Judea Pearl: Causal Reasoning, Counterfactuals, Bayesian Networks, and the Path to AGI
AI 与机器学习数学技术与编程心理与人性哲学与宗教
📋 章节目录
暂无章节信息
🔑 关键词
donmodelknowledgemachinequestionsputcausationcannotmetaphorlearningimportantmeansscienceexperimentprobabilityrobotdoingcorrelationdatadidn
💬 精彩语录
暂无语录
🎙️ 完整对话（1581 条）
Lex Fridman (00:00.000)
The following is a conversation with Judea Pearl,
以下是与 Judea Pearl 的对话，
Lex Fridman (00:03.240)
professor at UCLA and a winner of the Turing Award
加州大学洛杉矶分校教授、图灵奖获得者
Lex Fridman (00:06.720)
that's generally recognized as the Nobel Prize of Computing.
这被普遍认为是计算机领域的诺贝尔奖。
Lex Fridman (00:10.640)
He's one of the seminal figures
他是影响深远的人物之一
Lex Fridman (00:12.400)
in the field of artificial intelligence,
在人工智能领域，
Judea Pearl (00:14.440)
computer science, and statistics.
计算机科学和统计学。
Lex Fridman (00:16.640)
He has developed and championed probabilistic approaches
他开发并倡导概率方法
Judea Pearl (00:19.960)
to AI, including Beijing networks,
到AI，包括北京网络，
Lex Fridman (00:22.680)
and profound ideas in causality in general.
以及一般因果关系的深刻思想。
Judea Pearl (00:26.040)
These ideas are important not just to AI,
这些想法不仅对人工智能很重要，
Lex Fridman (00:29.080)
but to our understanding and practice of science.
而是我们对科学的理解和实践。
Lex Fridman (00:32.800)
But in the field of AI, the idea of causality,
但在人工智能领域，因果关系的理念，
Lex Fridman (00:35.840)
cause and effect, to many, lie at the core
对许多人来说，因果关系是核心
Judea Pearl (00:39.400)
of what is currently missing and what must be developed
目前缺少什么以及必须开发什么
Lex Fridman (00:42.080)
in order to build truly intelligent systems.
以构建真正的智能系统。
Judea Pearl (00:46.080)
For this reason, and many others,
由于这个原因以及许多其他原因，
Lex Fridman (00:48.000)
his work is worth returning to often.
他的作品值得经常回顾。
Judea Pearl (00:50.760)
I recommend his most recent book called Book of Why
我推荐他的最新书《为什么之书》
Lex Fridman (00:54.200)
that presents key ideas from a lifetime of work
提出了一生工作中的关键想法
Judea Pearl (00:57.160)
in a way that is accessible to the general public.
以公众可以访问的方式。
Lex Fridman (01:00.440)
This is the Artificial Intelligence Podcast.
Judea Pearl (01:03.480)
If you enjoy it, subscribe on YouTube,
Lex Fridman (01:05.880)
give it five stars on Apple Podcast,
Judea Pearl (01:07.840)
support it on Patreon, or simply connect with me on Twitter
Lex Fridman (01:11.360)
at Lex Friedman, spelled F R I D M A N.
Judea Pearl (01:15.320)
If you leave a review on Apple Podcasts especially,
Lex Fridman (01:18.200)
but also a cast box or comment on YouTube,
Judea Pearl (01:20.960)
consider mentioning topics, people, ideas, questions,
Lex Fridman (01:23.920)
quotes, and science, tech, and philosophy
Judea Pearl (01:26.120)
you find interesting, and I'll read them on this podcast.
Lex Fridman (01:29.560)
I won't call out names, but I love comments
Judea Pearl (01:31.960)
with kindness and thoughtfulness in them,
Lex Fridman (01:33.880)
so I thought I'd share them with you.
Judea Pearl (01:35.760)
Someone on YouTube highlighted a quote
Lex Fridman (01:37.840)
from the conversation with Noam Chomsky,
Judea Pearl (01:40.040)
where he said that the significance of your life
Lex Fridman (01:42.800)
is something you create.
Judea Pearl (01:44.780)
I like this line as well.
Lex Fridman (01:46.600)
On most days, the existentialist approach to life
Judea Pearl (01:49.880)
is one I find liberating and fulfilling.
Lex Fridman (01:53.660)
I recently started doing ads
Judea Pearl (01:55.040)
at the end of the introduction.
Lex Fridman (01:56.640)
I'll do one or two minutes after introducing the episode,
Lex Fridman (01:59.400)
and never any ads in the middle
Lex Fridman (02:01.080)
that break the flow of the conversation.
Judea Pearl (02:03.280)
I hope that works for you
Lex Fridman (02:04.760)
and doesn't hurt the listening experience.
Judea Pearl (02:08.280)
This show is presented by Cash App,
Lex Fridman (02:10.600)
the number one finance app in the App Store.
Judea Pearl (02:13.040)
I personally use Cash App to send money to friends,
Lex Fridman (02:15.520)
but you can also use it to buy, sell,
Lex Fridman (02:17.420)
and deposit Bitcoin in just seconds.
Lex Fridman (02:20.100)
Cash App also has a new investing feature.
Judea Pearl (02:22.800)
You can buy fractions of a stock, say $1 worth,
Lex Fridman (02:25.860)
no matter what the stock price is.
Judea Pearl (02:28.040)
Broker services are provided by Cash App Investing,
Lex Fridman (02:30.720)
a subsidiary of Square, a member of SIPC.
Judea Pearl (02:34.280)
I'm excited to be working with Cash App
Lex Fridman (02:36.600)
to support one of my favorite organizations called First,
Judea Pearl (02:39.940)
best known for their first robotics and Lego competitions.
Lex Fridman (02:43.440)
They educate and inspire hundreds of thousands of students
Judea Pearl (02:47.120)
in over 110 countries,
Lex Fridman (02:49.000)
and have a perfect rating on Charity Navigator,
Judea Pearl (02:51.760)
which means the donated money
Lex Fridman (02:53.120)
is used to the maximum effectiveness.
Judea Pearl (02:56.000)
When you get Cash App from the App Store or Google Play,
Lex Fridman (02:58.680)
and use the code LEXPODCAST, you'll get $10,
Lex Fridman (03:02.640)
and Cash App will also donate $10 to First,
Lex Fridman (03:05.600)
which again, is an organization
Judea Pearl (03:07.560)
that I've personally seen inspire girls and boys
Lex Fridman (03:10.120)
to dream of engineering a better world.
Lex Fridman (03:12.800)
And now, here's my conversation with Judea Pearl.
Lex Fridman (03:18.080)
You mentioned in an interview
Judea Pearl (03:19.720)
that science is not a collection of facts,
Lex Fridman (03:21.840)
but a constant human struggle
Judea Pearl (03:23.960)
with the mysteries of nature.
Lex Fridman (03:26.700)
What was the first mystery that you can recall
Lex Fridman (03:29.240)
that hooked you, that kept you in the creaset?
Lex Fridman (03:30.800)
Oh, the first mystery, that's a good one.
Judea Pearl (03:34.440)
Yeah, I remember that.
Lex Fridman (03:37.840)
I had a fever for three days.
Lex Fridman (03:41.440)
And when I learned about Descartes, analytic geometry,
Lex Fridman (03:45.680)
and I found out that you can do all the construction
Judea Pearl (03:49.880)
in geometry using algebra.
Lex Fridman (03:52.880)
And I couldn't get over it.
Judea Pearl (03:54.480)
I simply couldn't get out of bed.
Lex Fridman (03:58.280)
So what kind of world does analytic geometry unlock?
Judea Pearl (04:02.800)
Well, it connects algebra with geometry.
Lex Fridman (04:07.320)
Okay, so Descartes had the idea
Judea Pearl (04:09.360)
that geometrical construction and geometrical theorems
Lex Fridman (04:14.360)
and assumptions can be articulated
Judea Pearl (04:17.520)
in the language of algebra,
Lex Fridman (04:19.520)
which means that all the proof that we did in high school,
Lex Fridman (04:24.840)
and trying to prove that the three bisectors
Lex Fridman (04:28.560)
meet at one point, and that, okay,
Judea Pearl (04:33.640)
all this can be proven by just shuffling around notation.
Lex Fridman (04:39.440)
Yeah, that was a traumatic experience.
Judea Pearl (04:43.560)
That was a traumatic experience.
Lex Fridman (04:44.400)
For me, it was, I'm telling you.
Lex Fridman (04:45.520)
So it's the connection
Lex Fridman (04:46.960)
between the different mathematical disciplines,
Judea Pearl (04:49.120)
that they all.
Lex Fridman (04:49.960)
Not in between two different languages.
Judea Pearl (04:52.960)
Languages.
Lex Fridman (04:53.800)
Yeah.
Lex Fridman (04:54.640)
So which mathematic discipline is most beautiful?
Lex Fridman (04:57.200)
Is geometry it for you?
Judea Pearl (04:58.560)
Both are beautiful.
Lex Fridman (04:59.440)
They have almost the same power.
Lex Fridman (05:02.400)
But there's a visual element to geometry, being a.
Lex Fridman (05:04.920)
Visually, it's more transparent.
Lex Fridman (05:08.080)
But once you get over to algebra,
Lex Fridman (05:10.680)
then the linear equation is a straight line.
Lex Fridman (05:14.400)
This translation is easily absorbed, okay?
Lex Fridman (05:18.160)
And to pass a tangent to a circle,
Judea Pearl (05:22.760)
you know, you have the basic theorems,
Lex Fridman (05:25.480)
and you can do it with algebra.
Lex Fridman (05:27.480)
So but the transition from one to another was really,
Lex Fridman (05:31.520)
I thought that Descartes was the greatest mathematician
Judea Pearl (05:34.120)
of all times.
Lex Fridman (05:35.160)
So you have been at the, if you think of engineering
Lex Fridman (05:40.840)
and mathematics as a spectrum.
Lex Fridman (05:43.240)
Yes.
Judea Pearl (05:44.080)
You have been, you have walked casually along this spectrum
Lex Fridman (05:49.200)
throughout your life.
Judea Pearl (05:51.520)
You know, a little bit of engineering,
Lex Fridman (05:53.000)
and then, you know,
Judea Pearl (05:55.920)
you've done a little bit of mathematics here and there.
Lex Fridman (05:58.800)
Not a little bit.
Judea Pearl (05:59.640)
I mean, we got a very solid background in mathematics,
Lex Fridman (06:04.080)
because our teachers were geniuses.
Judea Pearl (06:07.080)
Our teachers came from Germany in the 1930s,
Lex Fridman (06:09.720)
running away from Hitler.
Judea Pearl (06:11.880)
They left their careers in Heidelberg and Berlin,
Lex Fridman (06:15.080)
and came to teach high school in Israel.
Lex Fridman (06:17.840)
And we were the beneficiary of that experiment.
Lex Fridman (06:21.600)
So I, and they taught us math the good way.
Lex Fridman (06:25.200)
What's the good way to teach math?
Lex Fridman (06:26.720)
Chronologically.
Judea Pearl (06:29.040)
The people.
Lex Fridman (06:29.880)
The people behind the theorems, yeah.
Judea Pearl (06:33.320)
Their cousins, and their nieces, and their faces.
Lex Fridman (06:39.040)
And how they jumped from the bathtub when they scream,
Judea Pearl (06:41.520)
Eureka!
Lex Fridman (06:43.480)
And ran naked in town.
Lex Fridman (06:46.000)
So you're almost educated as a historian of math.
Lex Fridman (06:49.240)
No, we just got a glimpse of that history
Judea Pearl (06:51.920)
together with a theorem,
Lex Fridman (06:53.760)
so every exercise in math was connected with a person.
Lex Fridman (06:59.600)
And the time of the person.
Lex Fridman (07:01.040)
The period.
Judea Pearl (07:03.520)
The period, also mathematically speaking.
Lex Fridman (07:05.600)
Mathematically speaking, yes.
Judea Pearl (07:06.880)
Not the politics, no.
Lex Fridman (07:09.040)
So, and then in university,
Judea Pearl (07:14.000)
you have gone on to do engineering.
Lex Fridman (07:16.240)
Yeah.
Lex Fridman (07:17.200)
I get a B.S. in engineering and a technical, right?
Lex Fridman (07:20.680)
And then I moved here for graduate work,
Lex Fridman (07:25.600)
and I got, I did engineering in addition to physics
Lex Fridman (07:30.280)
in Rutgers, and it combined very nicely with my thesis,
Judea Pearl (07:35.840)
which I did in RCA Laboratories in superconductivity.
Lex Fridman (07:40.480)
And then somehow thought to switch
Judea Pearl (07:43.760)
to almost computer science, software,
Lex Fridman (07:46.720)
even, not switch, but long to become,
Judea Pearl (07:50.800)
to get into software engineering a little bit.
Lex Fridman (07:53.000)
Yes.
Lex Fridman (07:53.840)
And programming, if you can call it that in the 70s.
Lex Fridman (07:56.200)
So there's all these disciplines.
Judea Pearl (07:58.120)
Yeah.
Lex Fridman (07:58.960)
So to pick a favorite, in terms of engineering
Lex Fridman (08:02.880)
and mathematics, which path do you think has more beauty?
Lex Fridman (08:07.120)
Which path has more power?
Judea Pearl (08:08.600)
It's hard to choose, no.
Lex Fridman (08:10.560)
I enjoy doing physics, and even have a vortex
Judea Pearl (08:14.640)
named on my name.
Lex Fridman (08:16.000)
So I have investment in immortality.
Lex Fridman (08:23.400)
So what is a vortex?
Lex Fridman (08:25.160)
Vortex is in superconductivity.
Judea Pearl (08:27.000)
In the superconductivity, yeah.
Lex Fridman (08:27.840)
You have permanent current swirling around.
Judea Pearl (08:30.880)
One way or the other, you can have a store one or zero
Lex Fridman (08:34.560)
for a computer.
Judea Pearl (08:35.400)
That's what we worked on in the 1960 in RCA.
Lex Fridman (08:39.680)
And I discovered a few nice phenomena with the vortices.
Judea Pearl (08:44.080)
You push current and they move.
Lex Fridman (08:44.920)
So that's a pearl vortex.
Lex Fridman (08:46.600)
Pearl vortex, right, you can Google it, right?
Lex Fridman (08:50.240)
I didn't know about it, but the physicists,
Judea Pearl (08:52.360)
they picked up on my thesis, on my PhD thesis,
Lex Fridman (08:57.200)
and it becomes popular when thin film superconductors
Judea Pearl (09:03.240)
became important for high temperature superconductors.
Lex Fridman (09:06.920)
So they called it pearl vortex without my knowledge.
Judea Pearl (09:10.840)
I discovered it only about 15 years ago.
Lex Fridman (09:14.760)
You have footprints in all of the sciences.
Lex Fridman (09:17.560)
So let's talk about the universe a little bit.
Lex Fridman (09:20.960)
Is the universe at the lowest level deterministic
Lex Fridman (09:23.880)
or stochastic in your amateur philosophy view?
Lex Fridman (09:27.400)
Put another way, does God play dice?
Lex Fridman (09:30.120)
We know it is stochastic, right?
Lex Fridman (09:33.000)
Today, today we think it is stochastic.
Judea Pearl (09:35.160)
Yes.
Lex Fridman (09:37.000)
We think because we have the Heisenberg uncertainty principle
Lex Fridman (09:40.120)
and we have some experiments to confirm that.
Lex Fridman (09:45.760)
All we have is experiments to confirm it.
Judea Pearl (09:47.960)
We don't understand why.
Lex Fridman (09:50.400)
Why is already?
Judea Pearl (09:51.440)
You wrote a book about why.
Lex Fridman (09:52.960)
Yeah, it's a puzzle.
Judea Pearl (09:57.240)
It's a puzzle that you have the dice flipping machine,
Lex Fridman (10:02.400)
oh God, and the result of the flipping
Judea Pearl (10:08.560)
propagate with the speed faster than the speed of light.
Lex Fridman (10:12.280)
We can't explain it, okay?
Judea Pearl (10:14.240)
So, but it only governs microscopic phenomena.
Lex Fridman (10:19.240)
Microscopic phenomena.
Lex Fridman (10:21.320)
So you don't think of quantum mechanics
Lex Fridman (10:23.560)
as useful for understanding the nature of reality?
Judea Pearl (10:28.240)
No, diversionary.
Lex Fridman (10:30.400)
So in your thinking, the world might
Judea Pearl (10:34.520)
as well be deterministic.
Lex Fridman (10:36.000)
The world is deterministic,
Lex Fridman (10:38.480)
and as far as the neuron firing is concerned,
Lex Fridman (10:42.800)
it is deterministic to first approximation.
Lex Fridman (10:47.240)
What about free will?
Lex Fridman (10:48.920)
Free will is also a nice exercise.
Judea Pearl (10:52.920)
Free will is an illusion that we AI people are gonna solve.
Lex Fridman (10:59.320)
So what do you think once we solve it,
Lex Fridman (11:01.800)
that solution will look like?
Lex Fridman (11:03.360)
Once we put it in the page.
Judea Pearl (11:04.200)
The solution will look like,
Lex Fridman (11:06.240)
first of all, it will look like a machine.
Judea Pearl (11:08.920)
A machine that act as though it has free will.
Lex Fridman (11:12.560)
It communicates with other machines
Judea Pearl (11:14.760)
as though they have free will,
Lex Fridman (11:17.160)
and you wouldn't be able to tell the difference
Judea Pearl (11:19.480)
between a machine that does
Lex Fridman (11:21.560)
and a machine that doesn't have free will, okay?
Lex Fridman (11:24.680)
So the illusion, it propagates the illusion
Lex Fridman (11:26.840)
of free will amongst the other machines.
Lex Fridman (11:28.960)
And faking it is having it, okay?
Lex Fridman (11:33.120)
That's what Turing test is all about.
Judea Pearl (11:35.120)
Faking intelligence is intelligent
Lex Fridman (11:37.160)
because it's not easy to fake.
Judea Pearl (11:41.080)
It's very hard to fake,
Lex Fridman (11:43.200)
and you can only fake if you have it.
Lex Fridman (11:45.200)
So that's such a beautiful statement.
Lex Fridman (11:54.080)
Yeah, you can't fake it if you don't have it, yeah.
Lex Fridman (11:59.520)
So let's begin at the beginning with probability,
Lex Fridman (12:06.800)
both philosophically and mathematically.
Lex Fridman (12:09.360)
What does it mean to say the probability
Lex Fridman (12:11.560)
of something happening is 50%?
Lex Fridman (12:15.160)
What is probability?
Lex Fridman (12:16.960)
It's a degree of uncertainty
Judea Pearl (12:18.680)
that an agent has about the world.
Lex Fridman (12:22.400)
You're still expressing some knowledge in that statement.
Judea Pearl (12:24.760)
Of course.
Lex Fridman (12:26.000)
If the probability is 90%,
Judea Pearl (12:27.840)
it's absolutely a different kind of knowledge
Lex Fridman (12:29.800)
than if it is 10%.
Lex Fridman (12:32.480)
But it's still not solid knowledge, it's...
Lex Fridman (12:36.080)
It is solid knowledge, but hey,
Judea Pearl (12:38.520)
if you tell me that 90% assurance smoking
Lex Fridman (12:43.520)
will give you lung cancer in five years versus 10%,
Judea Pearl (12:48.880)
it's a piece of useful knowledge.
Lex Fridman (12:52.400)
So the statistical view of the universe,
Lex Fridman (12:56.120)
why is it useful?
Lex Fridman (12:57.600)
So we're swimming in complete uncertainty,
Judea Pearl (13:00.760)
most of everything around us.
Lex Fridman (13:01.600)
It allows you to predict things with a certain probability,
Lex Fridman (13:06.120)
and computing those probabilities are very useful.
Lex Fridman (13:09.240)
That's the whole idea of prediction.
Lex Fridman (13:15.080)
And you need prediction to be able to survive.
Lex Fridman (13:18.160)
If you can't predict the future,
Judea Pearl (13:19.640)
then you're just crossing the street,
Lex Fridman (13:22.320)
will be extremely fearful.
Lex Fridman (13:26.080)
And so you've done a lot of work in causation,
Lex Fridman (13:28.840)
and so let's think about correlation.
Judea Pearl (13:32.120)
I started with probability.
Lex Fridman (13:34.320)
You started with probability.
Judea Pearl (13:35.640)
You've invented the Bayesian networks.
Lex Fridman (13:38.760)
Yeah.
Lex Fridman (13:39.600)
And so we'll dance back and forth
Lex Fridman (13:43.880)
between these levels of uncertainty.
Lex Fridman (13:47.480)
But what is correlation?
Lex Fridman (13:49.320)
What is it?
Lex Fridman (13:50.360)
So probability of something happening is something,
Lex Fridman (13:54.000)
but then there's a bunch of things happening.
Lex Fridman (13:56.480)
And sometimes they happen together,
Lex Fridman (13:58.680)
sometimes not, they're independent or not.
Lex Fridman (14:00.760)
So how do you think about correlation of things?
Lex Fridman (14:03.640)
Correlation occurs when two things vary together
Judea Pearl (14:06.280)
over a very long time is one way of measuring it.
Lex Fridman (14:09.720)
Or when you have a bunch of variables
Judea Pearl (14:11.840)
that they all vary cohesively,
Lex Fridman (14:15.880)
then because we have a correlation here.
Lex Fridman (14:18.600)
And usually when we think about correlation,
Lex Fridman (14:21.720)
we really think causally.
Judea Pearl (14:24.400)
Things cannot be correlated unless there is a reason
Lex Fridman (14:27.960)
for them to vary together.
Lex Fridman (14:30.280)
Why should they vary together?
Lex Fridman (14:32.040)
If they don't see each other, why should they vary together?
Lex Fridman (14:35.560)
So underlying it somewhere is causation.
Lex Fridman (14:38.240)
Yes.
Judea Pearl (14:39.240)
Hidden in our intuition, there is a notion of causation
Lex Fridman (14:43.160)
because we cannot grasp any other logic except causation.
Lex Fridman (14:48.600)
And how does conditional probability differ from causation?
Lex Fridman (14:55.600)
So what is conditional probability?
Judea Pearl (14:57.920)
Conditional probability, how things vary
Lex Fridman (15:00.560)
when one of them stays the same.
Judea Pearl (15:05.040)
Now staying the same means that I have chosen
Lex Fridman (15:09.320)
to look only at those incidents
Judea Pearl (15:11.720)
where the guy has the same value as previous one.
Lex Fridman (15:16.120)
It's my choice as an experimenter.
Lex Fridman (15:19.280)
So things that are not correlated before
Lex Fridman (15:22.280)
could become correlated.
Judea Pearl (15:24.240)
Like for instance, if I have two coins
Lex Fridman (15:26.840)
which are uncorrelated, okay,
Lex Fridman (15:29.240)
and I choose only those flippings, experiments
Lex Fridman (15:33.760)
in which a bell rings, and the bell rings
Judea Pearl (15:36.680)
when at least one of them is a tail, okay,
Lex Fridman (15:40.760)
then suddenly I see correlation between the two coins
Judea Pearl (15:44.360)
because I only look at the cases where the bell rang.
Lex Fridman (15:49.280)
You see, it's my design, with my ignorance essentially,
Judea Pearl (15:53.680)
with my audacity to ignore certain incidents,
Lex Fridman (15:58.680)
I suddenly create a correlation
Judea Pearl (16:04.760)
where it doesn't exist physically.
Lex Fridman (16:07.760)
Right, so that's, you just outlined one of the flaws
Judea Pearl (16:11.400)
of observing the world and trying to infer something
Lex Fridman (16:14.960)
from the math about the world
Judea Pearl (16:16.080)
from looking at the correlation.
Lex Fridman (16:17.520)
I don't look at it as a flaw, the world works like that.
Lex Fridman (16:20.960)
But the flaws comes if we try to impose
Lex Fridman (16:24.960)
causal logic on correlation, it doesn't work too well.
Judea Pearl (16:34.680)
I mean, but that's exactly what we do.
Lex Fridman (16:36.320)
That's what, that has been the majority of science.
Judea Pearl (16:40.040)
The majority of naive science, the decisions know it.
Lex Fridman (16:45.040)
The decisions know that if you condition
Judea Pearl (16:47.840)
on a third variable, then you can destroy
Lex Fridman (16:50.960)
or create correlations among two other variables.
Judea Pearl (16:55.640)
They know it, it's in their data.
Lex Fridman (16:58.480)
It's nothing surprising, that's why they all dismiss
Judea Pearl (17:01.120)
the Simpson Paradox, ah, we know it.
Lex Fridman (17:04.120)
They don't know anything about it.
Judea Pearl (17:07.280)
Well, there's disciplines like psychology
Lex Fridman (17:09.680)
where all the variables are hard to account for.
Lex Fridman (17:12.880)
And so oftentimes there's a leap
Lex Fridman (17:15.200)
between correlation to causation.
Judea Pearl (17:17.520)
You're imposing.
Lex Fridman (17:18.840)
You're implying a leap.
Lex Fridman (17:21.120)
Who is trying to get causation from correlation?
Lex Fridman (17:26.520)
You're not proving causation,
Lex Fridman (17:27.960)
but you're sort of discussing it,
Lex Fridman (17:31.680)
implying, sort of hypothesizing with our ability to prove.
Lex Fridman (17:35.360)
Which discipline you have in mind?
Lex Fridman (17:37.040)
I'll tell you if they are obsolete,
Judea Pearl (17:40.440)
or if they are outdated, or they are about to get outdated.
Lex Fridman (17:44.240)
Yes, yes.
Lex Fridman (17:45.440)
Tell me which one you have in mind.
Lex Fridman (17:46.760)
Oh, psychology, you know.
Lex Fridman (17:48.160)
Psychology, what, is it SEM, structural equation?
Lex Fridman (17:50.760)
No, no, I was thinking of applied psychology studying.
Judea Pearl (17:54.280)
For example, we work with human behavior
Lex Fridman (17:57.200)
in semi autonomous vehicles, how people behave.
Lex Fridman (18:00.360)
And you have to conduct these studies
Lex Fridman (18:02.560)
of people driving cars.
Judea Pearl (18:03.960)
Everything starts with the question.
Lex Fridman (18:05.520)
What is the research question?
Lex Fridman (18:07.800)
What is the research question?
Lex Fridman (18:10.400)
The research question, do people fall asleep
Lex Fridman (18:14.320)
when the car is driving itself?
Lex Fridman (18:18.560)
Do they fall asleep, or do they tend to fall asleep
Lex Fridman (18:22.240)
more frequently?
Lex Fridman (18:23.160)
More frequently.
Judea Pearl (18:24.000)
Than the car not driving itself.
Lex Fridman (18:25.960)
Not driving itself.
Judea Pearl (18:26.880)
That's a good question, okay.
Lex Fridman (18:28.720)
And so you measure, you put people in the car
Judea Pearl (18:32.480)
because it's real world.
Lex Fridman (18:33.720)
You can't conduct an experiment
Judea Pearl (18:35.160)
where you control everything.
Lex Fridman (18:36.320)
Why can't you control?
Judea Pearl (18:37.840)
You could.
Lex Fridman (18:38.680)
Why can't you control the automatic module on and off?
Judea Pearl (18:44.760)
Because it's on road, public.
Lex Fridman (18:47.720)
I mean, there's aspects to it that's unethical.
Judea Pearl (18:52.680)
Because it's testing on public roads.
Lex Fridman (18:54.920)
So you can only use vehicle.
Judea Pearl (18:56.640)
They have to, the people, the drivers themselves
Lex Fridman (19:00.240)
have to make that choice themselves.
Lex Fridman (19:02.840)
And so they regulate that.
Lex Fridman (19:05.080)
And so you just observe when they drive autonomously
Lex Fridman (19:09.040)
and when they don't.
Lex Fridman (19:10.320)
And then.
Lex Fridman (19:11.160)
But maybe they turn it off when they were very tired.
Lex Fridman (19:13.120)
Yeah, that kind of thing.
Lex Fridman (19:14.520)
But you don't know those variables.
Lex Fridman (19:16.600)
Okay, so that you have now uncontrolled experiment.
Judea Pearl (19:19.360)
Uncontrolled experiment.
Lex Fridman (19:20.680)
We call it observational study.
Lex Fridman (19:23.200)
And we form the correlation detected.
Lex Fridman (19:27.160)
We have to infer causal relationship.
Judea Pearl (19:30.520)
Whether it was the automatic piece
Lex Fridman (19:33.440)
has caused them to fall asleep, or.
Lex Fridman (19:35.960)
So that is an issue that is about 120 years old.
Lex Fridman (19:45.440)
I should only go 100 years old, okay.
Judea Pearl (19:51.400)
Well, maybe it's not.
Lex Fridman (19:52.840)
Actually I should say it's 2,000 years old.
Judea Pearl (19:55.240)
Because we have this experiment by Daniel.
Lex Fridman (19:58.520)
But the Babylonian king that wanted the exiled,
Lex Fridman (1:00:01.620)
Is a Turing test a reasonable test?
Lex Fridman (1:00:04.980)
A test of free will doesn't exist yet.
Lex Fridman (1:00:08.380)
How would you test free will?
Lex Fridman (1:00:10.940)
So far, we know only one thing.
Judea Pearl (1:00:14.940)
If robots can communicate with reward and punishment
Lex Fridman (1:00:20.500)
among themselves and hitting each other on the wrist
Lex Fridman (1:00:25.420)
and say, you shouldn't have done that, okay?
Lex Fridman (1:00:27.820)
Playing better soccer because they can do that.
Lex Fridman (1:00:33.900)
What do you mean, because they can do that?
Lex Fridman (1:00:35.940)
Because they can communicate among themselves.
Judea Pearl (1:00:38.100)
Because of the communication they can do.
Lex Fridman (1:00:40.100)
Because they communicate like us.
Judea Pearl (1:00:42.660)
Reward and punishment, yes.
Lex Fridman (1:00:44.580)
You didn't pass the ball at the right time,
Lex Fridman (1:00:47.580)
and so therefore you're gonna sit on the bench
Lex Fridman (1:00:50.060)
for the next two.
Judea Pearl (1:00:51.540)
If they start communicating like that,
Lex Fridman (1:00:53.660)
the question is, will they play better soccer?
Lex Fridman (1:00:56.380)
As opposed to what?
Lex Fridman (1:00:57.660)
As opposed to what they do now?
Judea Pearl (1:00:59.660)
Without this ability to reason about reward and punishment.
Lex Fridman (1:01:04.980)
Responsibility.
Lex Fridman (1:01:06.420)
And?
Lex Fridman (1:01:07.940)
Artifactions.
Lex Fridman (1:01:08.780)
So far, I can only think about communication.
Lex Fridman (1:01:11.700)
Communication is, and not necessarily natural language,
Lex Fridman (1:01:15.380)
but just communication.
Lex Fridman (1:01:16.380)
Just communication.
Lex Fridman (1:01:17.580)
And that's important to have a quick and effective means
Lex Fridman (1:01:21.980)
of communicating knowledge.
Judea Pearl (1:01:24.100)
If the coach tells you you should have passed the ball,
Lex Fridman (1:01:26.460)
pink, he conveys so much knowledge to you
Lex Fridman (1:01:28.740)
as opposed to what?
Lex Fridman (1:01:30.460)
Go down and change your software.
Judea Pearl (1:01:33.420)
That's the alternative.
Lex Fridman (1:01:35.300)
But the coach doesn't know your software.
Lex Fridman (1:01:37.700)
So how can the coach tell you
Lex Fridman (1:01:39.580)
you should have passed the ball?
Lex Fridman (1:01:41.540)
But our language is very effective.
Lex Fridman (1:01:44.260)
You should have passed the ball.
Judea Pearl (1:01:45.580)
You know your software.
Lex Fridman (1:01:47.020)
You tweak the right module, and next time you don't do it.
Judea Pearl (1:01:52.180)
Now that's for playing soccer,
Lex Fridman (1:01:53.620)
the rules are well defined.
Judea Pearl (1:01:55.180)
No, no, no, no, they're not well defined.
Lex Fridman (1:01:57.420)
When you should pass the ball.
Judea Pearl (1:01:58.820)
Is not well defined.
Lex Fridman (1:02:00.020)
No, it's very soft, very noisy.
Judea Pearl (1:02:04.420)
Yes, you have to do it under pressure.
Lex Fridman (1:02:06.700)
It's art.
Lex Fridman (1:02:07.820)
But in terms of aligning values
Lex Fridman (1:02:11.340)
between computers and humans,
Lex Fridman (1:02:15.260)
do you think this cause and effect type of thinking
Lex Fridman (1:02:20.220)
is important to align the values,
Judea Pearl (1:02:22.300)
values, morals, ethics under which the machines
Lex Fridman (1:02:25.540)
make decisions, is the cause effect
Lex Fridman (1:02:28.500)
where the two can come together?
Lex Fridman (1:02:32.100)
Cause and effect is necessary component
Judea Pearl (1:02:34.780)
to build an ethical machine.
Lex Fridman (1:02:38.300)
Because the machine has to empathize
Judea Pearl (1:02:40.460)
to understand what's good for you,
Lex Fridman (1:02:42.660)
to build a model of you as a recipient,
Lex Fridman (1:02:47.180)
which should be very much, what is compassion?
Lex Fridman (1:02:50.900)
They imagine that you suffer pain as much as me.
Judea Pearl (1:02:56.020)
As much as me.
Lex Fridman (1:02:57.020)
I do have already a model of myself, right?
Lex Fridman (1:03:00.300)
So it's very easy for me to map you to mine.
Lex Fridman (1:03:02.780)
I don't have to rebuild the model.
Judea Pearl (1:03:04.620)
It's much easier to say, oh, you're like me.
Lex Fridman (1:03:06.940)
Okay, therefore I would not hate you.
Lex Fridman (1:03:09.620)
And the machine has to imagine,
Lex Fridman (1:03:12.100)
has to try to fake to be human essentially
Lex Fridman (1:03:14.540)
so you can imagine that you're like me, right?
Lex Fridman (1:03:19.420)
And moreover, who is me?
Judea Pearl (1:03:21.660)
That's the first, that's consciousness.
Lex Fridman (1:03:24.180)
They have a model of yourself.
Lex Fridman (1:03:26.860)
Where do you get this model?
Lex Fridman (1:03:28.100)
You look at yourself as if you are a part
Judea Pearl (1:03:30.980)
of the environment.
Lex Fridman (1:03:32.380)
If you build a model of yourself
Judea Pearl (1:03:33.940)
versus the environment, then you can say,
Lex Fridman (1:03:36.100)
I need to have a model of myself.
Lex Fridman (1:03:38.220)
I have abilities, I have desires and so forth, okay?
Lex Fridman (1:03:41.820)
I have a blueprint of myself though.
Judea Pearl (1:03:44.340)
Not the full detail because I cannot
Lex Fridman (1:03:46.140)
get the whole thing problem.
Lex Fridman (1:03:49.220)
But I have a blueprint.
Lex Fridman (1:03:50.700)
So on that level of a blueprint, I can modify things.
Judea Pearl (1:03:54.220)
I can look at myself in the mirror and say,
Lex Fridman (1:03:56.580)
hmm, if I change this model, tweak this model,
Judea Pearl (1:03:59.180)
I'm gonna perform differently.
Lex Fridman (1:04:02.340)
That is what we mean by free will.
Lex Fridman (1:04:05.740)
And consciousness.
Lex Fridman (1:04:06.940)
And consciousness.
Lex Fridman (1:04:08.260)
What do you think is consciousness?
Lex Fridman (1:04:10.380)
Is it simply self awareness?
Lex Fridman (1:04:11.820)
So including yourself into the model of the world?
Lex Fridman (1:04:14.660)
That's right.
Judea Pearl (1:04:15.500)
Some people tell me, no, this is only part of consciousness.
Lex Fridman (1:04:19.620)
And then they start telling me what they really mean
Judea Pearl (1:04:21.460)
by consciousness, and I lose them.
Lex Fridman (1:04:24.740)
For me, consciousness is having a blueprint
Judea Pearl (1:04:29.420)
of your software.
Lex Fridman (1:04:31.580)
Do you have concerns about the future of AI?
Lex Fridman (1:04:37.020)
All the different trajectories of all of our research?
Lex Fridman (1:04:39.660)
Yes.
Judea Pearl (1:04:40.700)
Where's your hope, where the movement has,
Lex Fridman (1:04:43.300)
where are your concerns?
Judea Pearl (1:04:44.380)
I'm concerned, because I know we are building a new species
Lex Fridman (1:04:49.500)
that has a capability of exceeding our, exceeding us,
Judea Pearl (1:04:56.820)
exceeding our capabilities, and can breed itself
Lex Fridman (1:05:01.300)
and take over the world.
Judea Pearl (1:05:02.540)
Absolutely.
Lex Fridman (1:05:03.580)
It's a new species that is uncontrolled.
Judea Pearl (1:05:07.660)
We don't know the degree to which we control it.
Lex Fridman (1:05:10.140)
We don't even understand what it means
Judea Pearl (1:05:12.660)
to be able to control this new species.
Lex Fridman (1:05:16.500)
So I'm concerned.
Judea Pearl (1:05:18.660)
I don't have anything to add to that,
Lex Fridman (1:05:21.100)
because it's such a gray area, it's unknown.
Judea Pearl (1:05:26.100)
It never happened in history.
Lex Fridman (1:05:29.820)
The only time it happened in history
Judea Pearl (1:05:34.260)
was evolution with human beings.
Lex Fridman (1:05:37.580)
It wasn't very successful, was it?
Judea Pearl (1:05:39.420)
Some people say it was a great success.
Lex Fridman (1:05:42.700)
For us it was, but a few people along the way,
Judea Pearl (1:05:46.380)
a few creatures along the way would not agree.
Lex Fridman (1:05:49.340)
So it's just because it's such a gray area,
Judea Pearl (1:05:53.100)
there's nothing else to say.
Lex Fridman (1:05:54.980)
We have a sample of one.
Judea Pearl (1:05:56.820)
Sample of one.
Lex Fridman (1:05:58.100)
It's us.
Lex Fridman (1:05:59.860)
But some people would look at you and say,
Lex Fridman (1:06:04.860)
yeah, but we were looking to you to help us
Judea Pearl (1:06:09.780)
make sure that the sample two works out okay.
Lex Fridman (1:06:13.180)
We have more than a sample of one.
Judea Pearl (1:06:14.860)
We have theories, and that's a good.
Lex Fridman (1:06:18.700)
We don't need to be statisticians.
Lex Fridman (1:06:20.740)
So sample of one doesn't mean poverty of knowledge.
Lex Fridman (1:06:25.420)
It's not.
Judea Pearl (1:06:26.460)
Sample of one plus theory, conjectural theory,
Lex Fridman (1:06:30.540)
of what could happen.
Judea Pearl (1:06:32.940)
That we do have.
Lex Fridman (1:06:34.380)
But I really feel helpless in contributing
Judea Pearl (1:06:38.260)
to this argument, because I know so little,
Lex Fridman (1:06:41.980)
and my imagination is limited,
Lex Fridman (1:06:46.640)
and I know how much I don't know,
Lex Fridman (1:06:50.140)
and I, but I'm concerned.
Judea Pearl (1:06:55.540)
You were born and raised in Israel.
Lex Fridman (1:06:57.780)
Born and raised in Israel, yes.
Lex Fridman (1:06:59.260)
And later served in Israel military,
Lex Fridman (1:07:03.900)
defense forces.
Judea Pearl (1:07:05.220)
In the Israel Defense Force.
Lex Fridman (1:07:07.980)
Yeah.
Lex Fridman (1:07:09.720)
What did you learn from that experience?
Lex Fridman (1:07:13.820)
From this experience?
Judea Pearl (1:07:16.540)
There's a kibbutz in there as well.
Lex Fridman (1:07:18.140)
Yes, because I was in the nachal,
Judea Pearl (1:07:20.540)
which is a combination of agricultural work
Lex Fridman (1:07:26.140)
and military service.
Judea Pearl (1:07:28.540)
We were supposed, I was really idealist.
Lex Fridman (1:07:31.220)
I wanted to be a member of the kibbutz throughout my life,
Lex Fridman (1:07:36.180)
and to live a communal life,
Lex Fridman (1:07:38.700)
and so I prepared myself for that.
Judea Pearl (1:07:46.060)
Slowly, slowly, I wanted a greater challenge.
Lex Fridman (1:07:51.360)
So that's a far world away, both.
Lex Fridman (1:07:55.340)
What I learned from that, what I can add,
Lex Fridman (1:07:57.780)
it was a miracle.
Judea Pearl (1:08:01.340)
It was a miracle that I served in the 1950s.
Lex Fridman (1:08:07.700)
I don't know how we survived.
Judea Pearl (1:08:11.380)
The country was under austerity.
Lex Fridman (1:08:15.380)
It tripled its population from 600,000 to a million point eight
Judea Pearl (1:08:21.460)
when I finished college.
Lex Fridman (1:08:23.320)
No one went hungry.
Lex Fridman (1:08:24.860)
And austerity, yes.
Lex Fridman (1:08:29.380)
When you wanted to make an omelet in a restaurant,
Judea Pearl (1:08:34.220)
you had to bring your own egg.
Lex Fridman (1:08:38.140)
And they imprisoned people from bringing the food
Judea Pearl (1:08:43.740)
from farming here, from the villages, to the city.
Lex Fridman (1:08:49.380)
But no one went hungry.
Lex Fridman (1:08:50.780)
And I always add to it,
Lex Fridman (1:08:53.620)
and higher education did not suffer any budget cut.
Judea Pearl (1:09:00.220)
They still invested in me, in my wife, in our generation
Lex Fridman (1:09:05.460)
to get the best education that they could, okay?
Lex Fridman (1:09:09.700)
So I'm really grateful for the opportunity,
Lex Fridman (1:09:15.200)
and I'm trying to pay back now, okay?
Judea Pearl (1:09:18.620)
It's a miracle that we survived the war of 1948.
Lex Fridman (1:09:22.900)
We were so close to a second genocide.
Judea Pearl (1:09:27.260)
It was all planned.
Lex Fridman (1:09:30.160)
But we survived it by miracle,
Lex Fridman (1:09:32.180)
and then the second miracle
Lex Fridman (1:09:33.580)
that not many people talk about, the next phase.
Lex Fridman (1:09:37.780)
How no one went hungry,
Lex Fridman (1:09:40.280)
and the country managed to triple its population.
Lex Fridman (1:09:43.940)
You know what it means to triple?
Lex Fridman (1:09:45.300)
Imagine United States going from what, 350 million
Judea Pearl (1:09:50.220)
to a trillion, unbelievable.
Lex Fridman (1:09:53.860)
So it's a really tense part of the world.
Judea Pearl (1:09:57.020)
It's a complicated part of the world,
Lex Fridman (1:09:59.060)
Israel and all around.
Judea Pearl (1:10:01.820)
Religion is at the core of that complexity.
Lex Fridman (1:10:08.100)
One of the components.
Judea Pearl (1:10:09.460)
Religion is a strong motivating cause
Lex Fridman (1:10:12.420)
to many, many people in the Middle East, yes.
Lex Fridman (1:10:16.500)
In your view, looking back, is religion good for society?
Lex Fridman (1:10:23.180)
That's a good question for robotic, you know?
Judea Pearl (1:10:26.900)
There's echoes of that question.
Lex Fridman (1:10:28.300)
Equip robot with religious belief.
Judea Pearl (1:10:32.260)
Suppose we find out, or we agree
Lex Fridman (1:10:34.660)
that religion is good to you, to keep you in line, okay?
Lex Fridman (1:10:37.940)
Should we give the robot the metaphor of a god?
Lex Fridman (1:10:43.340)
As a matter of fact, the robot will get it without us also.
Lex Fridman (1:10:47.340)
Why?
Lex Fridman (1:10:48.180)
The robot will reason by metaphor.
Lex Fridman (1:10:51.020)
And what is the most primitive metaphor
Lex Fridman (1:10:56.380)
a child grows with?
Judea Pearl (1:10:58.940)
Mother smile, father teaching,
Lex Fridman (1:11:02.940)
father image and mother image, that's god.
Judea Pearl (1:11:05.380)
So, whether you want it or not,
Lex Fridman (1:11:08.800)
the robot will, well, assuming that the robot
Judea Pearl (1:11:12.560)
is gonna have a mother and a father,
Lex Fridman (1:11:14.800)
it may only have a programmer,
Judea Pearl (1:11:16.360)
which doesn't supply warmth and discipline.
Lex Fridman (1:11:20.960)
Well, discipline it does.
Lex Fridman (1:11:22.440)
So the robot will have a model of the trainer,
Lex Fridman (1:11:27.360)
and everything that happens in the world,
Judea Pearl (1:11:29.320)
cosmology and so on, is going to be mapped
Lex Fridman (1:11:32.280)
into the programmer, it's god.
Judea Pearl (1:11:36.520)
Man, the thing that represents the origin
Lex Fridman (1:11:41.380)
of everything for that robot.
Judea Pearl (1:11:43.960)
It's the most primitive relationship.
Lex Fridman (1:11:46.340)
So it's gonna arrive there by metaphor.
Lex Fridman (1:11:48.680)
And so the question is if overall
Lex Fridman (1:11:51.880)
that metaphor has served us well as humans.
Judea Pearl (1:11:56.200)
I really don't know.
Lex Fridman (1:11:58.000)
I think it did, but as long as you keep
Judea Pearl (1:12:01.680)
in mind it's only a metaphor.
Lex Fridman (1:12:05.160)
So, if you think we can, can we talk about your son?
Judea Pearl (1:12:11.580)
Yes, yes.
Lex Fridman (1:12:13.240)
Can you tell his story?
Lex Fridman (1:12:15.080)
His story?
Lex Fridman (1:12:17.180)
Daniel?
Judea Pearl (1:12:18.020)
His story is known, he was abducted
Lex Fridman (1:12:21.200)
in Pakistan by Al Qaeda driven sect,
Lex Fridman (1:12:26.200)
and under various pretenses.
Lex Fridman (1:12:32.040)
I don't even pay attention to what the pretence was.
Judea Pearl (1:12:35.200)
Originally they wanted to have the United States
Lex Fridman (1:12:42.720)
deliver some promised airplanes.
Judea Pearl (1:12:47.520)
It was all made up, and all these demands were bogus.
Lex Fridman (1:12:51.520)
Bogus, I don't know really, but eventually
Judea Pearl (1:12:57.000)
he was executed in front of a camera.
Lex Fridman (1:13:03.680)
At the core of that is hate and intolerance.
Judea Pearl (1:13:07.400)
At the core, yes, absolutely, yes.
Lex Fridman (1:13:10.000)
We don't really appreciate the depth of the hate
Judea Pearl (1:13:15.000)
at which billions of peoples are educated.
Lex Fridman (1:13:26.080)
We don't understand it.
Judea Pearl (1:13:27.560)
I just listened recently to what they teach you
Lex Fridman (1:13:31.360)
in Mogadishu.
Judea Pearl (1:13:32.320)
Okay, okay, when the water stopped in the tap,
Lex Fridman (1:13:45.680)
we knew exactly who did it, the Jews.
Judea Pearl (1:13:49.360)
The Jews.
Lex Fridman (1:13:50.600)
We didn't know how, but we knew who did it.
Judea Pearl (1:13:55.280)
We don't appreciate what it means to us.
Lex Fridman (1:13:58.080)
The depth is unbelievable.
Lex Fridman (1:14:00.440)
Do you think all of us are capable of evil?
Lex Fridman (1:14:06.920)
And the education, the indoctrination
Judea Pearl (1:14:09.840)
is really what creates evil.
Lex Fridman (1:14:10.680)
Absolutely we are capable of evil.
Judea Pearl (1:14:12.240)
If you're indoctrinated sufficiently long and in depth,
Lex Fridman (1:14:18.800)
you're capable of ISIS, you're capable of Nazism.
Judea Pearl (1:14:24.840)
Yes, we are, but the question is whether we,
Lex Fridman (1:14:28.400)
after we have gone through some Western education
Lex Fridman (1:14:32.840)
and we learn that everything is really relative.
Lex Fridman (1:14:35.680)
It is not absolute God.
Judea Pearl (1:14:37.680)
It's only a belief in God.
Lex Fridman (1:14:40.120)
Whether we are capable now of being transformed
Judea Pearl (1:14:43.520)
under certain circumstances to become brutal.
Lex Fridman (1:14:49.920)
Yeah.
Judea Pearl (1:14:51.680)
I'm worried about it because some people say yes,
Lex Fridman (1:14:55.080)
given the right circumstances,
Judea Pearl (1:14:57.360)
given bad economical crisis,
Lex Fridman (1:15:04.120)
you are capable of doing it too.
Judea Pearl (1:15:06.160)
That worries me.
Lex Fridman (1:15:08.420)
I want to believe it, I'm not capable.
Lex Fridman (1:15:12.200)
So seven years after Daniel's death,
Lex Fridman (1:15:14.560)
you wrote an article at the Wall Street Journal
Judea Pearl (1:15:16.800)
titled Daniel Pearl and the Normalization of Evil.
Lex Fridman (1:15:19.760)
Yes.
Lex Fridman (1:15:20.580)
What was your message back then
Lex Fridman (1:15:23.040)
and how did it change today over the years?
Judea Pearl (1:15:27.560)
I lost.
Lex Fridman (1:15:30.600)
What was the message?
Judea Pearl (1:15:31.840)
The message was that we are not treating terrorism
Lex Fridman (1:15:39.960)
as a taboo.
Judea Pearl (1:15:41.840)
We are treating it as a bargaining device that is accepted.
Lex Fridman (1:15:46.840)
People have grievance and they go and bomb restaurants.
Judea Pearl (1:15:54.000)
It's normal.
Lex Fridman (1:15:55.240)
Look, you're even not surprised when I tell you that.
Lex Fridman (1:15:59.600)
20 years ago you say, what?
Lex Fridman (1:16:01.640)
For grievance you go and blow a restaurant?
Judea Pearl (1:16:05.000)
Today it's becoming normalized.
Lex Fridman (1:16:07.960)
The banalization of evil.
Lex Fridman (1:16:11.080)
And we have created that to ourselves by normalizing,
Lex Fridman (1:16:16.080)
by making it part of political life.
Judea Pearl (1:16:23.960)
It's a political debate.
Lex Fridman (1:16:27.760)
Every terrorist yesterday becomes a freedom fighter today
Lex Fridman (1:16:34.440)
and tomorrow it becomes terrorist again.
Lex Fridman (1:16:36.600)
It's switchable.
Judea Pearl (1:16:38.600)
Right, and so we should call out evil when there's evil.
Lex Fridman (1:16:42.220)
If we don't want to be part of it.
Judea Pearl (1:16:46.860)
Becoming.
Lex Fridman (1:16:49.020)
Yeah, if we want to separate good from evil,
Judea Pearl (1:16:52.300)
that's one of the first things that,
Lex Fridman (1:16:55.060)
what was it, in the Garden of Eden,
Judea Pearl (1:16:57.540)
remember the first thing that God told him was,
Lex Fridman (1:17:02.900)
hey, you want some knowledge, here's a tree of good and evil.
Judea Pearl (1:17:07.380)
Yeah, so this evil touched your life personally.
Lex Fridman (1:17:12.260)
Does your heart have anger, sadness, or is it hope?
Judea Pearl (1:17:20.300)
Look, I see some beautiful people coming from Pakistan.
Lex Fridman (1:17:26.140)
I see beautiful people everywhere.
Lex Fridman (1:17:29.420)
But I see horrible propagation of evil in this country too.
Lex Fridman (1:17:38.700)
It shows you how populistic slogans
Judea Pearl (1:17:42.780)
can catch the mind of the best intellectuals.
Lex Fridman (1:17:48.220)
Today is Father's Day.
Judea Pearl (1:17:50.100)
I didn't know that.
Lex Fridman (1:17:51.060)
Yeah, what's a fond memory you have of Daniel?
Lex Fridman (1:17:56.060)
What's a fond memory you have of Daniel?
Lex Fridman (1:17:58.380)
Oh, very good memories, immense.
Judea Pearl (1:18:02.660)
He was my mentor.
Lex Fridman (1:18:06.100)
He had a sense of balance that I didn't have.
Judea Pearl (1:18:15.340)
He saw the beauty in every person.
Lex Fridman (1:18:19.460)
He was not as emotional as I am,
Judea Pearl (1:18:22.020)
the more looking things in perspective.
Lex Fridman (1:18:26.220)
He really liked every person.
Judea Pearl (1:18:29.340)
He really grew up with the idea that a foreigner
Lex Fridman (1:18:34.740)
is a reason for curiosity, not for fear.
Judea Pearl (1:18:42.780)
That one time we went in Berkeley,
Lex Fridman (1:18:45.280)
and a homeless came out from some dark alley,
Lex Fridman (1:18:49.220)
and said, hey, man, can you spare a dime?
Lex Fridman (1:18:51.180)
I retreated back, two feet back,
Lex Fridman (1:18:54.620)
and then I just hugged him and say,
Lex Fridman (1:18:56.220)
here's a dime, enjoy yourself.
Judea Pearl (1:18:58.420)
Maybe you want some money to take a bus or whatever.
Lex Fridman (1:19:05.340)
Where did you get it?
Judea Pearl (1:19:06.580)
Not from me.
Lex Fridman (1:19:10.460)
Do you have advice for young minds today,
Judea Pearl (1:19:12.400)
dreaming about creating as you have dreamt,
Lex Fridman (1:19:16.200)
creating intelligent systems?
Lex Fridman (1:19:17.880)
What is the best way to arrive at new breakthrough ideas
Lex Fridman (1:19:21.380)
and carry them through the fire of criticism
Lex Fridman (1:19:23.820)
and past conventional ideas?
Lex Fridman (1:19:28.780)
Ask your questions freely.
Judea Pearl (1:19:34.920)
Your questions are never dumb.
Lex Fridman (1:19:37.660)
And solve them your own way.
Lex Fridman (1:19:40.700)
And don't take no for an answer.
Lex Fridman (1:19:42.680)
Look, if they are really dumb,
Judea Pearl (1:19:46.240)
you will find out quickly by trying an arrow
Lex Fridman (1:19:49.880)
to see that they're not leading any place.
Lex Fridman (1:19:52.420)
But follow them and try to understand things your way.
Lex Fridman (1:19:59.440)
That is my advice.
Judea Pearl (1:20:01.600)
I don't know if it's gonna help anyone.
Lex Fridman (1:20:03.980)
Not as brilliantly.
Judea Pearl (1:20:05.520)
There is a lot of inertia in science, in academia.
Lex Fridman (1:20:15.120)
It is slowing down science.
Judea Pearl (1:20:18.560)
Yeah, those two words, your way, that's a powerful thing.
Lex Fridman (1:20:23.320)
It's against inertia, potentially, against the flow.
Judea Pearl (1:20:27.000)
Against your professor.
Lex Fridman (1:20:28.640)
Against your professor.
Judea Pearl (1:20:30.440)
I wrote the Book of Why in order to democratize
Lex Fridman (1:20:34.840)
common sense.
Judea Pearl (1:20:38.680)
In order to instill rebellious spirit in students
Lex Fridman (1:20:45.460)
so they wouldn't wait until the professor get things right.
Lex Fridman (1:20:53.720)
So you wrote the manifesto of the rebellion
Lex Fridman (1:20:56.640)
against the professor.
Judea Pearl (1:20:58.360)
Against the professor, yes.
Lex Fridman (1:21:00.360)
So looking back at your life of research,
Lex Fridman (1:21:02.840)
what ideas do you hope ripple through the next many decades?
Lex Fridman (1:21:06.200)
What do you hope your legacy will be?
Judea Pearl (1:21:10.600)
I already have a tombstone carved.
Lex Fridman (1:21:19.880)
Oh, boy.
Judea Pearl (1:21:21.560)
The fundamental law of counterfactuals.
Lex Fridman (1:21:25.840)
That's what, it's a simple equation.
Judea Pearl (1:21:29.800)
Counterfactual in terms of a model surgery.
Lex Fridman (1:21:35.560)
That's it, because everything follows from that.
Judea Pearl (1:21:40.000)
If you get that, all the rest, I can die in peace.
Lex Fridman (1:21:46.600)
And my student can derive all my knowledge
Judea Pearl (1:21:49.640)
by mathematical means.
Lex Fridman (1:21:51.920)
The rest follows.
Judea Pearl (1:21:53.520)
Yeah.
Lex Fridman (1:21:55.040)
Thank you so much for talking today.
Judea Pearl (1:21:56.420)
I really appreciate it.
Lex Fridman (1:21:57.260)
Thank you for being so attentive and instigating.
Judea Pearl (1:22:03.920)
We did it.
Lex Fridman (1:22:05.240)
We did it.
Judea Pearl (1:22:06.080)
The coffee helped.
Lex Fridman (1:22:07.520)
Thanks for listening to this conversation with Judea Pearl.
Lex Fridman (1:22:11.280)
And thank you to our presenting sponsor, Cash App.
Lex Fridman (1:22:14.280)
Download it, use code LexPodcast, you'll get $10,
Lex Fridman (1:22:18.320)
and $10 will go to FIRST, a STEM education nonprofit
Lex Fridman (1:22:21.960)
that inspires hundreds of thousands of young minds
Judea Pearl (1:22:24.880)
to learn and to dream of engineering our future.
Lex Fridman (1:22:28.320)
If you enjoy this podcast, subscribe on YouTube,
Judea Pearl (1:22:31.120)
give it five stars on Apple Podcast,
Lex Fridman (1:22:33.040)
support on Patreon, or simply connect with me on Twitter.
Lex Fridman (1:22:36.880)
And now, let me leave you with some words of wisdom
Lex Fridman (1:22:39.320)
from Judea Pearl.
Judea Pearl (1:22:41.080)
You cannot answer a question that you cannot ask,
Lex Fridman (1:22:44.000)
and you cannot ask a question that you have no words for.
Judea Pearl (1:22:47.400)
Thank you for listening, and hope to see you next time.
Lex Fridman (20:07.440)
the people from Israel that were taken in exile
Judea Pearl (20:12.760)
to Babylon to serve the king.
Lex Fridman (20:14.720)
He wanted to serve them king's food, which was meat.
Lex Fridman (20:18.080)
And Daniel as a good Jew couldn't eat non kosher food.
Lex Fridman (20:22.760)
So he asked them to eat vegetarian food.
Lex Fridman (20:26.640)
But the king overseer says, I'm sorry,
Lex Fridman (20:29.320)
but if the king sees that your performance falls below
Judea Pearl (20:34.760)
that of other kids, he's going to kill me.
Lex Fridman (20:39.360)
Daniel said, let's make an experiment.
Judea Pearl (20:41.600)
Let's take four of us from Jerusalem, okay.
Lex Fridman (20:44.320)
Give us vegetarian food.
Judea Pearl (20:46.440)
Let's take the other guys to eat the king's food.
Lex Fridman (20:50.280)
And in about a week's time, we'll test our performance.
Lex Fridman (20:54.080)
And you know the answer.
Lex Fridman (20:55.440)
Of course he did the experiment.
Lex Fridman (20:57.840)
And they were so much better than the others.
Lex Fridman (21:02.120)
And the kings nominated them to super position in his king.
Lex Fridman (21:08.000)
So it was a first experiment, yes.
Lex Fridman (21:10.160)
So there was a very simple,
Judea Pearl (21:12.760)
it's also the same research questions.
Lex Fridman (21:15.520)
We want to know if vegetarian food
Judea Pearl (21:18.640)
assist or obstruct your mental ability.
Lex Fridman (21:23.480)
And okay, so the question is very old one.
Judea Pearl (21:30.440)
Even Democritus said, if I could discover one cause
Lex Fridman (21:38.040)
of things, I would rather discover one cause
Lex Fridman (21:41.480)
and be a king of Persia, okay.
Lex Fridman (21:43.720)
The task of discovering causes was in the mind
Judea Pearl (21:48.720)
of ancient people from many, many years ago.
Lex Fridman (21:53.480)
But the mathematics of doing that was only developed
Judea Pearl (21:58.360)
in the 1920s.
Lex Fridman (22:00.480)
So science has left us orphan, okay.
Judea Pearl (22:05.120)
Science has not provided us with the mathematics
Lex Fridman (22:08.320)
to capture the idea of X causes Y
Lex Fridman (22:12.000)
and Y does not cause X.
Lex Fridman (22:14.320)
Because all the question of physics are symmetrical,
Judea Pearl (22:17.400)
algebraic, the equality sign goes both ways.
Lex Fridman (22:21.680)
Okay, let's look at machine learning.
Judea Pearl (22:23.120)
Machine learning today, if you look at deep neural networks,
Lex Fridman (22:26.880)
you can think of it as kind of conditional probability
Judea Pearl (22:32.840)
estimators.
Lex Fridman (22:33.680)
Correct, beautiful.
Lex Fridman (22:35.520)
So where did you say that?
Lex Fridman (22:39.280)
Conditional probability estimators.
Lex Fridman (22:41.520)
None of the machine learning people clevered you?
Lex Fridman (22:45.360)
Attacked you?
Judea Pearl (22:46.840)
Listen, most people, and this is why today's conversation
Lex Fridman (22:52.400)
I think is interesting, is most people would agree with you.
Judea Pearl (22:55.800)
There's certain aspects that are just effective today,
Lex Fridman (22:58.680)
but we're going to hit a wall and there's a lot of ideas.
Judea Pearl (23:02.440)
I think you're very right that we're gonna have to return to
Lex Fridman (23:05.400)
about causality and it would be, let's try to explore it.
Judea Pearl (23:11.560)
Let's even take a step back.
Lex Fridman (23:13.160)
You've invented Bayesian networks that look awfully a lot
Judea Pearl (23:19.800)
like they express something like causation,
Lex Fridman (23:22.040)
but they don't, not necessarily.
Lex Fridman (23:25.280)
So how do we turn Bayesian networks
Lex Fridman (23:28.600)
into expressing causation?
Lex Fridman (23:30.840)
How do we build causal networks?
Lex Fridman (23:33.160)
This A causes B, B causes C,
Lex Fridman (23:36.480)
how do we start to infer that kind of thing?
Lex Fridman (23:38.840)
We start asking ourselves question,
Lex Fridman (23:41.460)
what are the factors that would determine
Lex Fridman (23:44.520)
the value of X?
Judea Pearl (23:46.360)
X could be blood pressure, death, hunger.
Lex Fridman (23:53.820)
But these are hypotheses that we propose for ourselves.
Judea Pearl (23:56.240)
Hypothesis, everything which has to do with causality
Lex Fridman (23:59.040)
comes from a theory.
Judea Pearl (24:03.400)
The difference is only how you interrogate
Lex Fridman (24:06.960)
the theory that you have in your mind.
Lex Fridman (24:09.080)
So it still needs the human expert to propose.
Lex Fridman (24:14.280)
You need the human expert to specify the initial model.
Judea Pearl (24:20.980)
Initial model could be very qualitative.
Lex Fridman (24:24.040)
Just who listens to whom?
Judea Pearl (24:27.040)
By whom listen to, I mean one variable listen to the other.
Lex Fridman (24:31.240)
So I say, okay, the tide is listening to the moon.
Lex Fridman (24:34.720)
And not to the rooster crow.
Lex Fridman (24:42.280)
And so forth.
Judea Pearl (24:43.120)
This is our understanding of the world in which we live.
Lex Fridman (24:46.840)
Scientific understanding of reality.
Judea Pearl (24:51.540)
We have to start there.
Lex Fridman (24:53.480)
Because if we don't know how to handle
Judea Pearl (24:56.960)
cause and effect relationship,
Lex Fridman (24:58.560)
when we do have a model,
Lex Fridman (25:01.240)
and we certainly do not know how to handle it
Lex Fridman (25:03.720)
when we don't have a model.
Lex Fridman (25:05.400)
So let's start first.
Lex Fridman (25:07.240)
In AI, slogan is representation first, discovery second.
Lex Fridman (25:13.440)
But if I give you all the information that you need,
Lex Fridman (25:17.240)
can you do anything useful with it?
Judea Pearl (25:19.840)
That is the first, representation.
Lex Fridman (25:21.520)
How do you represent it?
Judea Pearl (25:22.520)
I give you all the knowledge in the world.
Lex Fridman (25:24.560)
How do you represent it?
Judea Pearl (25:26.920)
When you represent it, I ask you,
Lex Fridman (25:30.680)
can you infer X or Y or Z?
Lex Fridman (25:33.200)
Can you answer certain queries?
Lex Fridman (25:35.240)
Is it complex?
Lex Fridman (25:36.920)
Is it polynomial?
Lex Fridman (25:39.280)
All the computer science exercises we do,
Judea Pearl (25:42.040)
once you give me a representation for my knowledge,
Lex Fridman (25:47.260)
then you can ask me,
Judea Pearl (25:48.280)
now I understand how to represent things.
Lex Fridman (25:51.680)
How do I discover them?
Judea Pearl (25:52.800)
It's a secondary thing.
Lex Fridman (25:54.920)
So first of all, I should echo the statement
Judea Pearl (25:57.040)
that mathematics and the current,
Lex Fridman (25:59.840)
much of the machine learning world has not considered
Judea Pearl (26:04.280)
causation that A causes B, just in anything.
Lex Fridman (26:07.880)
So that seems like a non obvious thing
Judea Pearl (26:15.360)
that you think we would have really acknowledged it,
Lex Fridman (26:18.280)
but we haven't.
Lex Fridman (26:19.200)
So we have to put that on the table.
Lex Fridman (26:21.080)
So knowledge, how hard is it to create a knowledge
Lex Fridman (26:26.420)
from which to work?
Lex Fridman (26:28.480)
In certain area, it's easy,
Judea Pearl (26:31.280)
because we have only four or five major variables,
Lex Fridman (26:37.400)
and an epidemiologist or an economist can put them down,
Judea Pearl (26:43.440)
what, minimum wage, unemployment, policy, X, Y, Z,
Lex Fridman (26:52.040)
and start collecting data,
Lex Fridman (26:54.320)
and quantify the parameters that were left unquantified
Lex Fridman (27:00.240)
with the initial knowledge.
Judea Pearl (27:01.560)
That's the routine work that you find
Lex Fridman (27:07.680)
in experimental psychology, in economics, everywhere.
Judea Pearl (27:13.000)
In the health science, that's a routine thing.
Lex Fridman (27:16.600)
But I should emphasize,
Judea Pearl (27:18.800)
you should start with the research question.
Lex Fridman (27:21.240)
What do you want to estimate?
Judea Pearl (27:24.880)
Once you have that, you have to have a language
Lex Fridman (27:27.480)
of expressing what you want to estimate.
Lex Fridman (27:30.200)
You think it's easy?
Lex Fridman (27:31.680)
No.
Lex Fridman (27:32.760)
So we can talk about two things, I think.
Lex Fridman (27:35.840)
One is how the science of causation is very useful
Judea Pearl (27:42.920)
for answering certain questions.
Lex Fridman (27:47.360)
And then the other is, how do we create intelligent systems
Lex Fridman (27:51.240)
that need to reason with causation?
Lex Fridman (27:53.600)
So if my research question is,
Lex Fridman (27:55.200)
how do I pick up this water bottle from the table?
Lex Fridman (28:00.920)
All of the knowledge that is required
Judea Pearl (28:03.560)
to be able to do that,
Lex Fridman (28:05.280)
how do we construct that knowledge base?
Judea Pearl (28:07.960)
Do we return back to the problem
Lex Fridman (28:11.000)
that we didn't solve in the 80s with expert systems?
Judea Pearl (28:13.580)
Do we have to solve that problem
Lex Fridman (28:15.440)
of automated construction of knowledge?
Judea Pearl (28:19.680)
You're talking about the task
Lex Fridman (28:23.560)
of eliciting knowledge from an expert.
Judea Pearl (28:26.560)
Task of eliciting knowledge from an expert,
Lex Fridman (28:28.520)
or the self discovery of more knowledge,
Judea Pearl (28:32.720)
more and more knowledge.
Lex Fridman (28:34.240)
So automating the building of knowledge as much as possible.
Judea Pearl (28:38.600)
It's a different game in the causal domain,
Lex Fridman (28:42.400)
because it's essentially the same thing.
Judea Pearl (28:46.480)
You have to start with some knowledge,
Lex Fridman (28:48.680)
and you're trying to enrich it.
Lex Fridman (28:51.520)
But you don't enrich it by asking for more rules.
Lex Fridman (28:56.460)
You enrich it by asking for the data,
Judea Pearl (28:58.960)
to look at the data and quantifying,
Lex Fridman (29:01.800)
and ask queries that you couldn't answer when you started.
Judea Pearl (29:06.480)
You couldn't because the question is quite complex,
Lex Fridman (29:11.480)
and it's not within the capability
Judea Pearl (29:16.600)
of ordinary cognition, of ordinary person,
Lex Fridman (29:19.840)
or ordinary expert even, to answer.
Lex Fridman (29:23.200)
So what kind of questions do you think
Lex Fridman (29:24.920)
we can start to answer?
Judea Pearl (29:26.960)
Even a simple one.
Lex Fridman (29:27.800)
Suppose, yeah, I'll start with easy one.
Judea Pearl (29:31.200)
Let's do it.
Lex Fridman (29:32.040)
Okay, what's the effect of a drug on recovery?
Lex Fridman (29:37.520)
What is the aspirin that caused my headache
Lex Fridman (29:39.500)
to be cured, or what did the television program,
Lex Fridman (29:44.640)
or the good news I received?
Lex Fridman (29:47.680)
This is already, you see, it's a difficult question,
Judea Pearl (29:49.920)
because it's find the cause from effect.
Lex Fridman (29:53.680)
The easy one is find the effect from cause.
Judea Pearl (29:56.760)
That's right.
Lex Fridman (29:57.720)
So first you construct a model,
Judea Pearl (29:59.180)
saying that this is an important research question.
Lex Fridman (30:01.280)
This is an important question.
Judea Pearl (30:02.840)
Then you do.
Lex Fridman (30:04.320)
I didn't construct a model yet.
Judea Pearl (30:05.520)
I just said it's an important question.
Lex Fridman (30:07.680)
And the first exercise is express it mathematically.
Lex Fridman (30:12.280)
What do you want to do?
Lex Fridman (30:13.800)
Like, if I tell you what will be the effect
Judea Pearl (30:17.000)
of taking this drug, you have to say that in mathematics.
Lex Fridman (30:21.320)
How do you say that?
Judea Pearl (30:22.880)
Yes.
Lex Fridman (30:23.720)
Can you write down the question, not the answer?
Judea Pearl (30:27.720)
I want to find the effect of the drug on my headache.
Lex Fridman (30:32.420)
Right.
Judea Pearl (30:33.260)
Write down, write it down.
Lex Fridman (30:34.680)
That's where the do calculus comes in.
Judea Pearl (30:35.920)
Yes.
Lex Fridman (30:36.760)
The do operator, what is the do operator?
Judea Pearl (30:38.240)
Do operator, yeah.
Lex Fridman (30:40.080)
Which is nice.
Judea Pearl (30:40.920)
It's the difference in association and intervention.
Lex Fridman (30:43.280)
Very beautifully sort of constructed.
Judea Pearl (30:45.760)
Yeah, so we have a do operator.
Lex Fridman (30:48.880)
So the do calculus connected on the do operator itself
Judea Pearl (30:52.560)
connects the operation of doing
Lex Fridman (30:55.560)
to something that we can see.
Judea Pearl (30:57.880)
Right.
Lex Fridman (30:58.720)
So as opposed to the purely observing,
Judea Pearl (31:01.740)
you're making the choice to change a variable.
Lex Fridman (31:05.920)
That's what it expresses.
Lex Fridman (31:08.240)
And then the way that we interpret it,
Lex Fridman (31:11.840)
the mechanism by which we take your query
Lex Fridman (31:15.400)
and we translate it into something that we can work with
Lex Fridman (31:18.640)
is by giving it semantics,
Judea Pearl (31:21.040)
saying that you have a model of the world
Lex Fridman (31:23.320)
and you cut off all the incoming error into X
Lex Fridman (31:26.840)
and you're looking now in the modified mutilated model,
Lex Fridman (31:30.680)
you ask for the probability of Y.
Judea Pearl (31:33.660)
That is interpretation of doing X
Lex Fridman (31:36.360)
because by doing things, you've liberated them
Judea Pearl (31:40.240)
from all influences that acted upon them earlier
Lex Fridman (31:45.200)
and you subject them to the tyranny of your muscles.
Lex Fridman (31:50.080)
So you remove all the questions about causality
Lex Fridman (31:54.040)
by doing them.
Lex Fridman (31:55.740)
So you're now.
Lex Fridman (31:56.580)
There's one level of questions.
Judea Pearl (31:58.160)
Yeah.
Lex Fridman (31:59.000)
Answer questions about what will happen if you do things.
Judea Pearl (32:01.920)
If you do, if you drink the coffee,
Lex Fridman (32:03.320)
if you take the aspirin.
Judea Pearl (32:04.160)
Right.
Lex Fridman (32:05.320)
So how do we get the doing data?
Judea Pearl (32:11.360)
Now the question is, if we cannot run experiments,
Lex Fridman (32:16.560)
then we have to rely on observational study.
Lex Fridman (32:20.960)
So first we could, sorry to interrupt,
Lex Fridman (32:22.560)
we could run an experiment where we do something,
Judea Pearl (32:25.880)
where we drink the coffee and this,
Lex Fridman (32:28.200)
the do operator allows you to sort of be systematic
Judea Pearl (32:31.120)
about expressing.
Lex Fridman (32:31.960)
So imagine how the experiment will look like
Judea Pearl (32:34.560)
even though we cannot physically
Lex Fridman (32:36.880)
and technologically conduct it.
Judea Pearl (32:38.760)
I'll give you an example.
Lex Fridman (32:40.600)
What is the effect of blood pressure on mortality?
Judea Pearl (32:44.960)
I cannot go down into your vein
Lex Fridman (32:47.360)
and change your blood pressure,
Lex Fridman (32:49.400)
but I can ask the question,
Lex Fridman (32:52.060)
which means I can even have a model of your body.
Judea Pearl (32:55.120)
I can imagine the effect of your,
Lex Fridman (32:58.600)
how the blood pressure change will affect your mortality.
Lex Fridman (33:04.720)
How?
Lex Fridman (33:05.560)
I go into the model and I conduct this surgery
Judea Pearl (33:09.760)
about the blood pressure,
Lex Fridman (33:12.080)
even though physically I can do, I cannot do it.
Judea Pearl (33:17.200)
Let me ask the quantum mechanics question.
Lex Fridman (33:19.760)
Does the doing change the observation?
Judea Pearl (33:23.720)
Meaning the surgery of changing the blood pressure is,
Lex Fridman (33:28.240)
I mean.
Judea Pearl (33:29.080)
No, the surgery is,
Lex Fridman (33:31.160)
I call the very delicate.
Judea Pearl (33:35.440)
It's very delicate, infinitely delicate.
Lex Fridman (33:37.840)
Incisive and delicate, which means,
Judea Pearl (33:40.760)
do means, do X means,
Lex Fridman (33:44.720)
I'm gonna touch only X.
Judea Pearl (33:46.480)
Only X.
Lex Fridman (33:47.320)
Directly into X.
Lex Fridman (33:50.160)
So that means that I change only things
Lex Fridman (33:52.800)
which depends on X by virtue of X changing,
Lex Fridman (33:56.800)
but I don't depend things which are not depend on X.
Lex Fridman (34:00.400)
Like I wouldn't change your sex or your age,
Judea Pearl (34:04.240)
I just change your blood pressure.
Lex Fridman (34:06.880)
So in the case of blood pressure,
Judea Pearl (34:08.720)
it may be difficult or impossible
Lex Fridman (34:11.160)
to construct such an experiment.
Judea Pearl (34:12.800)
No, physically yes, but hypothetically no.
Lex Fridman (34:16.560)
Hypothetically no.
Judea Pearl (34:17.400)
If we have a model, that is what the model is for.
Lex Fridman (34:20.720)
So you conduct surgeries on a model,
Judea Pearl (34:24.640)
you take it apart, put it back,
Lex Fridman (34:26.880)
that's the idea of a model.
Judea Pearl (34:28.880)
It's the idea of thinking counterfactually, imagining,
Lex Fridman (34:31.640)
and that's the idea of creativity.
Lex Fridman (34:35.120)
So by constructing that model,
Lex Fridman (34:36.400)
you can start to infer if the higher the blood pressure
Judea Pearl (34:43.040)
leads to mortality, which increases or decreases by.
Lex Fridman (34:47.320)
I construct the model, I still cannot answer it.
Judea Pearl (34:50.800)
I have to see if I have enough information in the model
Lex Fridman (34:53.800)
that would allow me to find out the effects of intervention
Judea Pearl (34:58.360)
from a noninterventional study, hence of study.
Lex Fridman (35:04.520)
So what's needed?
Judea Pearl (35:06.240)
You need to have assumptions about who affects whom.
Lex Fridman (35:13.480)
If the graph had a certain property,
Judea Pearl (35:16.400)
the answer is yes, you can get it from observational study.
Lex Fridman (35:20.560)
If the graph is too mushy, bushy, bushy,
Judea Pearl (35:23.760)
the answer is no, you cannot.
Lex Fridman (35:25.680)
Then you need to find either different kind of observation
Judea Pearl (35:30.680)
that you haven't considered, or one experiment.
Lex Fridman (35:35.120)
So basically, that puts a lot of pressure on you
Judea Pearl (35:38.880)
to encode wisdom into that graph.
Lex Fridman (35:41.920)
Correct.
Lex Fridman (35:42.960)
But you don't have to encode more than what you know.
Lex Fridman (35:47.520)
God forbid, if you put the,
Judea Pearl (35:49.680)
like economists are doing this,
Lex Fridman (35:51.400)
they call identifying assumption.
Judea Pearl (35:52.880)
They put assumptions, even if they don't prevail
Lex Fridman (35:55.120)
in the world, they put assumptions
Lex Fridman (35:56.840)
so they can identify things.
Lex Fridman (35:59.240)
But the problem is, yes, beautifully put,
Lex Fridman (36:01.480)
but the problem is you don't know what you don't know.
Lex Fridman (36:04.440)
So.
Judea Pearl (36:05.920)
You know what you don't know.
Lex Fridman (36:07.560)
Because if you don't know, you say it's possible.
Judea Pearl (36:10.600)
It's possible that X affect the traffic tomorrow.
Lex Fridman (36:17.720)
It's possible.
Judea Pearl (36:18.640)
You put down an arrow which says it's possible.
Lex Fridman (36:20.880)
Every arrow in the graph says it's possible.
Lex Fridman (36:23.960)
So there's not a significant cost to adding arrows that.
Lex Fridman (36:28.000)
The more arrow you add, the less likely you are
Judea Pearl (36:32.240)
to identify things from purely observational data.
Lex Fridman (36:37.680)
So if the whole world is bushy,
Lex Fridman (36:41.360)
and everybody affect everybody else,
Lex Fridman (36:45.400)
the answer is, you can answer it ahead of time.
Judea Pearl (36:49.160)
I cannot answer my query from observational data.
Lex Fridman (36:54.240)
I have to go to experiments.
Lex Fridman (36:56.640)
So you talk about machine learning
Lex Fridman (36:58.360)
is essentially learning by association,
Judea Pearl (37:01.600)
or reasoning by association,
Lex Fridman (37:03.120)
and this do calculus is allowing for intervention.
Judea Pearl (37:07.120)
I like that word.
Lex Fridman (37:09.000)
Action.
Lex Fridman (37:09.840)
So you also talk about counterfactuals.
Lex Fridman (37:12.400)
Yeah.
Lex Fridman (37:13.220)
And trying to sort of understand the difference
Lex Fridman (37:15.880)
between counterfactuals and intervention.
Judea Pearl (37:19.360)
First of all, what is counterfactuals,
Lex Fridman (37:22.320)
and why are they useful?
Lex Fridman (37:26.200)
Why are they especially useful,
Lex Fridman (37:29.700)
as opposed to just reasoning what effect actions have?
Judea Pearl (37:34.880)
Well, counterfactual contains
Lex Fridman (37:37.040)
what we normally call explanations.
Lex Fridman (37:40.280)
Can you give an example of a counterfactual?
Lex Fridman (37:41.600)
If I tell you that acting one way affects something else,
Judea Pearl (37:45.240)
I didn't explain anything yet.
Lex Fridman (37:47.720)
But if I ask you, was it the aspirin that cured my headache?
Lex Fridman (37:55.400)
I'm asking for explanation, what cured my headache?
Lex Fridman (37:58.640)
And putting a finger on aspirin provide an explanation.
Judea Pearl (38:04.640)
It was aspirin that was responsible
Lex Fridman (38:08.160)
for your headache going away.
Judea Pearl (38:11.560)
If you didn't take the aspirin,
Lex Fridman (38:14.420)
you would still have a headache.
Lex Fridman (38:16.000)
So by saying if I didn't take aspirin,
Lex Fridman (38:20.280)
I would have a headache, you're thereby saying
Judea Pearl (38:22.760)
that aspirin is the thing that removes the headache.
Lex Fridman (38:26.360)
But you have to have another important information.
Judea Pearl (38:30.440)
I took the aspirin, and my headache is gone.
Lex Fridman (38:34.560)
It's very important information.
Judea Pearl (38:36.400)
Now I'm reasoning backward,
Lex Fridman (38:38.080)
and I said, was it the aspirin?
Judea Pearl (38:40.520)
Yeah.
Lex Fridman (38:42.080)
By considering what would have happened
Judea Pearl (38:44.440)
if everything else is the same, but I didn't take aspirin.
Lex Fridman (38:47.000)
That's right.
Lex Fridman (38:47.840)
So you know that things took place.
Lex Fridman (38:51.040)
Joe killed Schmoe, and Schmoe would be alive
Judea Pearl (38:56.360)
had John not used his gun.
Lex Fridman (38:59.320)
Okay, so that is the counterfactual.
Judea Pearl (39:02.080)
It had the conflict here, or clash,
Lex Fridman (39:06.640)
between observed fact,
Lex Fridman (39:09.600)
but he did shoot, okay?
Lex Fridman (39:13.600)
And the hypothetical predicate,
Judea Pearl (39:16.600)
which says had he not shot,
Lex Fridman (39:18.800)
you have a logical clash.
Judea Pearl (39:21.540)
They cannot exist together.
Lex Fridman (39:23.820)
That's the counterfactual.
Lex Fridman (39:24.920)
And that is the source of our explanation
Lex Fridman (39:28.280)
of the idea of responsibility, regret, and free will.
Judea Pearl (39:34.820)
Yeah, so it certainly seems
Lex Fridman (39:37.200)
that's the highest level of reasoning, right?
Judea Pearl (39:39.760)
Yeah, and physicists do it all the time.
Lex Fridman (39:41.880)
Who does it all the time?
Judea Pearl (39:42.760)
Physicists.
Lex Fridman (39:43.580)
Physicists.
Judea Pearl (39:44.920)
In every equation of physics,
Lex Fridman (39:47.120)
let's say you have a Hooke's law,
Lex Fridman (39:49.560)
and you put one kilogram on the spring,
Lex Fridman (39:52.240)
and the spring is one meter,
Lex Fridman (39:54.680)
and you say, had this weight been two kilogram,
Lex Fridman (39:58.400)
the spring would have been twice as long.
Judea Pearl (40:02.060)
It's no problem for physicists to say that,
Lex Fridman (40:05.560)
except that mathematics is only in the form of equation,
Judea Pearl (40:10.240)
okay, equating the weight,
Lex Fridman (40:14.680)
proportionality constant, and the length of the string.
Lex Fridman (40:18.520)
So you don't have the asymmetry
Lex Fridman (40:22.080)
in the equation of physics,
Judea Pearl (40:23.280)
although every physicist thinks counterfactually.
Lex Fridman (40:26.840)
Ask the high school kids,
Judea Pearl (40:29.160)
had the weight been three kilograms,
Lex Fridman (40:31.120)
what would be the length of the spring?
Judea Pearl (40:33.360)
They can answer it immediately,
Lex Fridman (40:35.160)
because they do the counterfactual processing in their mind,
Lex Fridman (40:38.760)
and then they put it into equation,
Lex Fridman (40:41.120)
algebraic equation, and they solve it, okay?
Lex Fridman (40:44.280)
But a robot cannot do that.
Lex Fridman (40:46.720)
How do you make a robot learn these relationships?
Lex Fridman (40:51.000)
Well, why you would learn?
Lex Fridman (40:53.240)
Suppose you tell him, can you do it?
Lex Fridman (40:55.560)
So before you go learning,
Lex Fridman (40:57.520)
you have to ask yourself,
Lex Fridman (40:59.360)
suppose I give you all the information, okay?
Lex Fridman (41:01.820)
Can the robot perform the task that I ask him to perform?
Judea Pearl (41:07.820)
Can he reason and say, no, it wasn't the aspirin.
Lex Fridman (41:10.980)
It was the good news you received on the phone.
Judea Pearl (41:15.500)
Right, because, well, unless the robot had a model,
Lex Fridman (41:19.620)
a causal model of the world.
Judea Pearl (41:23.620)
Right, right.
Lex Fridman (41:24.740)
I'm sorry I have to linger on this.
Lex Fridman (41:26.260)
But now we have to linger and we have to say,
Lex Fridman (41:27.860)
how do we do it?
Lex Fridman (41:29.100)
How do we build it?
Lex Fridman (41:29.940)
Yes.
Lex Fridman (41:30.760)
How do we build a causal model
Lex Fridman (41:32.180)
without a team of human experts running around?
Lex Fridman (41:37.260)
Why don't you go to learning right away?
Lex Fridman (41:39.560)
You're too much involved with learning.
Judea Pearl (41:41.180)
Because I like babies.
Lex Fridman (41:42.100)
Babies learn fast.
Judea Pearl (41:43.140)
I'm trying to figure out how they do it.
Lex Fridman (41:45.140)
Good.
Lex Fridman (41:45.980)
So that's another question.
Lex Fridman (41:47.660)
How do the babies come out with a counterfactual
Lex Fridman (41:50.340)
model of the world?
Lex Fridman (41:51.760)
And babies do that.
Judea Pearl (41:53.540)
They know how to play in the crib.
Lex Fridman (41:56.860)
They know which balls hit another one.
Lex Fridman (41:59.500)
And they learn it by playful manipulation of the world.
Lex Fridman (42:06.820)
Yes.
Judea Pearl (42:07.660)
The simple world involves only toys and balls and chimes.
Lex Fridman (42:13.620)
But if you think about it, it's a complex world.
Judea Pearl (42:17.260)
We take for granted how complicated.
Lex Fridman (42:20.700)
And kids do it by playful manipulation
Judea Pearl (42:23.740)
plus parents guidance, peer wisdom, and hearsay.
Lex Fridman (42:30.860)
They meet each other and they say,
Judea Pearl (42:34.780)
you shouldn't have taken my toy.
Lex Fridman (42:38.940)
Right.
Lex Fridman (42:40.820)
And these multiple sources of information,
Lex Fridman (42:43.540)
they're able to integrate.
Lex Fridman (42:45.780)
So the challenge is about how to integrate,
Lex Fridman (42:49.260)
how to form these causal relationships
Judea Pearl (42:52.620)
from different sources of data.
Lex Fridman (42:54.260)
Correct.
Lex Fridman (42:55.600)
So how much information is it to play,
Lex Fridman (42:59.980)
how much causal information is required
Lex Fridman (43:03.060)
to be able to play in the crib with different objects?
Lex Fridman (43:06.900)
I don't know.
Judea Pearl (43:08.260)
I haven't experimented with the crib.
Lex Fridman (43:11.340)
Okay, not a crib.
Judea Pearl (43:12.700)
Picking up, manipulating physical objects
Lex Fridman (43:15.860)
on this very, opening the pages of a book,
Judea Pearl (43:19.800)
all the tasks, the physical manipulation tasks.
Lex Fridman (43:23.740)
Do you have a sense?
Judea Pearl (43:25.260)
Because my sense is the world is extremely complicated.
Lex Fridman (43:27.980)
It's extremely complicated.
Judea Pearl (43:29.420)
I agree, and I don't know how to organize it
Lex Fridman (43:31.260)
because I've been spoiled by easy problems
Lex Fridman (43:34.660)
such as cancer and death, okay?
Lex Fridman (43:37.240)
And I'm a, but she's a.
Judea Pearl (43:39.820)
First we have to start trying to.
Lex Fridman (43:41.060)
No, but it's easy.
Judea Pearl (43:42.660)
There is in a sense that you have only 20 variables.
Lex Fridman (43:46.980)
And they are just variables and not mechanics.
Judea Pearl (43:51.060)
Okay, it's easy.
Lex Fridman (43:52.300)
You just put them on the graph and they speak to you.
Judea Pearl (43:57.980)
Yeah, and you're providing a methodology
Lex Fridman (44:00.540)
for letting them speak.
Judea Pearl (44:02.380)
Yeah.
Lex Fridman (44:03.220)
I'm working only in the abstract.
Judea Pearl (44:05.140)
The abstract was knowledge in, knowledge out,
Lex Fridman (44:08.980)
data in between.
Judea Pearl (44:11.820)
Now, can we take a leap to trying to learn
Lex Fridman (44:15.100)
in this very, when it's not 20 variables,
Lex Fridman (44:18.140)
but 20 million variables, trying to learn causation
Lex Fridman (44:23.180)
in this world?
Judea Pearl (44:24.060)
Not learn, but somehow construct models.
Lex Fridman (44:27.180)
I mean, it seems like you would only have to be able
Judea Pearl (44:29.620)
to learn because constructing it manually
Lex Fridman (44:33.920)
would be too difficult.
Lex Fridman (44:35.580)
Do you have ideas of?
Lex Fridman (44:37.900)
I think it's a matter of combining simple models
Judea Pearl (44:41.220)
from many, many sources, from many, many disciplines,
Lex Fridman (44:45.460)
and many metaphors.
Judea Pearl (44:48.220)
Metaphors are the basics of human intelligence, basis.
Lex Fridman (44:51.940)
Yeah, so how do you think of about a metaphor
Lex Fridman (44:53.940)
in terms of its use in human intelligence?
Lex Fridman (44:56.180)
Metaphors is an expert system.
Judea Pearl (45:01.900)
An expert, it's mapping problem
Lex Fridman (45:05.700)
from a problem with which you are not familiar
Judea Pearl (45:09.580)
to a problem with which you are familiar.
Lex Fridman (45:13.800)
Like, I'll give you a good example.
Judea Pearl (45:15.980)
The Greek believed that the sky is an opaque shell.
Lex Fridman (45:22.240)
It's not really infinite space.
Judea Pearl (45:25.940)
It's an opaque shell, and the stars are holes
Lex Fridman (45:30.580)
poked in the shells through which you see
Judea Pearl (45:32.980)
the eternal light.
Lex Fridman (45:34.260)
That was a metaphor.
Lex Fridman (45:35.940)
Why?
Lex Fridman (45:36.960)
Because they understand how you poke holes in the shells.
Judea Pearl (45:42.800)
They were not familiar with infinite space.
Lex Fridman (45:47.260)
And we are walking on a shell of a turtle,
Lex Fridman (45:52.780)
and if you get too close to the edge,
Lex Fridman (45:54.520)
you're gonna fall down to Hades or wherever.
Judea Pearl (45:58.940)
That's a metaphor.
Lex Fridman (46:00.780)
It's not true.
Lex Fridman (46:02.180)
But this kind of metaphor enabled Aristoteles
Lex Fridman (46:07.220)
to measure the radius of the Earth,
Judea Pearl (46:10.680)
because he said, Kamal, if we are walking on a turtle shell,
Lex Fridman (46:15.140)
then the ray of light coming to this place
Judea Pearl (46:20.700)
will be a different angle than coming to this place.
Lex Fridman (46:23.040)
I know the distance, I'll measure the two angles,
Lex Fridman (46:26.420)
and then I have the radius of the shell of the turtle.
Lex Fridman (46:33.640)
And he did, and he found his measurement
Judea Pearl (46:38.460)
very close to the measurements we have today,
Lex Fridman (46:43.300)
through the, what, 6,700 kilometers of the Earth.
Judea Pearl (46:52.020)
That's something that would not occur
Lex Fridman (46:55.060)
to Babylonian astronomer,
Judea Pearl (46:59.760)
even though the Babylonian experiment
Lex Fridman (47:01.900)
were the machine learning people of the time.
Judea Pearl (47:04.600)
They fit curves, and they could predict
Lex Fridman (47:07.900)
the eclipse of the moon much more accurately
Judea Pearl (47:12.220)
than the Greek, because they fit curve.
Lex Fridman (47:17.160)
That's a different metaphor.
Judea Pearl (47:19.220)
Something that you're familiar with,
Lex Fridman (47:20.580)
a game, a turtle shell.
Lex Fridman (47:21.820)
Okay?
Lex Fridman (47:24.380)
What does it mean if you are familiar?
Judea Pearl (47:27.520)
Familiar means that answers to certain questions
Lex Fridman (47:31.900)
are explicit.
Judea Pearl (47:33.460)
You don't have to derive them.
Lex Fridman (47:35.580)
And they were made explicit because somewhere in the past
Judea Pearl (47:39.700)
you've constructed a model of that.
Lex Fridman (47:43.540)
Yeah, you're familiar with,
Lex Fridman (47:44.940)
so the child is familiar with billiard balls.
Lex Fridman (47:48.340)
So the child could predict that if you let loose
Judea Pearl (47:51.420)
of one ball, the other one will bounce off.
Lex Fridman (47:55.740)
You obtain that by familiarity.
Judea Pearl (48:00.180)
Familiarity is answering questions,
Lex Fridman (48:02.920)
and you store the answer explicitly.
Judea Pearl (48:05.880)
You don't have to derive them.
Lex Fridman (48:08.020)
So this is the idea of a metaphor.
Judea Pearl (48:09.620)
All our life, all our intelligence
Lex Fridman (48:11.620)
is built around metaphors,
Judea Pearl (48:13.420)
mapping from the unfamiliar to the familiar.
Lex Fridman (48:16.220)
But the marriage between the two is a tough thing,
Judea Pearl (48:20.540)
which we haven't yet been able to algorithmatize.
Lex Fridman (48:24.740)
So you think of that process of using metaphor
Judea Pearl (48:29.300)
to leap from one place to another,
Lex Fridman (48:31.900)
we can call it reasoning?
Lex Fridman (48:34.540)
Is it a kind of reasoning?
Lex Fridman (48:35.900)
It is reasoning by metaphor, metaphorical reasoning.
Lex Fridman (48:39.500)
Do you think of that as learning?
Lex Fridman (48:44.220)
So learning is a popular terminology today
Judea Pearl (48:46.500)
in a narrow sense.
Lex Fridman (48:47.620)
It is, it is, it is definitely a form.
Lex Fridman (48:49.780)
So you may not, okay, right.
Lex Fridman (48:51.620)
It's one of the most important learnings,
Judea Pearl (48:53.780)
taking something which theoretically is derivable
Lex Fridman (48:57.620)
and store it in accessible format.
Lex Fridman (49:01.700)
I'll give you an example, chess, okay?
Lex Fridman (49:07.780)
Finding the winning starting move in chess is hard.
Judea Pearl (49:12.780)
It is hard, but there is an answer.
Lex Fridman (49:21.260)
Either there is a winning move for white
Lex Fridman (49:23.780)
or there isn't, or there is a draw, okay?
Lex Fridman (49:26.820)
So it is, the answer to that
Judea Pearl (49:30.260)
is available through the rule of the games.
Lex Fridman (49:33.700)
But we don't know the answer.
Lex Fridman (49:35.460)
So what does a chess master have that we don't have?
Lex Fridman (49:38.500)
He has stored explicitly an evaluation
Judea Pearl (49:41.780)
of certain complex pattern of the board.
Lex Fridman (49:45.340)
We don't have it.
Judea Pearl (49:46.180)
Ordinary people like me, I don't know about you,
Lex Fridman (49:50.940)
I'm not a chess master.
Lex Fridman (49:52.820)
So for me, I have to derive things that for him is explicit.
Lex Fridman (49:58.620)
He has seen it before, or he has seen the pattern before,
Lex Fridman (50:02.500)
or similar pattern, you see metaphor, yeah?
Lex Fridman (50:05.100)
And he generalize and said, don't move, it's a dangerous move.
Judea Pearl (50:13.460)
It's just that not in the game of chess,
Lex Fridman (50:15.540)
but in the game of billiard balls,
Judea Pearl (50:18.940)
we humans are able to initially derive very effectively
Lex Fridman (50:22.420)
and then reason by metaphor very effectively
Lex Fridman (50:25.100)
and make it look so easy that it makes one wonder
Lex Fridman (50:28.700)
how hard is it to build it in a machine.
Lex Fridman (50:31.260)
So in your sense, how far away are we
Lex Fridman (50:38.260)
to be able to construct?
Judea Pearl (50:40.660)
I don't know, I'm not a futurist.
Lex Fridman (50:42.780)
All I can tell you is that we are making tremendous progress
Judea Pearl (50:48.380)
in the causal reasoning domain.
Lex Fridman (50:52.100)
Something that I even dare to call it revolution,
Judea Pearl (50:57.100)
the code of revolution, because what we have achieved
Lex Fridman (51:02.780)
in the past three decades is something that dwarf
Judea Pearl (51:09.660)
everything that was derived in the entire history.
Lex Fridman (51:13.660)
So there's an excitement about
Judea Pearl (51:15.860)
current machine learning methodologies,
Lex Fridman (51:18.700)
and there's really important good work you're doing
Judea Pearl (51:22.100)
in causal inference.
Lex Fridman (51:24.620)
Where does the future, where do these worlds collide
Lex Fridman (51:32.980)
and what does that look like?
Lex Fridman (51:35.060)
First, they're gonna work without collision.
Judea Pearl (51:37.780)
It's gonna work in harmony.
Lex Fridman (51:40.580)
Harmony, it's not collision.
Judea Pearl (51:41.740)
The human is going to jumpstart the exercise
Lex Fridman (51:46.740)
by providing qualitative, noncommitting models
Judea Pearl (51:55.060)
of how the universe works, how in reality
Lex Fridman (52:00.300)
the domain of discourse works.
Judea Pearl (52:03.780)
The machine is gonna take over from that point of view
Lex Fridman (52:06.820)
and derive whatever the calculus says can be derived.
Judea Pearl (52:11.820)
Namely, quantitative answer to our questions.
Lex Fridman (52:15.100)
Now, these are complex questions.
Judea Pearl (52:18.420)
I'll give you some example of complex questions
Lex Fridman (52:21.220)
that will bug your mind if you think about it.
Judea Pearl (52:27.580)
You take result of studies in diverse population
Lex Fridman (52:33.660)
under diverse condition, and you infer the cause effect
Judea Pearl (52:38.660)
of a new population which doesn't even resemble
Lex Fridman (52:43.140)
any of the ones studied, and you do that by do calculus.
Judea Pearl (52:48.300)
You do that by generalizing from one study to another.
Lex Fridman (52:52.660)
See, what's common with Berto?
Lex Fridman (52:54.940)
What is different?
Lex Fridman (52:57.020)
Let's ignore the differences and pull out the commonality,
Lex Fridman (53:01.180)
and you do it over maybe 100 hospitals around the world.
Lex Fridman (53:06.140)
From that, you can get really mileage from big data.
Judea Pearl (53:11.140)
It's not only that you have many samples,
Lex Fridman (53:15.060)
you have many sources of data.
Lex Fridman (53:18.740)
So that's a really powerful thing, I think,
Lex Fridman (53:21.340)
especially for medical applications.
Lex Fridman (53:23.420)
I mean, cure cancer, right?
Lex Fridman (53:25.860)
That's how from data you can cure cancer.
Lex Fridman (53:28.580)
So we're talking about causation,
Lex Fridman (53:30.100)
which is the temporal relationships between things.
Judea Pearl (53:35.180)
Not only temporal, it's both structural and temporal.
Lex Fridman (53:38.700)
Temporal enough, temporal precedence by itself
Judea Pearl (53:43.020)
cannot replace causation.
Lex Fridman (53:46.820)
Is temporal precedence the arrow of time in physics?
Judea Pearl (53:50.900)
It's important, necessary.
Lex Fridman (53:52.220)
It's important.
Judea Pearl (53:53.060)
It's efficient, yes.
Lex Fridman (53:54.260)
Is it?
Judea Pearl (53:55.780)
Yes, I never seen cause propagate backward.
Lex Fridman (54:00.220)
But if we use the word cause,
Lex Fridman (54:03.940)
but there's relationships that are timeless.
Lex Fridman (54:07.060)
I suppose that's still forward in the arrow of time.
Lex Fridman (54:10.380)
But are there relationships, logical relationships,
Lex Fridman (54:14.860)
that fit into the structure?
Judea Pearl (54:18.580)
Sure, the whole do calculus is logical relationship.
Lex Fridman (54:21.980)
That doesn't require a temporal.
Judea Pearl (54:23.780)
It has just the condition that
Lex Fridman (54:26.940)
you're not traveling back in time.
Judea Pearl (54:28.580)
Yes, correct.
Lex Fridman (54:31.180)
So it's really a generalization of,
Lex Fridman (54:34.020)
a powerful generalization of what?
Lex Fridman (54:39.740)
Of Boolean logic.
Judea Pearl (54:40.700)
Yeah, Boolean logic.
Lex Fridman (54:41.700)
Yes.
Judea Pearl (54:43.460)
That is sort of simply put,
Lex Fridman (54:46.260)
and allows us to reason about the order of events,
Judea Pearl (54:53.620)
the source, the.
Lex Fridman (54:54.460)
Not about, between, we're not deriving the order of events.
Lex Fridman (54:58.020)
We are given cause effects relationship, okay?
Lex Fridman (55:01.340)
They ought to be obeying the time presidents relationship.
Judea Pearl (55:08.900)
We are given it.
Lex Fridman (55:09.940)
And now that we ask questions about
Judea Pearl (55:12.500)
other causes of relationship,
Lex Fridman (55:14.260)
that could be derived from the initial ones,
Lex Fridman (55:17.860)
but were not given to us explicitly.
Lex Fridman (55:20.940)
Like the case of the firing squad I gave you
Judea Pearl (55:26.260)
in the first chapter.
Lex Fridman (55:28.220)
And I ask, what if rifleman A declined to shoot?
Lex Fridman (55:33.860)
Would the prisoner still be dead?
Lex Fridman (55:37.940)
To decline to shoot, it means that he disobey order.
Lex Fridman (55:42.020)
And the rule of the games were that he is a
Lex Fridman (55:48.180)
obedient marksman, okay?
Judea Pearl (55:51.820)
That's how you start.
Lex Fridman (55:52.660)
That's the initial order.
Lex Fridman (55:53.500)
But now you ask question about breaking the rules.
Lex Fridman (55:56.540)
What if he decided not to pull the trigger?
Judea Pearl (56:00.420)
He just became a pacifist.
Lex Fridman (56:03.620)
And you and I can answer that.
Lex Fridman (56:06.020)
The other rifleman would have killed him, okay?
Lex Fridman (56:09.220)
I want the machine to do that.
Lex Fridman (56:12.300)
Is it so hard to ask a machine to do that?
Lex Fridman (56:15.420)
It's such a simple task.
Judea Pearl (56:16.740)
You have to have a calculus for that.
Lex Fridman (56:19.380)
Yes, yeah.
Lex Fridman (56:21.180)
But the curiosity, the natural curiosity for me is
Lex Fridman (56:24.300)
that yes, you're absolutely correct and important.
Lex Fridman (56:27.980)
And it's hard to believe that we haven't done this
Lex Fridman (56:31.060)
seriously extensively already a long time ago.
Lex Fridman (56:35.380)
So this is really important work.
Lex Fridman (56:37.020)
But I also wanna know, maybe you can philosophize
Judea Pearl (56:41.340)
about how hard is it to learn.
Lex Fridman (56:43.340)
Okay, let's assume we're learning.
Lex Fridman (56:44.420)
We wanna learn it, okay?
Lex Fridman (56:45.620)
We wanna learn.
Lex Fridman (56:46.460)
So what do we do?
Lex Fridman (56:47.280)
We put a learning machine that watches execution trials
Lex Fridman (56:51.980)
in many countries and many locations, okay?
Lex Fridman (56:57.220)
All the machine can learn is to see shut or not shut.
Judea Pearl (57:01.040)
Dead, not dead.
Lex Fridman (57:02.920)
A court issued an order or didn't, okay?
Judea Pearl (57:05.620)
Just the facts.
Lex Fridman (57:07.340)
For the fact you don't know who listens to whom.
Judea Pearl (57:10.140)
You don't know that the condemned person
Lex Fridman (57:13.740)
listened to the bullets,
Lex Fridman (57:15.280)
that the bullets are listening to the captain, okay?
Lex Fridman (57:19.300)
All we hear is one command, two shots, dead, okay?
Judea Pearl (57:24.300)
A triple of variable.
Lex Fridman (57:27.100)
Yes, no, yes, no.
Judea Pearl (57:29.300)
Okay, that you can learn who listens to whom
Lex Fridman (57:32.140)
and you can answer the question, no.
Judea Pearl (57:33.980)
Definitively no.
Lex Fridman (57:35.220)
But don't you think you can start proposing ideas
Lex Fridman (57:39.120)
for humans to review?
Lex Fridman (57:41.660)
You want machine to learn, you want a robot.
Lex Fridman (57:44.380)
So robot is watching trials like that, 200 trials,
Lex Fridman (57:50.780)
and then he has to answer the question,
Lex Fridman (57:52.580)
what if rifleman A refrain from shooting?
Lex Fridman (57:56.920)
Yeah.
Lex Fridman (57:58.660)
How do I do that?
Lex Fridman (58:00.260)
That's exactly my point.
Judea Pearl (58:03.620)
It's looking at the facts,
Lex Fridman (58:04.900)
don't give you the strings behind the facts.
Judea Pearl (58:07.140)
Absolutely, but do you think of machine learning
Lex Fridman (58:11.860)
as it's currently defined as only something
Lex Fridman (58:15.220)
that looks at the facts and tries to do?
Lex Fridman (58:17.620)
Right now, they only look at the facts, yeah.
Lex Fridman (58:19.180)
So is there a way to modify, in your sense?
Lex Fridman (58:23.500)
Playful manipulation.
Judea Pearl (58:25.140)
Playful manipulation.
Lex Fridman (58:26.140)
Yes, once in a while.
Judea Pearl (58:26.980)
Doing the interventionist kind of thing, intervention.
Lex Fridman (58:29.700)
But it could be at random.
Judea Pearl (58:31.140)
For instance, the rifleman is sick that day
Lex Fridman (58:34.540)
or he just vomits or whatever.
Lex Fridman (58:37.260)
So machine can observe this unexpected event
Lex Fridman (58:41.240)
which introduce noise.
Judea Pearl (58:43.580)
The noise still have to be random
Lex Fridman (58:46.420)
to be able to relate it to randomized experiment.
Lex Fridman (58:51.660)
And then you have observational studies
Lex Fridman (58:55.460)
from which to infer the strings behind the facts.
Judea Pearl (58:59.940)
It's doable to a certain extent.
Lex Fridman (59:02.980)
But now that we are expert in what you can do
Judea Pearl (59:06.260)
once you have a model, we can reason back and say,
Lex Fridman (59:09.080)
what kind of data you need to build a model.
Judea Pearl (59:13.020)
Got it, so I know you're not a futurist,
Lex Fridman (59:17.300)
but are you excited?
Judea Pearl (59:19.760)
Have you, when you look back at your life,
Lex Fridman (59:22.500)
longed for the idea of creating
Lex Fridman (59:24.540)
a human level intelligence system?
Lex Fridman (59:25.940)
Yeah, I'm driven by that.
Judea Pearl (59:28.340)
All my life, I'm driven just by one thing.
Lex Fridman (59:33.780)
But I go slowly.
Judea Pearl (59:34.900)
I go from what I know to the next step incrementally.
Lex Fridman (59:39.380)
So without imagining what the end goal looks like.
Lex Fridman (59:42.440)
Do you imagine what an eight?
Lex Fridman (59:44.860)
The end goal is gonna be a machine
Judea Pearl (59:47.780)
that can answer sophisticated questions,
Lex Fridman (59:50.900)
counterfactuals of regret, compassion,
Judea Pearl (59:55.500)
responsibility, and free will.
Lex Fridman (59:59.380)
So what is a good test?
🔗 相关节目