04:02.360)
bad. So instead, what you want to do is give it meta level advice, tactical and strategic advice,
Douglas Lenat (2:04:08.600)
that enables it to reason about what kind of knowledge is going to be relevant to this problem,
Lex Fridman (2:04:15.320)
what kind of tactics are going to be good to take in trying to attack this problem. When is it time
Douglas Lenat (2:04:21.040)
to start trying to prove the negation of this thing, because I'm knocking myself out trying to
Douglas Lenat (2:04:26.600)
prove it's true, and maybe it's false. And if I just spend a minute, I can see that it's false
Douglas Lenat (2:04:30.480)
or something. So it's like dynamically pruning the graph to only like, based on the particular
Douglas Lenat (2:04:37.680)
thing you're trying to infer. Yes. And so by now, we have about 150 of these sort of like
Douglas Lenat (2:04:45.960)
breakthrough ideas that have led to dramatic speed ups in the inference process, where one
Douglas Lenat (2:04:53.120)
of them was this ELHL split and lots of HL modules. Another one was using meta and meta
Douglas Lenat (2:04:59.800)
level reasoning to reason about the reasoning that's going on and so on. And 150 breakthroughs
Lex Fridman (2:05:08.080)
may sound like a lot, but if you divide by 37 years, it's not as impressive.
Lex Fridman (2:05:12.160)
So there's these kind of heuristic modules that really help improve the inference. How hard,
Douglas Lenat (2:05:21.200)
in general, is this? Because you mentioned higher order logic. In the general theorem prover sense,
Douglas Lenat (2:05:30.880)
it's intractable, very difficult problem. Yes. So how hard is this inference problem when we're not
Douglas Lenat (2:05:37.760)
talking about if we let go of the perfect and focus on the good? I would say it's half of the
Douglas Lenat (2:05:46.680)
problem in the following empirical sense, which is over the years, about half of our effort,
Douglas Lenat (2:05:54.320)
maybe 40% of our effort has been our team of inference programmers. And the other 50,
Douglas Lenat (2:06:02.160)
60% has been our ontologists or ontological engineers putting in knowledge. So our ontological
Douglas Lenat (2:06:08.280)
engineers in most cases don't even know how to program. They have degrees in things like
Douglas Lenat (2:06:12.840)
philosophy and so on. So it's almost like the... I love that. I love to hang out with
Douglas Lenat (2:06:17.520)
those people actually. Oh yes, it's wonderful. But it's very much like the Eloy and the Morlocks
Douglas Lenat (2:06:22.000)
in H.G. Wells Time Machine. So you have the Eloy who only program in the epistemological higher
Douglas Lenat (2:06:29.760)
order logic language. And then you have the Morlocks who are under the ground figuring
Douglas Lenat (2:06:36.800)
out what the machinery is that will make this efficiently operate and so on. And so, you know,
Douglas Lenat (2:06:43.440)
occasionally they'll toss messages back to each other and so on. But it really is almost this
Douglas Lenat (2:06:49.640)
50 50 split between finding clever ways to recoup efficiency when you have an expressive language
Lex Fridman (2:06:57.200)
and putting in the content of what the system needs to know. And yeah, both are fascinating.
Douglas Lenat (2:07:03.200)
To some degree, the entirety of the system, as far as I understand, is written in various variants
Douglas Lenat (2:07:10.560)
of Lisp. So my favorite program language is still Lisp. I don't program in it much anymore because,
Douglas Lenat (2:07:17.200)
you know, the world has in majority of its system has moved on. Like everybody respects Lisp,
Lex Fridman (2:07:24.840)
but many of the systems are not written in Lisp anymore. But Syke, as far as I understand,
Douglas Lenat (2:07:30.920)
maybe you can correct me, there's a bunch of Lisp in it. Yeah. So it's based on Lisp code that we
Douglas Lenat (2:07:37.200)
produced. Most of the programming is still going on in a dialect of Lisp. And then for efficiency
Douglas Lenat (2:07:44.920)
reasons, that gets automatically translated into things like Java or C. Nowadays, it's almost all
Douglas Lenat (2:07:51.800)
translated into Java because Java has gotten good enough that that's really all we need to do.
Lex Fridman (2:07:58.560)
So it's translated into Java, and then Java is compiled down to bytecode.
Lex Fridman (2:08:02.600)
Yes.
Douglas Lenat (2:08:03.040)
Okay, so that's sort of that's a that that that's a, you know, it's a process that probably has to
Douglas Lenat (2:08:11.280)
do with the fact that when Syke was originally written, and you build up a powerful system,
Douglas Lenat (2:08:16.160)
like there is some technical depth you have to deal with, as is the case with most powerful
Douglas Lenat (2:08:22.520)
systems that span years. Have you ever considered this, this would help me understand, because my
Douglas Lenat (2:08:31.200)
perspective, so much of the value of everything you've done with Syke and Cycorp is the is the
Douglas Lenat (2:08:38.720)
is the knowledge. Have you ever considered just like throwing away the code base and starting
Douglas Lenat (2:08:44.640)
from scratch, not really throwing away, but sort of moving it to like throwing away that technical
Douglas Lenat (2:08:53.480)
debt, starting with a more updated programming language? Is that throwing away a lot of value
Douglas Lenat (2:08:59.800)
or no? Like, what's your sense? How much of the value is in the silly software engineering aspect,
Lex Fridman (2:09:05.400)
and how much of the value is in the knowledge?
Lex Fridman (2:09:07.720)
So development of programs in Lisp proceeds, I think, somewhere between a thousand and fifty
Douglas Lenat (2:09:21.840)
thousand times faster than development in any of what you're calling modern or improved computer
Douglas Lenat (2:09:29.720)
languages.
Douglas Lenat (2:09:30.200)
Well, there's other functional languages like, you know, Clojure and all that. But I mean,
Douglas Lenat (2:09:34.960)
I'm with you. I like Lisp. I just wonder how many great programmers there are. There's still like...
Douglas Lenat (2:09:40.200)
Yes. So it is true when a new inference programmer comes on board, they need to learn some of Lisp.
Lex Fridman (2:09:48.760)
And in fact, we have a subset of Lisp, which we call cleverly Sub L, which is really all they
Douglas Lenat (2:09:55.080)
need to learn. And so the programming actually goes on in Sub L, not in full Lisp. And so it
Douglas Lenat (2:10:01.840)
does not take programmers very long at all to learn Sub L. And that's something which can then
Douglas Lenat (2:10:08.000)
be translated efficiently into Java. And for some of our programmers who are doing, say,
Douglas Lenat (2:10:14.620)
user interface work, then they never have to even learn Sub L. They just have to learn APIs into the
Lex Fridman (2:10:21.600)
basic psych engine.
Lex Fridman (2:10:23.440)
So you're not necessarily feeling the burden of like, it's extremely efficient. That's not a
Lex Fridman (2:10:29.920)
problem to solve. Okay.
Douglas Lenat (2:10:31.960)
Right. The other thing is, remember that we're talking about hiring programmers to do inference,
Douglas Lenat (2:10:37.320)
who are programmers interested in effectively automatic theorem proving. And so those are
Douglas Lenat (2:10:43.680)
people already predisposed to representing things in logic and so on. And Lisp really was the
Douglas Lenat (2:10:50.880)
programming language based on logic that John McCarthy and others who developed it basically
Douglas Lenat (2:10:58.360)
took the formalisms that Alonzo Church and other philosophers, other logicians, had come up with
Lex Fridman (2:11:06.120)
and basically said, can we basically make a programming language which is effectively logic?
Lex Fridman (2:11:12.840)
And so since we're talking about reasoning about expressions written in this epistemological
Douglas Lenat (2:11:22.520)
language and we're doing operations which are effectively like theorem proving type
Douglas Lenat (2:11:27.040)
operations and so on, there's a natural impedance match between Lisp and the knowledge, the
Lex Fridman (2:11:34.800)
way it's represented.
Lex Fridman (2:11:36.040)
So I guess you could say it's a perfectly logical language to use.
Lex Fridman (2:11:40.640)
Oh, yes.
Douglas Lenat (2:11:41.640)
Okay, I'm sorry.
Lex Fridman (2:11:42.640)
I'll even let you get away with that.
Douglas Lenat (2:11:46.120)
Okay, thank you. I appreciate it.
Lex Fridman (2:11:47.920)
So I'll probably use that in the future without credit.
Lex Fridman (2:11:53.240)
But no, I think the point is that the language you program in isn't really that important.
Douglas Lenat (2:12:01.340)
It's more that you have to be able to think in terms of, for instance, creating new helpful
Douglas Lenat (2:12:07.240)
HL modules and how they'll work with each other and looking at things that are taking
Douglas Lenat (2:12:13.580)
a long time and coming up with new specialized data structures that will make this efficient.
Lex Fridman (2:12:20.280)
So let me just give you one very simple example, which is when you have a transitive relation
Douglas Lenat (2:12:26.080)
like larger than, this is larger than that, which is larger than that, which is larger
Douglas Lenat (2:12:29.960)
than that.
Lex Fridman (2:12:30.960)
So the first thing must be larger than the last thing.
Douglas Lenat (2:12:33.440)
Whenever you have a transitive relation, if you're not careful, if I ask whether this
Douglas Lenat (2:12:38.660)
thing over here is larger than the thing over here, I'll have to do some kind of graph walk
Douglas Lenat (2:12:43.560)
or theorem proving that might involve like five or 10 or 20 or 30 steps.
Lex Fridman (2:12:48.680)
But if you store, redundantly store the transitive closure, the cleanly star of that transitive
Douglas Lenat (2:12:55.880)
relation, now you have this big table.
Lex Fridman (2:12:58.980)
But you can always guarantee that in one single step, you can just look up whether this is
Douglas Lenat (2:13:04.720)
larger than that.
Lex Fridman (2:13:06.700)
And so there are lots of cases where storage is cheap today.
Lex Fridman (2:13:12.600)
And so by having this extra redundant data structure, we can answer this commonly occurring
Lex Fridman (2:13:18.400)
type of question very, very efficiently.
Douglas Lenat (2:13:22.240)
Let me give you one other analogy, analog of that, which is something we call rule macro
Douglas Lenat (2:13:28.960)
predicates, which is we'll see this complicated rule and we'll notice that things very much
Douglas Lenat (2:13:36.840)
like it syntactically come up again and again and again.
Lex Fridman (2:13:41.080)
So we'll create a whole brand new relation or predicate or function that captures that
Lex Fridman (2:13:47.920)
and takes maybe not two arguments, takes maybe three, four or five arguments and so on.
Lex Fridman (2:13:54.760)
And now we have effectively converted some complicated if then rule that might have to
Douglas Lenat (2:14:04.120)
have inference done on it into some ground atomic formula, which is just the name of
Lex Fridman (2:14:10.160)
a relation and a few arguments and so on.
Lex Fridman (2:14:13.200)
And so converting commonly occurring types or schemas of rules into brand new predicates,
Lex Fridman (2:14:20.880)
brand new functions, turns out to enormously speed up the inference process.
Lex Fridman (2:14:27.520)
So now we've covered about four of the 150 good ideas I said.
Lex Fridman (2:14:32.720)
So that idea in particular is like a nice compression that turns out to be really useful.
Douglas Lenat (2:14:37.400)
That's really interesting.
Lex Fridman (2:14:38.400)
I mean, this whole thing is just fascinating from a philosophical.
Douglas Lenat (2:14:40.920)
There's part of me, I mean, it makes me a little bit sad because your work is both from
Douglas Lenat (2:14:48.200)
a computer science perspective fascinating and the inference engine from a epistemological
Douglas Lenat (2:14:53.600)
philosophical aspect fascinating, but you know, it is also you're running a company
Lex Fridman (2:14:59.400)
and there's some stuff that has to remain private and it's sad.
Douglas Lenat (2:15:03.560)
Well here's something that may make you feel better, a little bit better.
Douglas Lenat (2:15:09.080)
We've formed a not for profit company called the Knowledge Axe Immunization Institute,
Douglas Lenat (2:15:15.200)
NAX, KNAX.
Lex Fridman (2:15:17.040)
And I have this firm belief with a lot of empirical evidence to support it that the
Douglas Lenat (2:15:25.440)
education that people get in high schools, in colleges, in graduate schools and so on
Douglas Lenat (2:15:31.440)
is almost completely orthogonal to, almost completely irrelevant to how good they're
Douglas Lenat (2:15:38.280)
going to be at coming up to speed in doing this kind of ontological engineering and writing
Lex Fridman (2:15:45.080)
these assertions and rules and so on in psych.
Lex Fridman (2:15:49.560)
And so very often we'll interview candidates who have their PhD in philosophy, who've
Lex Fridman (2:15:54.880)
taught logic for years and so on, and they're just awful.
Lex Fridman (2:15:59.680)
But the converse is true.
Lex Fridman (2:16:00.840)
So one of the best ontological engineers we ever had never graduated high school.
Lex Fridman (2:16:06.340)
And so the purpose of Knowledge Axe Immunization Institute, if we can get some foundations
Douglas Lenat (2:16:13.560)
to help support it is identify people in the general population, maybe high school dropouts,
Douglas Lenat (2:16:20.640)
who have latent talent for this sort of thing, offer them effectively scholarships to train
Douglas Lenat (2:16:28.360)
them and then help place them in companies that need more trained ontological engineers,
Douglas Lenat (2:16:35.280)
some of which would be working for us, but mostly would be working for partners or customers
Lex Fridman (2:16:39.560)
or something.
Lex Fridman (2:16:40.680)
And if we could do that, that would create an enormous number of relatively very high
Douglas Lenat (2:16:46.080)
paying jobs for people who currently have no way out of some situation that they're
Douglas Lenat (2:16:53.880)
locked into.
Lex Fridman (2:16:55.100)
So is there something you can put into words that describes somebody who would be great
Lex Fridman (2:17:01.080)
at ontological engineering?
Lex Fridman (2:17:03.280)
So what characteristics about a person make them great at this task, this task of converting
Lex Fridman (2:17:12.200)
the messiness of human language and knowledge into formal logic?
Douglas Lenat (2:17:17.240)
This is very much like what Alan Turing had to do during World War II in trying to find
Douglas Lenat (2:17:22.760)
people to bring to Bletchley Park, where he would publish in the London Times cryptic
Douglas Lenat (2:17:28.400)
crossword puzzles along with some innocuous looking note, which essentially said, if you
Douglas Lenat (2:17:34.760)
were able to solve this puzzle in less than 15 minutes, please call this phone number
Lex Fridman (2:17:40.820)
and so on.
Douglas Lenat (2:17:42.960)
Or back when I was young, there was the practice of having matchbooks, where on the inside
Lex Fridman (2:17:49.840)
of the matchbook, there would be a, can you draw this?
Douglas Lenat (2:17:54.040)
You have a career in art, commercial art, if you can copy this drawing and so on.
Lex Fridman (2:18:00.100)
So yes, the analog of that.
Lex Fridman (2:18:02.400)
Is there a little test to get to the core of whether you're going to be good or not?
Lex Fridman (2:18:06.000)
So part of it has to do with being able to make and appreciate and react negatively appropriately
Douglas Lenat (2:18:14.020)
to puns and other jokes.
Lex Fridman (2:18:16.160)
So you have to have a kind of sense of humor.
Lex Fridman (2:18:18.740)
And if you're good at telling jokes and good at understanding jokes, that's one
Lex Fridman (2:18:24.640)
indicator.
Lex Fridman (2:18:25.640)
Like puns?
Lex Fridman (2:18:26.640)
Yes.
Lex Fridman (2:18:27.640)
Like dad jokes?
Lex Fridman (2:18:28.640)
Yes.
Douglas Lenat (2:18:29.640)
Well, maybe not dad jokes, but funny jokes.
Lex Fridman (2:18:32.360)
I think I'm applying to work at SACOR.
Douglas Lenat (2:18:36.360)
Another is if you're able to introspect.
Lex Fridman (2:18:38.240)
So very often, we'll give someone a simple question and we'll say like, why is this?
Lex Fridman (2:18:48.360)
And sometimes they'll just say, because it is, okay, that's a bad sign.
Lex Fridman (2:18:53.480)
But very often, they'll be able to introspect and so on.
Lex Fridman (2:18:56.560)
So one of the questions I often ask is I'll point to a sentence with a pronoun in it and
Lex Fridman (2:19:01.800)
I'll say, the referent of that pronoun is obviously this noun over here.
Lex Fridman (2:19:07.700)
How would you or I or an AI or a five year old, 10 year old child know that that pronoun
Lex Fridman (2:19:14.980)
refers to that noun over here?
Lex Fridman (2:19:18.240)
And often the people who are going to be good at ontological engineering will give me some
Lex Fridman (2:19:25.760)
causal explanation or will refer to some things that are true in the world.
Lex Fridman (2:19:30.100)
So if you imagine a sentence like, the horse was led into the barn while its head was still
Lex Fridman (2:19:35.120)
wet.
Lex Fridman (2:19:36.120)
And so its head refers to the horse's head.
Lex Fridman (2:19:38.980)
But how do you know that?
Lex Fridman (2:19:40.420)
And so some people will say, I just know it.
Lex Fridman (2:19:42.140)
Some people will say, well, the horse was the subject of the sentence.
Lex Fridman (2:19:45.620)
And I'll say, okay, well, what about the horse was led into the barn while its roof was still
Lex Fridman (2:19:50.080)
wet?
Douglas Lenat (2:19:51.080)
Now, its roof obviously refers to the barn.
Lex Fridman (2:19:54.360)
And so then they'll say, oh, well, that's because it's the closest noun.
Lex Fridman (2:19:58.440)
And so basically, if they try to give me answers which are based on syntax and grammar and
Lex Fridman (2:20:05.380)
so on, that's a really bad sign.
Lex Fridman (2:20:07.580)
But if they're able to say things like, well, horses have heads and barns don't and barns
Douglas Lenat (2:20:12.120)
have roofs and horses don't, then that's a positive sign that they're going to be good
Douglas Lenat (2:20:16.860)
at this because they can introspect on what's true in the world that leads you to know certain
Lex Fridman (2:20:22.180)
things.
Lex Fridman (2:20:23.180)
How fascinating is it that getting a Ph.D. makes you less capable to introspect deeply
Lex Fridman (2:20:28.060)
about this?
Douglas Lenat (2:20:29.060)
Oh, I wouldn't go that far.
Lex Fridman (2:20:30.860)
I'm not saying that it makes you less capable.
Douglas Lenat (2:20:32.820)
Let's just say it's independent of how good people are.
Lex Fridman (2:20:37.500)
You're not saying that.
Douglas Lenat (2:20:38.500)
I'm saying that.
Douglas Lenat (2:20:39.500)
It's interesting that for a lot of people, Ph.D.s, sorry, philosophy aside, that sometimes
Douglas Lenat (2:20:47.740)
education narrows your thinking versus expands it.
Lex Fridman (2:20:52.060)
It's kind of fascinating.
Lex Fridman (2:20:53.600)
And for certain when you're trying to do ontological engineering, which is essentially teach our
Douglas Lenat (2:20:58.900)
future AI overlords how to reason deeply about this world and how to understand it, that
Douglas Lenat (2:21:05.980)
requires that you think deeply about the world.
Lex Fridman (2:21:08.700)
So I'll tell you a sad story about mathcraft, which is why is that not widely used in schools
Lex Fridman (2:21:14.820)
today?
Lex Fridman (2:21:16.100)
We're not really trying to make big profit on it or anything like that.
Douglas Lenat (2:21:20.700)
When we've gone to schools, their attitude has been, well, if a student spends 20 hours
Douglas Lenat (2:21:27.060)
going through this mathcraft program from start to end and so on, will it improve their
Douglas Lenat (2:21:34.220)
score on this standardized test more than if they spent 20 hours just doing mindless
Lex Fridman (2:21:39.980)
drills of problem after problem after problem?
Lex Fridman (2:21:43.700)
And the answer is, well, no, but it'll increase their understanding more.
Lex Fridman (2:21:47.820)
And their attitude is, well, if it doesn't increase their score on this test, then we're
Douglas Lenat (2:21:54.260)
not going to adopt it.
Lex Fridman (2:21:55.660)
That's sad.
Douglas Lenat (2:21:56.660)
I mean, that's a whole another three, four hour conversation about the education system.
Lex Fridman (2:22:01.900)
But let me ask you, let me go super philosophical, as if we weren't already.
Lex Fridman (2:22:06.660)
So in 1950, Alan Turing wrote the paper that formulated the Turing test.
Lex Fridman (2:22:11.620)
Yes.
Lex Fridman (2:22:12.620)
And he opened the paper with the question, can machines think?
Lex Fridman (2:22:16.020)
So what do you think?
Lex Fridman (2:22:17.020)
Can machines think?
Lex Fridman (2:22:18.020)
Let me ask you this question.
Douglas Lenat (2:22:20.820)
Absolutely.
Lex Fridman (2:22:21.980)
Machines can think, certainly as well as humans can think, right?
Douglas Lenat (2:22:27.980)
We're meat machines just because they're not currently made out of meat is just an engineering
Lex Fridman (2:22:34.380)
solution decision and so on.
Lex Fridman (2:22:38.300)
So of course machines can think.
Douglas Lenat (2:22:42.180)
I think that there was a lot of damage done by people misunderstanding Turing's imitation
Douglas Lenat (2:22:51.860)
game and focus on trying to get a chat bot to fool other people into thinking it was
Lex Fridman (2:23:03.220)
human and so on.
Douglas Lenat (2:23:06.540)
That's not a terrible test in and of itself, but it shouldn't be your one and only test
Lex Fridman (2:23:10.980)
for intelligence.
Douglas Lenat (2:23:13.380)
In terms of tests of intelligence, you know, with the Lobner Prize, which is a very kind
Douglas Lenat (2:23:19.180)
of, you want to say a more strict formulation of the Turing test as originally formulated.
Lex Fridman (2:23:25.340)
And then there's something like Alexa Prize, which is more, I would say a more interesting
Douglas Lenat (2:23:31.180)
formulation of the test, which is like, ultimately the metric is how long does a human want to
Lex Fridman (2:23:37.740)
talk to the AI system?
Lex Fridman (2:23:38.740)
So it's like if the goal is you want it to be 20 minutes, it's basically not just have
Douglas Lenat (2:23:46.060)
a convincing conversation, but more like a compelling one or a fun one or an interesting
Lex Fridman (2:23:52.620)
one.
Lex Fridman (2:23:53.620)
And that seems like more to the spirit maybe of what Turing was imagining.
Lex Fridman (2:24:01.120)
But what for you do you think in the space of tests is a good test?
Douglas Lenat (2:24:06.700)
When you see a system based on psych that passes that test, you'd be like, damn, we've
Lex Fridman (2:24:12.900)
created something special here.
Douglas Lenat (2:24:17.020)
The test has to be something involving depth of reasoning and recursiveness of reasoning,
Lex Fridman (2:24:23.700)
the ability to answer repeated why questions about the answer you just gave.
Lex Fridman (2:24:30.100)
How many why questions in a row can you keep answering?
Lex Fridman (2:24:33.140)
Something like that.
Douglas Lenat (2:24:36.140)
Just have like a young curious child and an AI system and how long will an AI system last
Lex Fridman (2:24:41.820)
before it wants to quit?
Douglas Lenat (2:24:43.300)
Yes.
Lex Fridman (2:24:44.300)
And again, that's not the only test.
Douglas Lenat (2:24:45.660)
Another one has to do with argumentation.
Douglas Lenat (2:24:48.020)
In other words, here's a proposition, come up with pro and con arguments for it and try
Lex Fridman (2:24:57.020)
and give me convincing arguments on both sides.
Lex Fridman (2:25:02.300)
And so that's another important kind of ability that the system needs to be able to exhibit
Douglas Lenat (2:25:09.660)
in order to really be intelligent, I think.
Lex Fridman (2:25:12.860)
So there's certain, I mean, if you look at IBM Watson and like certain impressive accomplishments
Lex Fridman (2:25:18.180)
for very specific tests, almost like a demo, right?
Douglas Lenat (2:25:24.500)
There's some, like I talked to the guy who led the Jeopardy effort, and there's some
Douglas Lenat (2:25:34.740)
kind of hard coding heuristics tricks that you try to pull it all together to make the
Lex Fridman (2:25:40.260)
thing work in the end for this thing, right?
Douglas Lenat (2:25:43.060)
That seems to be one of the lessons with AI is like, that's the fastest way to get a solution
Lex Fridman (2:25:49.060)
that's pretty damn impressive.
Lex Fridman (2:25:50.580)
So here's what I would say is that as impressive as that was, it made some mistakes, but more
Douglas Lenat (2:25:59.620)
importantly, many of the mistakes it made were mistakes which no human would have made.
Lex Fridman (2:26:07.260)
And so part of the new or augmented Turing tests would have to be, and the mistakes you
Lex Fridman (2:26:17.540)
make are ones which humans don't basically look at and say, what?
Lex Fridman (2:26:24.300)
So for example, there was a question about which 16th century Italian politician, blah,
Lex Fridman (2:26:33.540)
blah, blah, and Watson said Ronald Reagan.
Lex Fridman (2:26:37.260)
So most Americans would have gotten that question wrong, but they would never have said Ronald
Douglas Lenat (2:26:42.000)
Reagan as an answer because among the things they know is that he lived relatively recently
Lex Fridman (2:26:49.860)
and people don't really live 400 years and things like that.
Lex Fridman (2:26:53.960)
So that's, I think, a very important thing, which is if it's making mistakes which no
Douglas Lenat (2:27:00.860)
normal sane human would have made, then that's a really bad sign.
Lex Fridman (2:27:05.780)
And if it's not making those kinds of mistakes, then that's a good sign.
Lex Fridman (2:27:10.140)
And I don't think it's any one very, very simple test.
Douglas Lenat (2:27:12.980)
I think it's all of the things you mentioned, all the things I mentioned is really a battery
Douglas Lenat (2:27:17.300)
of tests, which together, if it passes almost all of these tests, it'd be hard to argue
Lex Fridman (2:27:23.700)
that it's not intelligent.
Lex Fridman (2:27:25.340)
And if it fails several of these tests, it's really hard to argue that it really understands
Lex Fridman (2:27:30.980)
what it's doing and that it really is generally intelligent.
Lex Fridman (2:27:33.460)
So to pass all of those tests, we've talked a lot about psych and knowledge and reasoning.
Lex Fridman (2:27:40.820)
Do you think this AI system would need to have some other human like elements, for example,
Lex Fridman (2:27:47.540)
a body or a physical manifestation in this world?
Lex Fridman (2:27:52.820)
And another one which seems to be fundamental to the human experience is consciousness.
Douglas Lenat (2:27:59.980)
The subjective experience of what it's like to actually be you.
Lex Fridman (2:28:04.740)
Do you think it needs those to be able to pass all of those tests and to achieve general
Lex Fridman (2:28:08.700)
intelligence?
Lex Fridman (2:28:09.700)
It's a good question.
Douglas Lenat (2:28:10.700)
I think in the case of a body, no, I know there are a lot of people like Penrose who
Douglas Lenat (2:28:15.340)
would have disagreed with me and others, but no, I don't think it needs to have a body
Douglas Lenat (2:28:21.940)
in order to be intelligent.
Douglas Lenat (2:28:24.300)
I think that it needs to be able to talk about having a body and having sensations and having
Douglas Lenat (2:28:32.780)
emotions and so on.
Douglas Lenat (2:28:33.980)
It doesn't actually have to have all of that, but it has to understand it in the same way
Douglas Lenat (2:28:39.380)
that Helen Keller was perfectly intelligent and able to talk about colors and sounds and
Douglas Lenat (2:28:47.300)
shapes and so on, even though she didn't directly experience all the same things that the rest
Douglas Lenat (2:28:54.100)
of us do.
Lex Fridman (2:28:55.100)
So knowledge of it and being able to correctly make use of that is certainly an important
Douglas Lenat (2:29:04.060)
facility, but actually having a body, if you believe that that's just a kind of religious
Lex Fridman (2:29:09.420)
or mystical belief, you can't really argue for or against it, I suppose.
Douglas Lenat (2:29:15.740)
It's just something that some people believe.
Lex Fridman (2:29:19.340)
What about an extension of the body, which is consciousness?
Douglas Lenat (2:29:24.540)
It feels like something to be here.
Lex Fridman (2:29:27.780)
Sure.
Lex Fridman (2:29:28.780)
But what does that really mean?
Douglas Lenat (2:29:30.820)
It's like, well, if I talk to you, you say things which make me believe that you're conscious.
Douglas Lenat (2:29:35.940)
I know that I'm conscious, but you're just taking my word for it now.
Lex Fridman (2:29:40.540)
But in the same sense, psych is conscious in that same sense already, where of course
Douglas Lenat (2:29:46.620)
it's a computer program, it understands where and when it's running, it understands who's
Douglas Lenat (2:29:50.620)
talking to it, it understands what its task is, what its goals are, what its current problem
Douglas Lenat (2:29:55.300)
is that it's working on, it understands how long it's spent on things, what it's tried,
Lex Fridman (2:29:59.600)
it understands what it's done in the past, and so on.
Douglas Lenat (2:30:06.300)
If we want to call that consciousness, then yes, psych is already conscious.
Lex Fridman (2:30:11.340)
But I don't think that I would ascribe anything mystical to that.
Douglas Lenat (2:30:15.700)
Again, some people would, but I would say that other than our own personal experience
Douglas Lenat (2:30:21.220)
of consciousness, we're just treating everyone else in the world, so to speak, at their word
Douglas Lenat (2:30:27.980)
about being conscious.
Lex Fridman (2:30:29.820)
And so if a computer program, if an AI is able to exhibit all the same kinds of response
Douglas Lenat (2:30:39.300)
as you would expect of a conscious entity, then doesn't it deserve the label of consciousness
Lex Fridman (2:30:46.500)
just as much?
Lex Fridman (2:30:47.500)
So there's another burden that comes with this whole intelligence thing that humans
Douglas Lenat (2:30:51.200)
got is the extinguishing of the light of consciousness, which is kind of realizing that we're going
Douglas Lenat (2:30:59.780)
to be dead someday.
Lex Fridman (2:31:02.660)
And there's a bunch of philosophers like Ernest Becker, who kind of think that this realization
Douglas Lenat (2:31:09.060)
of mortality, and then fear, sometimes they call it terror of mortality, is one of the
Lex Fridman (2:31:18.180)
creative forces behind human condition, like, it's the thing that drives us.
Lex Fridman (2:31:24.980)
Do you think it's important for an AI system?
Douglas Lenat (2:31:27.900)
You know, when Psych proposed that it's not human, and it's one of the moderators of
Douglas Lenat (2:31:36.660)
his contents, you know, there's another question it could ask, which is like, it kind of knows
Lex Fridman (2:31:43.980)
that humans are mortal, am I mortal?
Lex Fridman (2:31:47.760)
And I think one really important thing that's possible when you're conscious is to fear
Lex Fridman (2:31:54.620)
the extinguishing of that consciousness, the fear of mortality.
Lex Fridman (2:31:59.020)
Do you think that's useful for intelligence, thinking like, I might die, and I really don't
Lex Fridman (2:32:04.340)
want to die?
Douglas Lenat (2:32:05.340)
I don't think so.
Lex Fridman (2:32:06.660)
I think it may help some humans to be better people.
Douglas Lenat (2:32:12.940)
It may help some humans to be more creative, and so on.
Douglas Lenat (2:32:16.140)
I don't think it's necessary for AIs to believe that they have limited lifespans, and therefore
Douglas Lenat (2:32:23.660)
they should make the most of their behavior.
Douglas Lenat (2:32:26.020)
Maybe eventually the answer to that and my answer to that will change, but as of now
Douglas Lenat (2:32:30.980)
I would say that that's almost like a frill or a side effect that is not, in fact, if
Douglas Lenat (2:32:36.980)
you look at most humans, most humans ignore the fact that they're going to die most of
Douglas Lenat (2:32:42.780)
the time.
Lex Fridman (2:32:43.780)
Well, but that's like, this goes to the white space between the words.
Lex Fridman (2:32:49.700)
So what Ernest Becker argues is that that ignoring is we're living in an illusion that
Lex Fridman (2:32:54.300)
we constructed on the foundation of this terror.
Lex Fridman (2:32:57.780)
So we're escape life as we know it, pursuing things, creating things, love, everything
Douglas Lenat (2:33:05.100)
we can think of that's beautiful about humanity is just trying to escape this realization
Douglas Lenat (2:33:11.620)
that we're going to die one day.
Douglas Lenat (2:33:13.420)
That's his idea, and I think, I don't know if I 100% believe in this, but it certainly
Douglas Lenat (2:33:21.420)
rhymes.
Lex Fridman (2:33:22.700)
It seems like to me like it rhymes with the truth.
Douglas Lenat (2:33:25.620)
Yeah.
Douglas Lenat (2:33:26.620)
I think that for some people that's going to be a more powerful factor than others.
Douglas Lenat (2:33:33.340)
Clearly Doug is talking about Russians.
Lex Fridman (2:33:35.860)
So I'm Russian, so clearly it infiltrates all of Russian literature.
Lex Fridman (2:33:44.700)
And AI doesn't have to have fear of death as a motivating force in that we can build
Lex Fridman (2:33:53.860)
in motivation.
Lex Fridman (2:33:55.700)
So we can build in the motivation of obeying users and making users happy and making others
Douglas Lenat (2:34:03.180)
happy and so on, and that can substitute for this sort of personal fear of death that sometimes
Douglas Lenat (2:34:12.380)
leads to bursts of creativity in humans.
Lex Fridman (2:34:16.940)
Yeah, I don't know.
Douglas Lenat (2:34:18.780)
I think AI really needs to understand death deeply in order to be able to drive a car,
Lex Fridman (2:34:23.420)
for example.
Douglas Lenat (2:34:24.860)
I think there's just some, like, there...
Lex Fridman (2:34:28.100)
No, I really disagree.
Douglas Lenat (2:34:30.060)
I think it needs to understand the value of human life, especially the value of human
Douglas Lenat (2:34:34.740)
life to other humans, and understand that certain things are more important than other
Douglas Lenat (2:34:41.940)
things.
Lex Fridman (2:34:42.940)
So it has to have a lot of knowledge about ethics and morality and so on.
Lex Fridman (2:34:48.060)
But some of it is so messy that it's impossible to encode.
Lex Fridman (2:34:51.220)
For example, there's...
Douglas Lenat (2:34:52.220)
I disagree.
Lex Fridman (2:34:53.780)
So if there's a person dying right in front of us, most human beings would help that person,
Lex Fridman (2:34:59.320)
but they would not apply that same ethics to everybody else in the world.
Douglas Lenat (2:35:04.580)
This is the tragedy of how difficult it is to be a doctor, because they know when they
Douglas Lenat (2:35:09.100)
help a dying child, they know that the money they're spending on this child cannot possibly
Lex Fridman (2:35:15.720)
be spent on every other child that's dying.
Lex Fridman (2:35:18.920)
And that's a very difficult to encode decision.
Lex Fridman (2:35:24.700)
Perhaps it is...
Douglas Lenat (2:35:26.540)
Perhaps it could be formalized.
Lex Fridman (2:35:27.540)
Oh, but I mean, you're talking about autonomous vehicles, right?
Lex Fridman (2:35:31.900)
So autonomous vehicles are going to have to make those decisions all the time of, what
Lex Fridman (2:35:39.340)
is the chance of this bad event happening?
Lex Fridman (2:35:43.060)
How bad is that compared to this chance of that bad event happening?
Lex Fridman (2:35:46.660)
And so on.
Lex Fridman (2:35:47.660)
And when a potential accident is about to happen, is it worth taking this risk?
Lex Fridman (2:35:52.940)
If I have to make a choice, which of these two cars am I going to hit and why?
Douglas Lenat (2:35:56.980)
See, I was thinking about a very different choice when I'm talking about hero mortality,
Lex Fridman (2:36:01.220)
which is just observing Manhattan style driving.
Douglas Lenat (2:36:06.020)
I think that humans as an effective driver needs to threaten pedestrians lives a lot.
Douglas Lenat (2:36:14.820)
There's a dance, I've watched pedestrians a lot, I worked on this problem, and it seems
Douglas Lenat (2:36:19.500)
like the, if I could summarize the problem of a pedestrian crossing is the car with this
Lex Fridman (2:36:27.220)
movement is saying, I'm going to kill you.
Lex Fridman (2:36:30.240)
And the pedestrian is saying, maybe.
Lex Fridman (2:36:33.320)
And then they decide and they say, no, I don't think you have the guts to kill me.
Lex Fridman (2:36:36.780)
And you walk and they walk in front and they look away.
Lex Fridman (2:36:39.460)
And there's that dance, the pedestrian, this is a social contract that the pedestrian trusts
Douglas Lenat (2:36:46.300)
that once they're in front of the car and the car is sufficiently, from a physics perspective,
Lex Fridman (2:36:51.020)
able to stop, they're going to stop.
Lex Fridman (2:36:53.220)
But the car also has to threaten that pedestrian is like, I'm late for work, so you're being
Lex Fridman (2:36:58.540)
kind of an asshole by crossing in front of me.
Lex Fridman (2:37:01.140)
But life and death is in like, it's part of the calculation here.
Lex Fridman (2:37:06.060)
And it's that equation is being solved millions of times a day.
Douglas Lenat (2:37:11.220)
Yes.
Lex Fridman (2:37:12.220)
Very effectively, that game theory, whatever that formulation is.
Douglas Lenat (2:37:15.540)
Absolutely.
Lex Fridman (2:37:16.540)
I just I don't know if it's as simple as some formalizable game theory problem.
Douglas Lenat (2:37:22.260)
It could very well be in the case of driving and in the case of most of human society.
Lex Fridman (2:37:28.300)
I don't know.
Lex Fridman (2:37:29.300)
But yeah, you might be right that sort of the fear of death is just one of the quirks
Douglas Lenat (2:37:34.660)
of like the way our brains have evolved, but it's not a necessary feature of intelligence.
Douglas Lenat (2:37:42.500)
Others certainly are always doing this kind of estimate, even if it's unconscious, subconscious,
Lex Fridman (2:37:48.740)
of what are the chances of various bad outcomes happening?
Douglas Lenat (2:37:52.860)
Like for instance, if I don't wait for this pedestrian or something like that, and what
Douglas Lenat (2:37:59.060)
is the downside to me going to be in terms of time wasted talking to the police or getting
Lex Fridman (2:38:07.680)
sent to jail or things like that?
Lex Fridman (2:38:11.980)
And there's also emotion, like people in their cars tend to get irrationally angry.
Douglas Lenat (2:38:17.580)
That's dangerous.
Douglas Lenat (2:38:18.580)
But, you know, think about this is all part of why I think that autonomous vehicles, truly
Douglas Lenat (2:38:24.260)
autonomous vehicles are farther out than most people do, because there is this enormous
Lex Fridman (2:38:31.060)
level of complexity which goes beyond mechanically controlling the car.
Lex Fridman (2:38:38.580)
And I can see the autonomous vehicles as a kind of metaphorical and literal accident
Lex Fridman (2:38:45.220)
waiting to happen.
Lex Fridman (2:38:47.100)
And not just because of their overall incurring versus preventing accidents and so on, but
Douglas Lenat (2:38:56.900)
just because of the almost voracious appetite people have for bad stories about powerful
Douglas Lenat (2:39:10.340)
companies and powerful entities.
Douglas Lenat (2:39:12.940)
When I was at a, coincidentally, Japanese fifth generation computing system conference
Douglas Lenat (2:39:19.640)
in 1987, while I happened to be there, there was a worker at an auto plant who was despondent
Lex Fridman (2:39:26.220)
and committed suicide by climbing under the safety chains and so on and getting stamped
Douglas Lenat (2:39:30.900)
to death by a machine.
Lex Fridman (2:39:32.940)
And instead of being a small story that said despondent worker commit suicide, it was front
Douglas Lenat (2:39:38.940)
page news that effectively said robot kills worker, because the public is just waiting
Lex Fridman (2:39:46.860)
for stories about like AI kills phonogenic family of five type stories.
Lex Fridman (2:39:54.220)
And even if you could show that nationwide, this system saved more lives than it cost
Lex Fridman (2:40:01.300)
and prevented more injuries than it caused and so on, the media, the public, the government
Douglas Lenat (2:40:09.100)
is just coiled and ready to pounce on stories where in fact it failed, even if they're relatively
Lex Fridman (2:40:18.180)
few.
Douglas Lenat (2:40:19.180)
Yeah, it's so fascinating to watch us humans resisting the cutting edge of science and
Douglas Lenat (2:40:26.220)
technology and almost like hoping for it to fail and constant, you know, this just happens
Douglas Lenat (2:40:31.940)
over and over and over throughout history.
Lex Fridman (2:40:33.820)
Or even if we're not hoping for it to fail, we're fascinated by it.
Lex Fridman (2:40:37.860)
And in terms of what we find interesting, the one in a thousand failures, much more
Lex Fridman (2:40:43.420)
interesting than the 999 boring successes.
Lex Fridman (2:40:48.160)
So once we build an AGI system, say psych is some part of it and say it's very possible
Douglas Lenat (2:40:57.860)
that you would be one of the first people that can sit down in the room, let's say with
Lex Fridman (2:41:03.420)
her and have a conversation, what would you ask her?
Lex Fridman (2:41:07.260)
What would you talk about?
Douglas Lenat (2:41:09.580)
Looking at all of the content out there on the web and so on, what are some possible
Douglas Lenat (2:41:26.460)
solutions to big problems that the world has that people haven't really thought of before
Lex Fridman (2:41:33.120)
that are not being properly or at least adequately pursued?
Lex Fridman (2:41:40.000)
What are some novel solutions that you can think of that we haven't that might work and
Lex Fridman (2:41:46.540)
that might be worth considering?
Lex Fridman (2:41:48.600)
So that is a damn good question.
Douglas Lenat (2:41:51.020)
Given that the AGI is going to be somewhat different from human intelligence, it's still
Douglas Lenat (2:41:56.360)
going to make some mistakes that we wouldn't make, but it's also possibly going to notice
Douglas Lenat (2:42:02.020)
some blind spots we have.
Lex Fridman (2:42:04.420)
And I would love as a test of is it really on a par with our intelligences, can it help
Lex Fridman (2:42:13.460)
spot some of the blind spots that we have?
Lex Fridman (2:42:17.660)
So the two part question of can you help identify what are the big problems in the world?
Lex Fridman (2:42:23.420)
And two, what are some novel solutions to those problems?
Lex Fridman (2:42:27.360)
That are not being talked about by anyone.
Lex Fridman (2:42:31.660)
And some of those may become infeasible or reprehensible or something, but some of them
Lex Fridman (2:42:37.040)
might be actually great things to look at.
Douglas Lenat (2:42:40.640)
If you go back and look at some of the most powerful discoveries that have been made,
Douglas Lenat (2:42:45.960)
like relativity and superconductivity and so on, a lot of them were cases where someone
Douglas Lenat (2:42:56.180)
took seriously the idea that there might actually be a non obvious answer to a question.
Lex Fridman (2:43:04.620)
So in Einstein's case, it was, yeah, the Lorentz transformation is known.
Douglas Lenat (2:43:09.720)
Nobody believes that it's actually the way reality works.
Lex Fridman (2:43:12.200)
What if it were the way that reality actually worked?
Lex Fridman (2:43:15.380)
So a lot of people don't realize he didn't actually work out that equation, he just sort
Lex Fridman (2:43:19.040)
of took it seriously.
Douglas Lenat (2:43:20.940)
Or in the case of superconductivity, you have this V equals IR equation where R is resistance
Lex Fridman (2:43:26.680)
and so on.
Lex Fridman (2:43:28.000)
And it was being mapped at lower and lower temperatures, but everyone thought that was
Lex Fridman (2:43:33.480)
just bump on a log research to show that V equals IR always held.
Lex Fridman (2:43:39.840)
And then when some graduate student got to a slightly lower temperature and showed that
Lex Fridman (2:43:46.180)
resistance suddenly dropped off, everyone just assumed that they did it wrong.
Lex Fridman (2:43:50.960)
And it was only a little while later that they realized it was actually a new phenomenon.
Douglas Lenat (2:43:56.680)
Or in the case of the H. pylori bacteria causing stomach ulcers, where everyone thought that
Douglas Lenat (2:44:04.660)
stress and stomach acid caused ulcers.
Lex Fridman (2:44:08.040)
And when a doctor in Australia claimed it was actually a bacterial infection, he couldn't
Douglas Lenat (2:44:15.800)
get anyone seriously to listen to him and he had to ultimately inject himself with the
Douglas Lenat (2:44:21.880)
bacteria to show that he suddenly developed a life threatening ulcer in order to get other
Douglas Lenat (2:44:27.640)
doctors to seriously consider that.
Lex Fridman (2:44:29.880)
So there are all sorts of things where humans are locked into paradigms, what Thomas Kuhn
Douglas Lenat (2:44:35.680)
called paradigms, and we can't get out of them very easily.
Lex Fridman (2:44:40.520)
So a lot of AI is locked into the deep learning machine learning paradigm right now.
Lex Fridman (2:44:47.720)
And almost all of us and almost all sciences are locked into current paradigms.
Lex Fridman (2:44:52.800)
And Kuhn's point was pretty much you have to wait for people to die in order for the
Douglas Lenat (2:45:00.000)
new generation to escape those paradigms.
Lex Fridman (2:45:03.280)
And I think that one of the things that would change that sad reality is if we had trusted
Douglas Lenat (2:45:09.120)
AGI's that could help take a step back and question some of the paradigms that we're
Lex Fridman (2:45:16.120)
currently locked into.
Douglas Lenat (2:45:17.400)
Yeah, it would accelerate the paradigm shifts in human science and progress.
Douglas Lenat (2:45:25.160)
You've lived a very interesting life where you thought about big ideas and you stuck
Douglas Lenat (2:45:30.600)
with them.
Lex Fridman (2:45:32.600)
Can you give advice to young people today, somebody in high school, somebody undergrad,
Lex Fridman (2:45:38.520)
about career, about life?
Lex Fridman (2:45:43.880)
I'd say you can make a difference.
Lex Fridman (2:45:47.840)
But in order to make a difference, you're going to have to have the courage to follow
Douglas Lenat (2:45:53.200)
through with ideas which other people might not immediately understand or support.
Douglas Lenat (2:46:02.480)
You have to realize that if you make some plan that's going to take an extended period
Lex Fridman (2:46:12.880)
of time to carry out, don't be afraid of that.
Douglas Lenat (2:46:16.640)
That's true of physical training of your body.
Lex Fridman (2:46:20.920)
That's true of learning some profession.
Douglas Lenat (2:46:27.020)
It's also true of innovation, that some innovations are not great ideas you can write down on
Lex Fridman (2:46:33.580)
a napkin and become an instant success if you turn out to be right.
Douglas Lenat (2:46:38.500)
Some of them are paths you have to follow, but remember that you're mortal.
Douglas Lenat (2:46:45.900)
Remember that you have a limited number of decade sized debts to make with your life
Lex Fridman (2:46:53.320)
and you should make each one of them count.
Lex Fridman (2:46:55.720)
And that's true in personal relationships.
Douglas Lenat (2:46:58.160)
That's true in career choice.
Lex Fridman (2:47:00.360)
That's true in making discoveries and so on.
Lex Fridman (2:47:03.920)
And if you follow the path of least resistance, you'll find that you're optimizing for short
Lex Fridman (2:47:10.960)
periods of time.
Lex Fridman (2:47:12.800)
And before you know it, you turn around and long periods of time have gone by without
Lex Fridman (2:47:17.280)
you ever really making a difference in the world.
Douglas Lenat (2:47:21.400)
When you look at the field that I really love is artificial intelligence and there's not
Douglas Lenat (2:47:26.280)
many projects, there's not many little flames of hope that have been carried out for many
Douglas Lenat (2:47:33.040)
years, for decades and psych represents one of them.
Lex Fridman (2:47:36.920)
And I mean that in itself is just a really inspiring thing.
Lex Fridman (2:47:42.880)
So I'm deeply grateful that you would be carrying that flame for so many years and I think that's
Lex Fridman (2:47:47.960)
an inspiration to young people.
Douglas Lenat (2:47:50.040)
That said, you said life is finite and we talked about mortality as a feature of AGI.
Lex Fridman (2:47:55.360)
Do you think about your own mortality?
Lex Fridman (2:47:57.480)
Are you afraid of death?
Lex Fridman (2:47:59.480)
Sure, I'd be crazy if I weren't.
Lex Fridman (2:48:03.280)
And as I get older, I'm now over 70.
Lex Fridman (2:48:07.560)
So as I get older, it's more on my mind, especially as acquaintances and friends and especially
Douglas Lenat (2:48:14.760)
mentors, one by one are dying, so I can't avoid thinking about mortality.
Lex Fridman (2:48:22.880)
And I think that the good news from the point of view and the rest of the world is that
Douglas Lenat (2:48:28.760)
that adds impetus to my need to succeed in a small number of years in the future.
Lex Fridman (2:48:34.640)
You have a deadline.
Douglas Lenat (2:48:36.520)
Exactly.
Lex Fridman (2:48:37.520)
I'm not going to have another 37 years to continue working on this.
Lex Fridman (2:48:41.440)
So we really do want Psyche to make an impact in the world commercially, physically, metaphysically
Douglas Lenat (2:48:50.400)
in the next small number of years, two, three, five years, not two, three, five decades anymore.
Lex Fridman (2:48:56.560)
And so this is really driving me toward this sort of commercialization and increasingly
Lex Fridman (2:49:04.760)
widespread application of Psyche.
Douglas Lenat (2:49:08.080)
Whereas before, I felt that I could just sort of sit back, roll my eyes, wait till the world
Lex Fridman (2:49:13.560)
caught up.
Lex Fridman (2:49:14.560)
And now I don't feel that way anymore.
Douglas Lenat (2:49:16.600)
I feel like I need to put in some effort to make the world aware of what we have and what
Douglas Lenat (2:49:22.440)
it can do.
Lex Fridman (2:49:23.920)
And the good news from your point of view is that that's why I'm sitting here.
Douglas Lenat (2:49:27.360)
You're going to be more productive.
Lex Fridman (2:49:30.640)
I love it.
Lex Fridman (2:49:31.720)
And if I can help in any way, I would love to.
Douglas Lenat (2:49:34.360)
From a programmer perspective, I love, especially these days, just contributing in small and
Douglas Lenat (2:49:41.760)
big ways.
Lex Fridman (2:49:42.840)
So if there's any open sourcing from an MIT side and the research, I would love to help.
Lex Fridman (2:49:48.680)
But bigger than Psyche, like I said, it's that little flame that you're carrying of
Lex Fridman (2:49:53.520)
artificial intelligence, the big dream.
Lex Fridman (2:49:58.600)
What do you hope your legacy is?
Lex Fridman (2:50:02.360)
That's a good question.
Douglas Lenat (2:50:05.080)
People think of me as one of the pioneers or inventors of the AI that is ubiquitous
Lex Fridman (2:50:15.920)
and that they take for granted and so on.
Douglas Lenat (2:50:19.480)
Much the way that today we look back on the pioneers of electricity or the pioneers of
Lex Fridman (2:50:28.440)
similar types of technologies and so on.
Douglas Lenat (2:50:33.200)
It's hard to imagine what life would be like if these people hadn't done what they did.
Lex Fridman (2:50:40.000)
So that's one thing that I'd like to be remembered as.
Douglas Lenat (2:50:44.000)
Another is that the creator, one of the originators of this gigantic knowledge store and acquisition
Douglas Lenat (2:50:53.680)
system that is likely to be at the center of whatever this future AI thing will look
Douglas Lenat (2:51:00.360)
like.
Lex Fridman (2:51:01.360)
Yes, exactly.
Lex Fridman (2:51:02.360)
And I'd also like to be remembered as someone who wasn't afraid to spend several decades
Douglas Lenat (2:51:11.440)
on a project in a time when almost all of the other forces, institutional forces and
Douglas Lenat (2:51:23.400)
commercial forces, are incenting people to go for short term rewards.
Lex Fridman (2:51:29.920)
And a lot of people gave up.
Douglas Lenat (2:51:31.400)
A lot of people that dreamt the same dream as you gave up and you didn't.
Lex Fridman (2:51:40.320)
I mean, Doug, it's truly an honor.
Douglas Lenat (2:51:42.800)
This is a long time coming.
Douglas Lenat (2:51:45.200)
A lot of people bring up your work specifically and more broadly, philosophically of this
Douglas Lenat (2:51:53.120)
is the dream of artificial intelligence.
Lex Fridman (2:51:55.580)
This is likely a part of the future.
Douglas Lenat (2:51:57.800)
We're so sort of focused on machine learning applications, all that kind of stuff today.
Lex Fridman (2:52:02.080)
But it seems like the ideas that Cite carries forward is something that will be at the center
Douglas Lenat (2:52:08.520)
of this problem they're all trying to solve, which is the problem of intelligence, emotional
Lex Fridman (2:52:15.640)
and otherwise.
Lex Fridman (2:52:16.640)
So thank you so much.
Douglas Lenat (2:52:18.440)
It's such a huge honor that you would talk to me and spend your valuable time with me
Douglas Lenat (2:52:22.760)
today.
Lex Fridman (2:52:23.760)
Thanks for talking.
Douglas Lenat (2:52:24.760)
Thanks, Lex.
Lex Fridman (2:52:25.760)
It's been great.
Douglas Lenat (2:52:26.760)
Thanks for listening to this conversation with Doug Lenat.
Lex Fridman (2:52:29.480)
To support this podcast, please check out our sponsors in the description.
Lex Fridman (2:52:33.760)
And now, let me leave you with some words from Mark Twain about the nature of truth.
Lex Fridman (2:52:40.160)
If you tell the truth, you don't have to remember anything.
Douglas Lenat (2:52:44.320)
Thank you for listening and hope to see you next time.
Douglas Lenat (30:02.000)
know that it's flat if we're talking about resting something on it, and so on. So one of the problems
Douglas Lenat (30:09.680)
was that they wanted a kind of Dewey decimal numbering system for all of their concepts,
Douglas Lenat (30:15.280)
which meant that each node could only have at most 10 children, and each node could only have
Douglas Lenat (30:21.600)
one parent. And while that does enable the Dewey decimal type numbering of concepts, labeling of
Douglas Lenat (30:30.800)
concepts, it prevents you from representing all the things you need to about objects in our world.
Lex Fridman (30:37.760)
And that was one of the things which they never were able to overcome, and I think that was one
Douglas Lenat (30:42.720)
of the main reasons that that project failed. So we'll return to some of the doors you've
Douglas Lenat (30:47.120)
opened, but if we can go back to that room in 1984 around there with Marvin Minsky and Stanford.
Douglas Lenat (30:53.440)
By the way, I should mention that Marvin wouldn't do his estimate until someone brought him an
Douglas Lenat (30:59.520)
envelope so that he could literally do a back of the envelope calculation to come up with his number.
Douglas Lenat (31:07.280)
Well, because I feel like the conversation in that room is an important one. You know,
Douglas Lenat (31:13.680)
this is how sometimes science is done in this way. A few people get together
Lex Fridman (31:19.040)
and plant the seed of ideas, and they reverberate throughout history.
Lex Fridman (31:23.040)
And some kind of dissipate and disappear, and some, you know, Drake equation, and, you know,
Douglas Lenat (31:29.040)
they, you know, seems like a meaningless equation, somewhat meaningless, but I think it drives and
Douglas Lenat (31:33.920)
motivates a lot of scientists. And when the aliens finally show up, that equation will get even more
Douglas Lenat (31:39.360)
valuable because then we'll get, be able to, in the long arc of history, the Drake equation
Douglas Lenat (31:45.760)
will prove to be quite useful, I think. And in that same way, a conversation of just how many facts
Douglas Lenat (31:53.920)
are required to capture the basic common sense knowledge of the world. That's a fascinating
Douglas Lenat (31:57.760)
question. I want to distinguish between what you think of as facts and the kind of things that we
Douglas Lenat (32:02.960)
represent. So we map to and essentially make sure that psych has the ability to, as it were, read
Lex Fridman (32:10.960)
and access the kind of facts you might find, say, in Wikidata or stated in a Wikipedia article or
Douglas Lenat (32:18.960)
something like that. So what we're representing, the things that we need a small number of tens
Douglas Lenat (32:23.280)
of millions of, are more like rules of thumb, rules of good guessing, things which are usually
Douglas Lenat (32:29.440)
true and which help you to make sense of the facts that are sort of sitting off in some database or
Douglas Lenat (32:37.840)
some other more static storage. So they're almost like platonic forms. So like when you read stuff
Douglas Lenat (32:43.680)
on Wikipedia, that's going to be like projections of those ideas. You read an article about the fact
Douglas Lenat (32:48.640)
that Elvis died, that's a projection of the idea that humans are mortal. And very few
Douglas Lenat (32:56.640)
Wikipedia articles will write, humans are mortal. Exactly. That's what I meant about
Douglas Lenat (33:01.920)
ferreting out the unstated things in text. What are all the things that were assumed? And so those
Douglas Lenat (33:07.440)
are things like if you have a problem with something, turning it off and on often fixes
Douglas Lenat (33:13.280)
it for reasons we don't really understand and we're not happy about. Or people can't be both
Douglas Lenat (33:18.080)
alive and dead at the same time. Or water flows downhill. If you search online for water flowing
Douglas Lenat (33:25.440)
uphill and water flowing downhill, you'll find more references for water flowing uphill because
Douglas Lenat (33:29.840)
it's used as a kind of a metaphorical reference for some unlikely thing because of course,
Lex Fridman (33:36.640)
everyone already knows that water flows downhill. So why would anyone bother saying that?
Lex Fridman (33:41.680)
Do you have a word you prefer? Because we said facts isn't the right word. Is there a word like
Douglas Lenat (33:46.800)
concepts? I would say assertions. Assertions or rules? Because I'm not talking about rigid rules,
Lex Fridman (33:53.280)
but rules of thumb. But assertions is a nice one that covers all of these things.
Douglas Lenat (33:59.760)
Yeah. As a programmer, to me, assert has a very dogmatic authoritarian feel to them.
Lex Fridman (34:06.240)
Oh, I'm sorry.
Douglas Lenat (34:08.240)
I'm so sorry. Okay. But assertions works. Okay. So if we go back to that room with
Douglas Lenat (34:13.280)
Marvin Minsky with you, all these seminal figures, Ed Fagamon, thinking about this very
Douglas Lenat (34:22.160)
philosophical, but also engineering question. We can also go back a couple of decades before then
Lex Fridman (34:29.600)
and thinking about artificial intelligence broadly when people were thinking about,
Douglas Lenat (34:34.160)
you know, how do you create super intelligent systems, general intelligence. And I think
Douglas Lenat (34:40.560)
people's intuition was off at the time. And I mean, this continues to be the case that we're not,
Douglas Lenat (34:48.240)
when we're grappling with these exceptionally difficult ideas, we're not always, it's very
Douglas Lenat (34:53.040)
difficult to truly understand ourselves when we're thinking about the human mind to introspect how
Douglas Lenat (35:00.640)
difficult it is to engineer intelligence, to solve intelligence. We're not very good at estimating
Douglas Lenat (35:05.840)
that. And you are somebody who has really stayed with this question for decades.
Douglas Lenat (35:11.600)
What's your sense from the 1984 to today? Have you gotten a stronger sense of just how much
Douglas Lenat (35:22.240)
knowledge is required? You've kind of said with some level of certainty that it's still on the
Douglas Lenat (35:27.680)
order of magnitude of tens of millions. Right. For the first several years, I would have said that
Douglas Lenat (35:32.560)
it was on the order of one or two million. And so it took us about five or six years to realize
Douglas Lenat (35:40.720)
that we were off by a factor of 10. But I guess what I'm asking, you know, Marvin Misk is very
Douglas Lenat (35:47.440)
confident in the 60s. Yes. Right. What's your sense if you, you know, 200 years from now,
Douglas Lenat (35:59.440)
you're still, you know, you're not going to be any longer in this particular biological body,
Lex Fridman (36:05.280)
but your brain will still be in the digital form and you'll be looking back. Would you think you
Lex Fridman (36:11.520)
were smart today? Like your intuition was right? Or do you think you may be really off?
Lex Fridman (36:19.120)
So I think I'm right enough. And let me explain what I mean by that, which is sometimes like if
Douglas Lenat (36:27.680)
you have an old fashioned pump, you have to prime the pump and then eventually it starts. So I think
Douglas Lenat (36:34.000)
I'm right enough in the sense that what we've built, even if it isn't, so to speak, everything
Douglas Lenat (36:41.200)
you need, it's primed the knowledge pump enough that psych can now itself help to learn more and
Douglas Lenat (36:51.120)
more automatically on its own by reading things and understanding and occasionally asking questions
Douglas Lenat (36:56.560)
like a student would or something and by doing experiments and discovering things on its own
Lex Fridman (37:02.240)
and so on. So through a combination of psych powered discovery and psych powered reading,
Douglas Lenat (37:09.760)
it will be able to bootstrap itself. Maybe it's the final 2%, maybe it's the final 99%.
Lex Fridman (37:16.240)
So even if I'm wrong, all I really need to build is a system which has primed the pump enough
Douglas Lenat (37:24.000)
that it can begin that cascade upward, that self reinforcing sort of quadratically,
Douglas Lenat (37:31.200)
or maybe even exponentially increasing path upward that we get from, for instance, talking with each
Douglas Lenat (37:39.200)
other. That's why humans today know so much more than humans 100,000 years ago. We're not really
Douglas Lenat (37:45.760)
that much smarter than people were 100,000 years ago, but there's so much more knowledge and we
Douglas Lenat (37:50.720)
have language and we can communicate, we can check things on Google and so on. So effectively,
Douglas Lenat (37:56.560)
we have this enormous power at our fingertips and there's almost no limit to how much you could
Douglas Lenat (38:02.560)
learn if you wanted to because you've already gotten to a certain level of understanding of
Douglas Lenat (38:07.200)
the world that enables you to read all these articles and understand them, that enables you
Douglas Lenat (38:12.240)
to go out and if necessary, do experiments although that's slower as a way of gathering data
Lex Fridman (38:18.160)
and so on. And I think this is really an important point, which is if we have artificial
Douglas Lenat (38:24.160)
intelligence, real general artificial intelligence, human level artificial intelligence,
Douglas Lenat (38:29.440)
then people will become smarter. It's not so much that it'll be us versus the AIs, it's more like
Douglas Lenat (38:37.600)
us and the AIs together. We'll be able to do things that require more creativity, that would
Douglas Lenat (38:43.920)
take too long right now, but we'll be able to do lots of things in parallel. We'll be able to
Douglas Lenat (38:48.640)
misunderstand each other less. There's all sorts of value that effectively for an individual would
Douglas Lenat (38:56.640)
mean that individual will for all intents and purposes be smarter and that means that humanity
Douglas Lenat (39:02.960)
as a species will be smarter. And when was the last time that any invention qualitatively
Douglas Lenat (39:10.720)
made a huge difference in human intelligence? You have to go back a long ways. It wasn't like the
Douglas Lenat (39:16.160)
internet or the computer or mathematics or something. It was all the way back to the
Lex Fridman (39:22.400)
development of language. We sort of look back on prelinguistic cavemen as well.
Douglas Lenat (39:29.840)
They weren't really intelligent, were they? They weren't really human, were they? And I think that
Lex Fridman (39:36.320)
as you said, 50, 100, 200 years from now, people will look back on people today
Douglas Lenat (39:42.240)
right before the advent of these sort of lifelong general AI uses and say,
Lex Fridman (39:51.760)
you know, those poor people, they weren't really human, were they?
Douglas Lenat (39:55.520)
Mm hmm. Exactly. So you said a lot of really interesting things. By the way, I would maybe
Douglas Lenat (40:00.800)
try to argue that the internet is on the order of the kind of big leap in improvement that the
Douglas Lenat (40:12.960)
invention of language was. Well, it's certainly a big leap in one direction. We're not sure whether
Douglas Lenat (40:17.200)
it's upward or downward. Well, I mean very specific parts of the internet, which is access to information
Douglas Lenat (40:22.720)
like a website like Wikipedia, like ability for human beings from across the world to access
Douglas Lenat (40:28.240)
information very quickly. So I could take either side of this argument. And since you just took
Douglas Lenat (40:33.120)
one side, I'll give you the other side, which is that almost nothing has done more harm than
Douglas Lenat (40:40.880)
something like the internet and access to that information in two ways. One is it's made people
Douglas Lenat (40:47.520)
more globally ignorant in the same way that calculators made us more or less innumerate.
Lex Fridman (40:56.800)
So when I was growing up, we had to use slide rules. We had to be able to estimate and so on.
Douglas Lenat (41:02.800)
Today, people don't really understand numbers. They don't really understand math. They don't
Douglas Lenat (41:08.320)
really estimate very well at all and so on. They don't really understand the difference
Douglas Lenat (41:13.360)
between trillions and billions and millions and so on very well because calculators do that all
Douglas Lenat (41:20.320)
for us. And thanks to things like the internet and search engines, that same kind of juvenileism
Douglas Lenat (41:30.560)
is reinforced in making people essentially be able to live their whole lives, not just without
Douglas Lenat (41:35.600)
being able to do arithmetic and estimate, but now without actually having to really know almost
Douglas Lenat (41:40.880)
anything because anytime they need to know something, they'll just go and look it up.
Douglas Lenat (41:44.880)
You're right. And I could tell you could play both sides of this and it is a double edged sword.
Douglas Lenat (41:48.640)
You can, of course, say the same thing about language. Probably people when they invented
Douglas Lenat (41:52.160)
language, they would criticize. It used to be if we're angry, we would just kill a person. And if
Douglas Lenat (41:58.560)
we're in love, we would just have sex with them. And now everybody's writing poetry and bullshit.
Douglas Lenat (42:04.160)
You should just be direct. You should have physical contact. Enough of this words and books.
Douglas Lenat (42:11.040)
You're not actually experiencing. If you read a book, you're not experiencing the thing. This
Douglas Lenat (42:15.040)
is nonsense. That's right. If you read a book about how to make butter, that's not the same
Douglas Lenat (42:19.120)
as if you had to learn it and do it yourself and so on. So let's just say that something is gained,
Lex Fridman (42:24.800)
but something is lost every time you have these sorts of dependencies on technology.
Lex Fridman (42:33.600)
And overall, I think that having smarter individuals and having smarter AI augmented
Douglas Lenat (42:41.040)
human species will be one of the few ways that we'll actually be able to overcome some of the
Douglas Lenat (42:47.840)
global problems we have involving poverty and starvation and global warming and overcrowding,
Douglas Lenat (42:54.880)
all the other problems that are besetting the planet. We really need to be smarter.
Lex Fridman (43:01.840)
And there are really only two routes to being smarter. One is through biochemistry and genetics.
Douglas Lenat (43:09.280)
Genetic engineering. The other route is through having general AIs that augment our intelligence.
Lex Fridman (43:17.680)
And hopefully one of those two ways of paths to salvation will come through before it's too late.
Douglas Lenat (43:27.680)
Yeah, so I agree with you. And obviously, as an engineer, I have a better sense and an optimism
Douglas Lenat (43:35.440)
about the technology side of things because you can control things there more. Biology is just
Douglas Lenat (43:39.600)
such a giant mess. We're living through a pandemic now. There's so many ways that nature can just be
Douglas Lenat (43:45.520)
just destructive and destructive in a way where it doesn't even notice you. It's not like a battle
Douglas Lenat (43:51.840)
of humans versus virus. It's just like, huh, okay. And then you can just wipe out an entire species.
Douglas Lenat (43:57.440)
The other problem with the internet is that it has enabled us to surround ourselves with an
Douglas Lenat (44:07.600)
echo chamber, with a bubble of like minded people, which means that you can have truly bizarre
Douglas Lenat (44:16.560)
theories, conspiracy theories, fake news, and so on, promulgate and surround yourself with people
Douglas Lenat (44:23.600)
who essentially reinforce what you want to believe or what you already believe about the world.
Lex Fridman (44:30.720)
And in the old days, that was much harder to do when you had, say, only three TV networks,
Douglas Lenat (44:37.520)
or even before when you had no TV networks and you had to actually look at the world and make your
Douglas Lenat (44:42.560)
own reasoned decisions. I like the push and pull of our dance that we're doing because then I'll
Douglas Lenat (44:47.280)
just say in the old world, having come from the Soviet Union, because you had one or a couple of
Douglas Lenat (44:52.240)
networks, then propaganda could be much more effective. And then the government can overpower
Douglas Lenat (44:56.480)
its people by telling you the truth and then starving millions and torturing millions and
Douglas Lenat (45:03.760)
putting millions into camps and starting wars with a propaganda machine, allowing you to believe
Douglas Lenat (45:09.360)
that you're actually doing good in the world. With the internet, because of all the quote unquote
Douglas Lenat (45:14.240)
conspiracy theories, some of them are actually challenging the power centers, the very kind of
Douglas Lenat (45:19.600)
power centers that a century ago would have led to the death of millions. So there's, again, this
Douglas Lenat (45:26.800)
double edged sword. And I very much agree with you on the AI side. It's often an intuition that
Douglas Lenat (45:32.720)
people have that somehow AI will be used to maybe overpower people by certain select groups. And to
Douglas Lenat (45:40.640)
me, it's not at all obvious that that's the likely scenario. To me, the likely scenario, especially
Douglas Lenat (45:46.000)
just having observed the trajectory of technology, is it'll be used to empower people. It'll be used
Douglas Lenat (45:51.200)
to extend the capabilities of individuals across the world, because there's a lot of money to be
Douglas Lenat (45:59.360)
made that way. Improving people's lives, you can make a lot of money. I agree. I think that the
Douglas Lenat (46:05.600)
main thing that AI prostheses, AI amplifiers will do for people is make it easier, maybe even
Douglas Lenat (46:15.520)
unavoidable, for them to do good critical thinking. So pointing out logical fallacies,
Douglas Lenat (46:22.880)
logical contradictions and so on, in things that they otherwise would just blithely believe,
Douglas Lenat (46:31.040)
pointing out essentially data which they should take into consideration if they really want to
Douglas Lenat (46:39.920)
learn the truth about something and so on. So I think doing not just educating in the sense of
Douglas Lenat (46:47.120)
pouring facts into people's heads, but educating in the sense of arming people with the ability to do
Douglas Lenat (46:53.360)
good critical thinking is enormously powerful. The education system that we have in the US and
Lex Fridman (47:01.360)
worldwide generally don't do a good job of that. But I believe that the AI...
Douglas Lenat (47:08.160)
The AIs will. The AIs will, the AIs can and will. In the same way that everyone can have their own
Douglas Lenat (47:15.920)
Alexa or Siri or Google Assistant or whatever, everyone will have this sort of cradle to grave
Douglas Lenat (47:24.160)
assistant which will get to know you, which you'll get to trust, it'll model you, you'll model it,
Lex Fridman (47:30.560)
and it'll call to your attention things which will in some sense make your life better, easier,
Douglas Lenat (47:37.200)
less mistake ridden and so on, less regret ridden if you listen to it.
Douglas Lenat (47:45.600)
Yeah, I'm in full agreement with you about this space of technologies and I think it's super
Douglas Lenat (47:51.920)
exciting. And from my perspective, integrating emotional intelligence, so even things like
Douglas Lenat (47:57.440)
friendship and companionship and love into those kinds of systems, as opposed to helping you just
Douglas Lenat (48:04.320)
grow intellectually as a human being, allow you to grow emotionally, which is ultimately what makes
Douglas Lenat (48:09.520)
life amazing, is to sort of, you know, the old pursuit of happiness. So it's not just the pursuit
Douglas Lenat (48:16.960)
of reason, it's the pursuit of happiness too. The full spectrum. Well, let me sort of, because you
Douglas Lenat (48:22.880)
mentioned so many fascinating things, let me jump back to the idea of automated reasoning. So the
Douglas Lenat (48:30.000)
the acquisition of new knowledge has been done in this very interesting way, but primarily by humans
Douglas Lenat (48:37.920)
doing this. Just you can think of monks in their cells in medieval Europe, you know, carefully
Douglas Lenat (48:45.600)
illuminating manuscripts and so on. It's a very difficult and amazing process actually because
Douglas Lenat (48:51.600)
it allows you to truly ask the question about the in the white space, what is assumed. I think this
Douglas Lenat (48:58.640)
exercise is like very few people do this, right? They just do it subconsciously. They perform this.
Douglas Lenat (49:07.040)
By definition, right? Because those pieces of elided, of omitted information, of those missing
Douglas Lenat (49:14.720)
steps, as it were, are pieces of common sense. If you actually included all of them, it would
Douglas Lenat (49:21.920)
almost be offensive or confusing to the reader. It's like, why are they telling me all these? Of
Douglas Lenat (49:26.560)
course I know all these things. And so it's one of these things which almost by its very nature
Douglas Lenat (49:35.760)
has almost never been explicitly written down anywhere because by the time you're old enough
Douglas Lenat (49:42.640)
to talk to other people and so on, if you survived to that age, presumably you already got pieces of
Douglas Lenat (49:49.840)
common sense. Like if something causes you pain whenever you do it, probably not a good idea to
Douglas Lenat (49:55.600)
keep doing it. So what ideas do you have, given how difficult this step is, what ideas are there
Douglas Lenat (50:04.160)
for how to do it automatically without using humans or at least not doing like a large
Lex Fridman (50:12.720)
percentage of the work for humans and then humans only do the very high level supervisory work?
Lex Fridman (50:18.080)
So we have, in fact, two directions we're pushing on very, very heavily currently at PsychCorp. And
Douglas Lenat (50:25.920)
one involves natural language understanding and the ability to read what people have explicitly
Douglas Lenat (50:30.880)
written down and to pull knowledge in that way. But the other is to build a series of knowledge
Douglas Lenat (50:40.160)
editing tools, knowledge entry tools, knowledge capture tools, knowledge testing tools and so on.
Douglas Lenat (50:49.040)
Think of them as like user interface suite of software tools if you want, something that will
Douglas Lenat (50:55.280)
help people to more or less automatically expand and extend the system in areas where, for instance,
Douglas Lenat (51:03.920)
they want to build some app, have it do some application or something like that. So I'll give
Douglas Lenat (51:08.560)
you an example of one, which is something called abduction. So you've probably heard of like
Douglas Lenat (51:14.800)
deduction and induction and so on. But abduction is unlike those, abduction is not sound, it's just
Douglas Lenat (51:25.120)
useful. So for instance, deductively, if someone is out in the rain and they're going to get all
Douglas Lenat (51:33.840)
wet and when they enter a room, they might be all wet and so on. So that's deduction. But if someone
Douglas Lenat (51:42.000)
were to walk into the room right now and they were dripping wet, we would immediately look
Douglas Lenat (51:47.840)
outside to say, oh, did it start to rain or something like that. Now, why did we say maybe
Douglas Lenat (51:54.160)
it started to rain? That's not a sound logical inference, but it's certainly a reasonable
Douglas Lenat (51:59.760)
abductive leap to say, well, one of the most common ways that a person would have gotten
Douglas Lenat (52:06.320)
dripping wet is if they had gotten caught out in the rain or something like that. So what does that
Douglas Lenat (52:14.160)
have to do with what we were talking about? So suppose you're building one of these applications
Lex Fridman (52:18.480)
and the system gets some answer wrong and you say, oh, yeah, the answer to this question is
Douglas Lenat (52:24.400)
this one, not the one you came up with. Then what the system can do is it can use everything it
Douglas Lenat (52:30.400)
already knows about common sense, general knowledge, the domain you've already been
Douglas Lenat (52:34.400)
telling it about, and context like we talked about and so on and say, well, here are seven
Douglas Lenat (52:41.760)
alternatives, each of which I believe is plausible, given everything I already know. And if any of
Douglas Lenat (52:48.400)
these seven things were true, I would have come up with the answer you just gave me instead of the
Douglas Lenat (52:53.520)
wrong answer I came up with. Is one of these seven things true? And then you, the expert, will look
Douglas Lenat (52:59.200)
at those seven things and say, oh, yeah, number five is actually true. And so without actually
Douglas Lenat (53:04.560)
having to tinker down at the level of logical assertions and so on, you'll be able to educate
Douglas Lenat (53:11.920)
the system in the same way that you would help educate another person who you were trying to
Douglas Lenat (53:16.720)
apprentice or something like that. So that significantly reduces the mental effort
Douglas Lenat (53:22.880)
or significantly increases the efficiency of the teacher, the human teacher. Exactly. And it makes
Douglas Lenat (53:28.400)
more or less anyone able to be a teacher in that way. So that's part of the answer. And then the
Douglas Lenat (53:36.400)
other is that the system on its own will be able to, through reading, through conversations with
Douglas Lenat (53:44.160)
other people and so on, learn the same way that you or I or other humans do. First of all, that's
Douglas Lenat (53:52.560)
a beautiful vision. I'll have to ask you about Semantic Web in a second here. But first,
Douglas Lenat (53:57.600)
are there, when we talk about specific techniques, do you find something inspiring or directly useful
Douglas Lenat (54:04.000)
from the whole space of machine learning, deep learning, these kinds of spaces of techniques that
Lex Fridman (54:08.960)
have been shown effective for certain kinds of problems in the recent, now, decade and a half?
Douglas Lenat (54:15.760)
I think of the machine learning work as more or less what our right brain has been able to do.
Douglas Lenat (54:23.840)
I think of the machine learning work as more or less what our right brain hemispheres do. So
Douglas Lenat (54:30.880)
being able to take a bunch of data and recognize patterns, being able to statistically infer
Douglas Lenat (54:39.200)
things and so on. And I certainly wouldn't want to not have a right brain hemisphere,
Lex Fridman (54:47.360)
but I'm also glad that I have a left brain hemisphere as well, something that can
Douglas Lenat (54:51.680)
metaphorically sit back and puff on its pipe and think about this thing over here. It's like,
Lex Fridman (54:57.520)
why might this have been true? And what are the implications of it? How should I feel about that
Lex Fridman (55:03.520)
and why and so on? So thinking more deeply and slowly, what Kahneman called thinking slowly
Douglas Lenat (55:11.120)
versus thinking quickly, whereas you want machine learning to think quickly, but you want the
Douglas Lenat (55:16.160)
ability to think deeply, even if it's a little slower. So I'll give you an example of a project
Douglas Lenat (55:22.480)
we did recently with NIH involving the Cleveland Clinic and a couple other institutions that we ran
Douglas Lenat (55:30.960)
a project for. And what it did was it took GWAS's genome wide association studies.
Douglas Lenat (55:37.200)
Those are big databases of patients that came into a hospital. They got their DNA sequenced
Douglas Lenat (55:46.720)
because the cost of doing that has gone from infinity to billions of dollars to $100 or so.
Lex Fridman (55:54.880)
And so now patients routinely get their DNA sequenced. So you have these big databases
Douglas Lenat (55:59.760)
of the SNPs, the single nucleotide polymorphisms, the point mutations in a patient's DNA,
Lex Fridman (56:06.320)
and the disease that happened to bring them into the hospital. So now you can do correlation
Douglas Lenat (56:11.840)
studies, machine learning studies of which mutations are associated with and led to which
Douglas Lenat (56:20.880)
physiological problems and diseases and so on, like getting arthritis and so on. And the problem
Douglas Lenat (56:27.920)
is that those correlations turn out to be very spurious. They turn out to be very noisy. Very
Douglas Lenat (56:34.080)
many of them have led doctors onto wild goose chases and so on. And so they wanted a way of
Douglas Lenat (56:40.960)
eliminating or the bad ones are focusing on the good ones. And so this is where psych comes in,
Douglas Lenat (56:46.960)
which is psych takes those sort of A to Z correlations between point mutations and
Douglas Lenat (56:53.760)
the medical condition that needs treatment. And we say, okay, let's use all this public knowledge
Lex Fridman (57:00.240)
and common sense knowledge about what reactions occur where in the human body,
Lex Fridman (57:06.240)
what polymerizes what, what catalyzes what reactions and so on. And let's try to put together
Lex Fridman (57:12.400)
a 10 or 20 or 30 step causal explanation of why that mutation might have caused
Douglas Lenat (57:20.400)
that medical condition. And so psych would put together in some sense, some Rube Goldberg like
Douglas Lenat (57:25.920)
a chain that would say, oh yeah, that mutation if it got expressed would be this altered protein,
Douglas Lenat (57:35.920)
which because of that, if it got to this part of the body would catalyze this reaction. And by the
Douglas Lenat (57:40.480)
way, that would cause more bioactive vitamin D in the person's blood. And anyway, 10 steps later,
Douglas Lenat (57:46.240)
that screws up bone resorption and that's why this person got osteoporosis early in life and so on.
Lex Fridman (57:52.640)
So that's human interpretable or at least docs are human interpretable.
Douglas Lenat (57:55.760)
Exactly. And the important thing even more than that is you shouldn't really trust that 20 step
Douglas Lenat (58:05.520)
Rube Goldberg chain any more than you trust that initial A to Z correlation except two things. One,
Douglas Lenat (58:12.160)
if you can't even think of one causal chain to explain this, then that correlation probably was
Douglas Lenat (58:19.760)
just noise to begin with. And secondly, and even more powerfully, along the way that causal chain
Douglas Lenat (58:27.040)
will make predictions like the one about having more bioactive vitamin D in your blood. So you
Douglas Lenat (58:32.240)
can now go back to the data about these patients and say, by the way, did they have slightly
Douglas Lenat (58:38.800)
elevated levels of bioactive vitamin D in their blood and so on? And if the answer is no, that
Douglas Lenat (58:44.400)
strongly disconfirms your whole causal chain. And if the answer is yes, that somewhat confirms
Douglas Lenat (58:50.800)
that causal chain. And so using that, we were able to take these correlations from this GWAS
Douglas Lenat (58:57.280)
database and we were able to essentially focus the researchers attention on the very small
Douglas Lenat (59:06.400)
percentage of correlations that had some explanation and even better some explanation
Douglas Lenat (59:12.720)
that also made some independent prediction that they could confirm or disconfirm by looking at
Douglas Lenat (59:17.280)
the data. So think of it like this kind of synergy where you want the right brain machine learning
Douglas Lenat (59:23.120)
to quickly come up with possible answers. You want the left brain psych like AI to think about that
Lex Fridman (59:31.360)
and think about why that might have been the case and what else would be the case if that were true
Lex Fridman (59:36.320)
and so on, and then suggest things back to the right brain to quickly check out again. So it's
Douglas Lenat (59:43.520)
that kind of synergy back and forth, which I think is really what's going to lead to general AI, not
Lex Fridman (59:50.480)
narrow, brittle machine learning systems and not just something like psych.
Douglas Lenat (59:55.520)
Okay. So that's a brilliant synergy. But I was also thinking in terms of the automated expansion
🔗 相关节目