back to indexJudea Pearl: Causal Reasoning, Counterfactuals, and the Path to AGI | Lex Fridman Podcast #56
link |
The following is a conversation with Judea Pearl,
link |
professor at UCLA and a winner of the Turing Award
link |
that's generally recognized as the Nobel Prize of Computing.
link |
He's one of the seminal figures
link |
in the field of artificial intelligence,
link |
computer science, and statistics.
link |
He has developed and championed probabilistic approaches
link |
to AI, including Beijing networks,
link |
and profound ideas in causality in general.
link |
These ideas are important not just to AI,
link |
but to our understanding and practice of science.
link |
But in the field of AI, the idea of causality,
link |
cause and effect, to many, lie at the core
link |
of what is currently missing and what must be developed
link |
in order to build truly intelligent systems.
link |
For this reason, and many others,
link |
his work is worth returning to often.
link |
I recommend his most recent book called Book of Why
link |
that presents key ideas from a lifetime of work
link |
in a way that is accessible to the general public.
link |
This is the Artificial Intelligence Podcast.
link |
If you enjoy it, subscribe on YouTube,
link |
give it five stars on Apple Podcast,
link |
support it on Patreon, or simply connect with me on Twitter
link |
at Lex Friedman, spelled F R I D M A N.
link |
If you leave a review on Apple Podcasts especially,
link |
but also a cast box or comment on YouTube,
link |
consider mentioning topics, people, ideas, questions,
link |
quotes, and science, tech, and philosophy
link |
you find interesting, and I'll read them on this podcast.
link |
I won't call out names, but I love comments
link |
with kindness and thoughtfulness in them,
link |
so I thought I'd share them with you.
link |
Someone on YouTube highlighted a quote
link |
from the conversation with Noam Chomsky,
link |
where he said that the significance of your life
link |
is something you create.
link |
I like this line as well.
link |
On most days, the existentialist approach to life
link |
is one I find liberating and fulfilling.
link |
I recently started doing ads
link |
at the end of the introduction.
link |
I'll do one or two minutes after introducing the episode,
link |
and never any ads in the middle
link |
that break the flow of the conversation.
link |
I hope that works for you
link |
and doesn't hurt the listening experience.
link |
This show is presented by Cash App,
link |
the number one finance app in the App Store.
link |
I personally use Cash App to send money to friends,
link |
but you can also use it to buy, sell,
link |
and deposit Bitcoin in just seconds.
link |
Cash App also has a new investing feature.
link |
You can buy fractions of a stock, say $1 worth,
link |
no matter what the stock price is.
link |
Broker services are provided by Cash App Investing,
link |
a subsidiary of Square, a member of SIPC.
link |
I'm excited to be working with Cash App
link |
to support one of my favorite organizations called First,
link |
best known for their first robotics and Lego competitions.
link |
They educate and inspire hundreds of thousands of students
link |
in over 110 countries,
link |
and have a perfect rating on Charity Navigator,
link |
which means the donated money
link |
is used to the maximum effectiveness.
link |
When you get Cash App from the App Store or Google Play,
link |
and use the code LEXPODCAST, you'll get $10,
link |
and Cash App will also donate $10 to First,
link |
which again, is an organization
link |
that I've personally seen inspire girls and boys
link |
to dream of engineering a better world.
link |
And now, here's my conversation with Judea Pearl.
link |
You mentioned in an interview
link |
that science is not a collection of facts,
link |
but a constant human struggle
link |
with the mysteries of nature.
link |
What was the first mystery that you can recall
link |
that hooked you, that kept you in the creaset?
link |
Oh, the first mystery, that's a good one.
link |
Yeah, I remember that.
link |
I had a fever for three days.
link |
And when I learned about Descartes, analytic geometry,
link |
and I found out that you can do all the construction
link |
in geometry using algebra.
link |
And I couldn't get over it.
link |
I simply couldn't get out of bed.
link |
So what kind of world does analytic geometry unlock?
link |
Well, it connects algebra with geometry.
link |
Okay, so Descartes had the idea
link |
that geometrical construction and geometrical theorems
link |
and assumptions can be articulated
link |
in the language of algebra,
link |
which means that all the proof that we did in high school,
link |
and trying to prove that the three bisectors
link |
meet at one point, and that, okay,
link |
all this can be proven by just shuffling around notation.
link |
Yeah, that was a traumatic experience.
link |
That was a traumatic experience.
link |
For me, it was, I'm telling you.
link |
So it's the connection
link |
between the different mathematical disciplines,
link |
Not in between two different languages.
link |
So which mathematic discipline is most beautiful?
link |
Is geometry it for you?
link |
Both are beautiful.
link |
They have almost the same power.
link |
But there's a visual element to geometry, being a.
link |
Visually, it's more transparent.
link |
But once you get over to algebra,
link |
then the linear equation is a straight line.
link |
This translation is easily absorbed, okay?
link |
And to pass a tangent to a circle,
link |
you know, you have the basic theorems,
link |
and you can do it with algebra.
link |
So but the transition from one to another was really,
link |
I thought that Descartes was the greatest mathematician
link |
So you have been at the, if you think of engineering
link |
and mathematics as a spectrum.
link |
You have been, you have walked casually along this spectrum
link |
throughout your life.
link |
You know, a little bit of engineering,
link |
and then, you know,
link |
you've done a little bit of mathematics here and there.
link |
I mean, we got a very solid background in mathematics,
link |
because our teachers were geniuses.
link |
Our teachers came from Germany in the 1930s,
link |
running away from Hitler.
link |
They left their careers in Heidelberg and Berlin,
link |
and came to teach high school in Israel.
link |
And we were the beneficiary of that experiment.
link |
So I, and they taught us math the good way.
link |
What's the good way to teach math?
link |
The people behind the theorems, yeah.
link |
Their cousins, and their nieces, and their faces.
link |
And how they jumped from the bathtub when they scream,
link |
And ran naked in town.
link |
So you're almost educated as a historian of math.
link |
No, we just got a glimpse of that history
link |
together with a theorem,
link |
so every exercise in math was connected with a person.
link |
And the time of the person.
link |
The period, also mathematically speaking.
link |
Mathematically speaking, yes.
link |
Not the politics, no.
link |
So, and then in university,
link |
you have gone on to do engineering.
link |
I get a B.S. in engineering and a technical, right?
link |
And then I moved here for graduate work,
link |
and I got, I did engineering in addition to physics
link |
in Rutgers, and it combined very nicely with my thesis,
link |
which I did in RCA Laboratories in superconductivity.
link |
And then somehow thought to switch
link |
to almost computer science, software,
link |
even, not switch, but long to become,
link |
to get into software engineering a little bit.
link |
And programming, if you can call it that in the 70s.
link |
So there's all these disciplines.
link |
So to pick a favorite, in terms of engineering
link |
and mathematics, which path do you think has more beauty?
link |
Which path has more power?
link |
It's hard to choose, no.
link |
I enjoy doing physics, and even have a vortex
link |
So I have investment in immortality.
link |
So what is a vortex?
link |
Vortex is in superconductivity.
link |
In the superconductivity, yeah.
link |
You have permanent current swirling around.
link |
One way or the other, you can have a store one or zero
link |
That's what we worked on in the 1960 in RCA.
link |
And I discovered a few nice phenomena with the vortices.
link |
You push current and they move.
link |
So that's a pearl vortex.
link |
Pearl vortex, right, you can Google it, right?
link |
I didn't know about it, but the physicists,
link |
they picked up on my thesis, on my PhD thesis,
link |
and it becomes popular when thin film superconductors
link |
became important for high temperature superconductors.
link |
So they called it pearl vortex without my knowledge.
link |
I discovered it only about 15 years ago.
link |
You have footprints in all of the sciences.
link |
So let's talk about the universe a little bit.
link |
Is the universe at the lowest level deterministic
link |
or stochastic in your amateur philosophy view?
link |
Put another way, does God play dice?
link |
We know it is stochastic, right?
link |
Today, today we think it is stochastic.
link |
We think because we have the Heisenberg uncertainty principle
link |
and we have some experiments to confirm that.
link |
All we have is experiments to confirm it.
link |
We don't understand why.
link |
You wrote a book about why.
link |
Yeah, it's a puzzle.
link |
It's a puzzle that you have the dice flipping machine,
link |
oh God, and the result of the flipping
link |
propagate with the speed faster than the speed of light.
link |
We can't explain it, okay?
link |
So, but it only governs microscopic phenomena.
link |
Microscopic phenomena.
link |
So you don't think of quantum mechanics
link |
as useful for understanding the nature of reality?
link |
So in your thinking, the world might
link |
as well be deterministic.
link |
The world is deterministic,
link |
and as far as the neuron firing is concerned,
link |
it is deterministic to first approximation.
link |
What about free will?
link |
Free will is also a nice exercise.
link |
Free will is an illusion that we AI people are gonna solve.
link |
So what do you think once we solve it,
link |
that solution will look like?
link |
Once we put it in the page.
link |
The solution will look like,
link |
first of all, it will look like a machine.
link |
A machine that act as though it has free will.
link |
It communicates with other machines
link |
as though they have free will,
link |
and you wouldn't be able to tell the difference
link |
between a machine that does
link |
and a machine that doesn't have free will, okay?
link |
So the illusion, it propagates the illusion
link |
of free will amongst the other machines.
link |
And faking it is having it, okay?
link |
That's what Turing test is all about.
link |
Faking intelligence is intelligent
link |
because it's not easy to fake.
link |
It's very hard to fake,
link |
and you can only fake if you have it.
link |
So that's such a beautiful statement.
link |
Yeah, you can't fake it if you don't have it, yeah.
link |
So let's begin at the beginning with probability,
link |
both philosophically and mathematically.
link |
What does it mean to say the probability
link |
of something happening is 50%?
link |
What is probability?
link |
It's a degree of uncertainty
link |
that an agent has about the world.
link |
You're still expressing some knowledge in that statement.
link |
If the probability is 90%,
link |
it's absolutely a different kind of knowledge
link |
than if it is 10%.
link |
But it's still not solid knowledge, it's...
link |
It is solid knowledge, but hey,
link |
if you tell me that 90% assurance smoking
link |
will give you lung cancer in five years versus 10%,
link |
it's a piece of useful knowledge.
link |
So the statistical view of the universe,
link |
So we're swimming in complete uncertainty,
link |
most of everything around us.
link |
It allows you to predict things with a certain probability,
link |
and computing those probabilities are very useful.
link |
That's the whole idea of prediction.
link |
And you need prediction to be able to survive.
link |
If you can't predict the future,
link |
then you're just crossing the street,
link |
will be extremely fearful.
link |
And so you've done a lot of work in causation,
link |
and so let's think about correlation.
link |
I started with probability.
link |
You started with probability.
link |
You've invented the Bayesian networks.
link |
And so we'll dance back and forth
link |
between these levels of uncertainty.
link |
But what is correlation?
link |
So probability of something happening is something,
link |
but then there's a bunch of things happening.
link |
And sometimes they happen together,
link |
sometimes not, they're independent or not.
link |
So how do you think about correlation of things?
link |
Correlation occurs when two things vary together
link |
over a very long time is one way of measuring it.
link |
Or when you have a bunch of variables
link |
that they all vary cohesively,
link |
then because we have a correlation here.
link |
And usually when we think about correlation,
link |
we really think causally.
link |
Things cannot be correlated unless there is a reason
link |
for them to vary together.
link |
Why should they vary together?
link |
If they don't see each other, why should they vary together?
link |
So underlying it somewhere is causation.
link |
Hidden in our intuition, there is a notion of causation
link |
because we cannot grasp any other logic except causation.
link |
And how does conditional probability differ from causation?
link |
So what is conditional probability?
link |
Conditional probability, how things vary
link |
when one of them stays the same.
link |
Now staying the same means that I have chosen
link |
to look only at those incidents
link |
where the guy has the same value as previous one.
link |
It's my choice as an experimenter.
link |
So things that are not correlated before
link |
could become correlated.
link |
Like for instance, if I have two coins
link |
which are uncorrelated, okay,
link |
and I choose only those flippings, experiments
link |
in which a bell rings, and the bell rings
link |
when at least one of them is a tail, okay,
link |
then suddenly I see correlation between the two coins
link |
because I only look at the cases where the bell rang.
link |
You see, it's my design, with my ignorance essentially,
link |
with my audacity to ignore certain incidents,
link |
I suddenly create a correlation
link |
where it doesn't exist physically.
link |
Right, so that's, you just outlined one of the flaws
link |
of observing the world and trying to infer something
link |
from the math about the world
link |
from looking at the correlation.
link |
I don't look at it as a flaw, the world works like that.
link |
But the flaws comes if we try to impose
link |
causal logic on correlation, it doesn't work too well.
link |
I mean, but that's exactly what we do.
link |
That's what, that has been the majority of science.
link |
The majority of naive science, the decisions know it.
link |
The decisions know that if you condition
link |
on a third variable, then you can destroy
link |
or create correlations among two other variables.
link |
They know it, it's in their data.
link |
It's nothing surprising, that's why they all dismiss
link |
the Simpson Paradox, ah, we know it.
link |
They don't know anything about it.
link |
Well, there's disciplines like psychology
link |
where all the variables are hard to account for.
link |
And so oftentimes there's a leap
link |
between correlation to causation.
link |
You're implying a leap.
link |
Who is trying to get causation from correlation?
link |
You're not proving causation,
link |
but you're sort of discussing it,
link |
implying, sort of hypothesizing with our ability to prove.
link |
Which discipline you have in mind?
link |
I'll tell you if they are obsolete,
link |
or if they are outdated, or they are about to get outdated.
link |
Tell me which one you have in mind.
link |
Oh, psychology, you know.
link |
Psychology, what, is it SEM, structural equation?
link |
No, no, I was thinking of applied psychology studying.
link |
For example, we work with human behavior
link |
in semi autonomous vehicles, how people behave.
link |
And you have to conduct these studies
link |
of people driving cars.
link |
Everything starts with the question.
link |
What is the research question?
link |
What is the research question?
link |
The research question, do people fall asleep
link |
when the car is driving itself?
link |
Do they fall asleep, or do they tend to fall asleep
link |
Than the car not driving itself.
link |
Not driving itself.
link |
That's a good question, okay.
link |
And so you measure, you put people in the car
link |
because it's real world.
link |
You can't conduct an experiment
link |
where you control everything.
link |
Why can't you control?
link |
Why can't you control the automatic module on and off?
link |
Because it's on road, public.
link |
I mean, there's aspects to it that's unethical.
link |
Because it's testing on public roads.
link |
So you can only use vehicle.
link |
They have to, the people, the drivers themselves
link |
have to make that choice themselves.
link |
And so they regulate that.
link |
And so you just observe when they drive autonomously
link |
and when they don't.
link |
But maybe they turn it off when they were very tired.
link |
Yeah, that kind of thing.
link |
But you don't know those variables.
link |
Okay, so that you have now uncontrolled experiment.
link |
Uncontrolled experiment.
link |
We call it observational study.
link |
And we form the correlation detected.
link |
We have to infer causal relationship.
link |
Whether it was the automatic piece
link |
has caused them to fall asleep, or.
link |
So that is an issue that is about 120 years old.
link |
I should only go 100 years old, okay.
link |
Well, maybe it's not.
link |
Actually I should say it's 2,000 years old.
link |
Because we have this experiment by Daniel.
link |
But the Babylonian king that wanted the exiled,
link |
the people from Israel that were taken in exile
link |
to Babylon to serve the king.
link |
He wanted to serve them king's food, which was meat.
link |
And Daniel as a good Jew couldn't eat non kosher food.
link |
So he asked them to eat vegetarian food.
link |
But the king overseer says, I'm sorry,
link |
but if the king sees that your performance falls below
link |
that of other kids, he's going to kill me.
link |
Daniel said, let's make an experiment.
link |
Let's take four of us from Jerusalem, okay.
link |
Give us vegetarian food.
link |
Let's take the other guys to eat the king's food.
link |
And in about a week's time, we'll test our performance.
link |
And you know the answer.
link |
Of course he did the experiment.
link |
And they were so much better than the others.
link |
And the kings nominated them to super position in his king.
link |
So it was a first experiment, yes.
link |
So there was a very simple,
link |
it's also the same research questions.
link |
We want to know if vegetarian food
link |
assist or obstruct your mental ability.
link |
And okay, so the question is very old one.
link |
Even Democritus said, if I could discover one cause
link |
of things, I would rather discover one cause
link |
and be a king of Persia, okay.
link |
The task of discovering causes was in the mind
link |
of ancient people from many, many years ago.
link |
But the mathematics of doing that was only developed
link |
So science has left us orphan, okay.
link |
Science has not provided us with the mathematics
link |
to capture the idea of X causes Y
link |
and Y does not cause X.
link |
Because all the question of physics are symmetrical,
link |
algebraic, the equality sign goes both ways.
link |
Okay, let's look at machine learning.
link |
Machine learning today, if you look at deep neural networks,
link |
you can think of it as kind of conditional probability
link |
Correct, beautiful.
link |
So where did you say that?
link |
Conditional probability estimators.
link |
None of the machine learning people clevered you?
link |
Listen, most people, and this is why today's conversation
link |
I think is interesting, is most people would agree with you.
link |
There's certain aspects that are just effective today,
link |
but we're going to hit a wall and there's a lot of ideas.
link |
I think you're very right that we're gonna have to return to
link |
about causality and it would be, let's try to explore it.
link |
Let's even take a step back.
link |
You've invented Bayesian networks that look awfully a lot
link |
like they express something like causation,
link |
but they don't, not necessarily.
link |
So how do we turn Bayesian networks
link |
into expressing causation?
link |
How do we build causal networks?
link |
This A causes B, B causes C,
link |
how do we start to infer that kind of thing?
link |
We start asking ourselves question,
link |
what are the factors that would determine
link |
X could be blood pressure, death, hunger.
link |
But these are hypotheses that we propose for ourselves.
link |
Hypothesis, everything which has to do with causality
link |
comes from a theory.
link |
The difference is only how you interrogate
link |
the theory that you have in your mind.
link |
So it still needs the human expert to propose.
link |
You need the human expert to specify the initial model.
link |
Initial model could be very qualitative.
link |
Just who listens to whom?
link |
By whom listen to, I mean one variable listen to the other.
link |
So I say, okay, the tide is listening to the moon.
link |
And not to the rooster crow.
link |
This is our understanding of the world in which we live.
link |
Scientific understanding of reality.
link |
We have to start there.
link |
Because if we don't know how to handle
link |
cause and effect relationship,
link |
when we do have a model,
link |
and we certainly do not know how to handle it
link |
when we don't have a model.
link |
So let's start first.
link |
In AI, slogan is representation first, discovery second.
link |
But if I give you all the information that you need,
link |
can you do anything useful with it?
link |
That is the first, representation.
link |
How do you represent it?
link |
I give you all the knowledge in the world.
link |
How do you represent it?
link |
When you represent it, I ask you,
link |
can you infer X or Y or Z?
link |
Can you answer certain queries?
link |
All the computer science exercises we do,
link |
once you give me a representation for my knowledge,
link |
then you can ask me,
link |
now I understand how to represent things.
link |
How do I discover them?
link |
It's a secondary thing.
link |
So first of all, I should echo the statement
link |
that mathematics and the current,
link |
much of the machine learning world has not considered
link |
causation that A causes B, just in anything.
link |
So that seems like a non obvious thing
link |
that you think we would have really acknowledged it,
link |
So we have to put that on the table.
link |
So knowledge, how hard is it to create a knowledge
link |
from which to work?
link |
In certain area, it's easy,
link |
because we have only four or five major variables,
link |
and an epidemiologist or an economist can put them down,
link |
what, minimum wage, unemployment, policy, X, Y, Z,
link |
and start collecting data,
link |
and quantify the parameters that were left unquantified
link |
with the initial knowledge.
link |
That's the routine work that you find
link |
in experimental psychology, in economics, everywhere.
link |
In the health science, that's a routine thing.
link |
But I should emphasize,
link |
you should start with the research question.
link |
What do you want to estimate?
link |
Once you have that, you have to have a language
link |
of expressing what you want to estimate.
link |
You think it's easy?
link |
So we can talk about two things, I think.
link |
One is how the science of causation is very useful
link |
for answering certain questions.
link |
And then the other is, how do we create intelligent systems
link |
that need to reason with causation?
link |
So if my research question is,
link |
how do I pick up this water bottle from the table?
link |
All of the knowledge that is required
link |
to be able to do that,
link |
how do we construct that knowledge base?
link |
Do we return back to the problem
link |
that we didn't solve in the 80s with expert systems?
link |
Do we have to solve that problem
link |
of automated construction of knowledge?
link |
You're talking about the task
link |
of eliciting knowledge from an expert.
link |
Task of eliciting knowledge from an expert,
link |
or the self discovery of more knowledge,
link |
more and more knowledge.
link |
So automating the building of knowledge as much as possible.
link |
It's a different game in the causal domain,
link |
because it's essentially the same thing.
link |
You have to start with some knowledge,
link |
and you're trying to enrich it.
link |
But you don't enrich it by asking for more rules.
link |
You enrich it by asking for the data,
link |
to look at the data and quantifying,
link |
and ask queries that you couldn't answer when you started.
link |
You couldn't because the question is quite complex,
link |
and it's not within the capability
link |
of ordinary cognition, of ordinary person,
link |
or ordinary expert even, to answer.
link |
So what kind of questions do you think
link |
we can start to answer?
link |
Even a simple one.
link |
Suppose, yeah, I'll start with easy one.
link |
Okay, what's the effect of a drug on recovery?
link |
What is the aspirin that caused my headache
link |
to be cured, or what did the television program,
link |
or the good news I received?
link |
This is already, you see, it's a difficult question,
link |
because it's find the cause from effect.
link |
The easy one is find the effect from cause.
link |
So first you construct a model,
link |
saying that this is an important research question.
link |
This is an important question.
link |
I didn't construct a model yet.
link |
I just said it's an important question.
link |
And the first exercise is express it mathematically.
link |
What do you want to do?
link |
Like, if I tell you what will be the effect
link |
of taking this drug, you have to say that in mathematics.
link |
How do you say that?
link |
Can you write down the question, not the answer?
link |
I want to find the effect of the drug on my headache.
link |
Write down, write it down.
link |
That's where the do calculus comes in.
link |
The do operator, what is the do operator?
link |
Do operator, yeah.
link |
It's the difference in association and intervention.
link |
Very beautifully sort of constructed.
link |
Yeah, so we have a do operator.
link |
So the do calculus connected on the do operator itself
link |
connects the operation of doing
link |
to something that we can see.
link |
So as opposed to the purely observing,
link |
you're making the choice to change a variable.
link |
That's what it expresses.
link |
And then the way that we interpret it,
link |
the mechanism by which we take your query
link |
and we translate it into something that we can work with
link |
is by giving it semantics,
link |
saying that you have a model of the world
link |
and you cut off all the incoming error into X
link |
and you're looking now in the modified mutilated model,
link |
you ask for the probability of Y.
link |
That is interpretation of doing X
link |
because by doing things, you've liberated them
link |
from all influences that acted upon them earlier
link |
and you subject them to the tyranny of your muscles.
link |
So you remove all the questions about causality
link |
There's one level of questions.
link |
Answer questions about what will happen if you do things.
link |
If you do, if you drink the coffee,
link |
if you take the aspirin.
link |
So how do we get the doing data?
link |
Now the question is, if we cannot run experiments,
link |
then we have to rely on observational study.
link |
So first we could, sorry to interrupt,
link |
we could run an experiment where we do something,
link |
where we drink the coffee and this,
link |
the do operator allows you to sort of be systematic
link |
So imagine how the experiment will look like
link |
even though we cannot physically
link |
and technologically conduct it.
link |
I'll give you an example.
link |
What is the effect of blood pressure on mortality?
link |
I cannot go down into your vein
link |
and change your blood pressure,
link |
but I can ask the question,
link |
which means I can even have a model of your body.
link |
I can imagine the effect of your,
link |
how the blood pressure change will affect your mortality.
link |
I go into the model and I conduct this surgery
link |
about the blood pressure,
link |
even though physically I can do, I cannot do it.
link |
Let me ask the quantum mechanics question.
link |
Does the doing change the observation?
link |
Meaning the surgery of changing the blood pressure is,
link |
No, the surgery is,
link |
I call the very delicate.
link |
It's very delicate, infinitely delicate.
link |
Incisive and delicate, which means,
link |
do means, do X means,
link |
I'm gonna touch only X.
link |
So that means that I change only things
link |
which depends on X by virtue of X changing,
link |
but I don't depend things which are not depend on X.
link |
Like I wouldn't change your sex or your age,
link |
I just change your blood pressure.
link |
So in the case of blood pressure,
link |
it may be difficult or impossible
link |
to construct such an experiment.
link |
No, physically yes, but hypothetically no.
link |
Hypothetically no.
link |
If we have a model, that is what the model is for.
link |
So you conduct surgeries on a model,
link |
you take it apart, put it back,
link |
that's the idea of a model.
link |
It's the idea of thinking counterfactually, imagining,
link |
and that's the idea of creativity.
link |
So by constructing that model,
link |
you can start to infer if the higher the blood pressure
link |
leads to mortality, which increases or decreases by.
link |
I construct the model, I still cannot answer it.
link |
I have to see if I have enough information in the model
link |
that would allow me to find out the effects of intervention
link |
from a noninterventional study, hence of study.
link |
You need to have assumptions about who affects whom.
link |
If the graph had a certain property,
link |
the answer is yes, you can get it from observational study.
link |
If the graph is too mushy, bushy, bushy,
link |
the answer is no, you cannot.
link |
Then you need to find either different kind of observation
link |
that you haven't considered, or one experiment.
link |
So basically, that puts a lot of pressure on you
link |
to encode wisdom into that graph.
link |
But you don't have to encode more than what you know.
link |
God forbid, if you put the,
link |
like economists are doing this,
link |
they call identifying assumption.
link |
They put assumptions, even if they don't prevail
link |
in the world, they put assumptions
link |
so they can identify things.
link |
But the problem is, yes, beautifully put,
link |
but the problem is you don't know what you don't know.
link |
You know what you don't know.
link |
Because if you don't know, you say it's possible.
link |
It's possible that X affect the traffic tomorrow.
link |
You put down an arrow which says it's possible.
link |
Every arrow in the graph says it's possible.
link |
So there's not a significant cost to adding arrows that.
link |
The more arrow you add, the less likely you are
link |
to identify things from purely observational data.
link |
So if the whole world is bushy,
link |
and everybody affect everybody else,
link |
the answer is, you can answer it ahead of time.
link |
I cannot answer my query from observational data.
link |
I have to go to experiments.
link |
So you talk about machine learning
link |
is essentially learning by association,
link |
or reasoning by association,
link |
and this do calculus is allowing for intervention.
link |
So you also talk about counterfactuals.
link |
And trying to sort of understand the difference
link |
between counterfactuals and intervention.
link |
First of all, what is counterfactuals,
link |
and why are they useful?
link |
Why are they especially useful,
link |
as opposed to just reasoning what effect actions have?
link |
Well, counterfactual contains
link |
what we normally call explanations.
link |
Can you give an example of a counterfactual?
link |
If I tell you that acting one way affects something else,
link |
I didn't explain anything yet.
link |
But if I ask you, was it the aspirin that cured my headache?
link |
I'm asking for explanation, what cured my headache?
link |
And putting a finger on aspirin provide an explanation.
link |
It was aspirin that was responsible
link |
for your headache going away.
link |
If you didn't take the aspirin,
link |
you would still have a headache.
link |
So by saying if I didn't take aspirin,
link |
I would have a headache, you're thereby saying
link |
that aspirin is the thing that removes the headache.
link |
But you have to have another important information.
link |
I took the aspirin, and my headache is gone.
link |
It's very important information.
link |
Now I'm reasoning backward,
link |
and I said, was it the aspirin?
link |
By considering what would have happened
link |
if everything else is the same, but I didn't take aspirin.
link |
So you know that things took place.
link |
Joe killed Schmoe, and Schmoe would be alive
link |
had John not used his gun.
link |
Okay, so that is the counterfactual.
link |
It had the conflict here, or clash,
link |
between observed fact,
link |
but he did shoot, okay?
link |
And the hypothetical predicate,
link |
which says had he not shot,
link |
you have a logical clash.
link |
They cannot exist together.
link |
That's the counterfactual.
link |
And that is the source of our explanation
link |
of the idea of responsibility, regret, and free will.
link |
Yeah, so it certainly seems
link |
that's the highest level of reasoning, right?
link |
Yeah, and physicists do it all the time.
link |
Who does it all the time?
link |
In every equation of physics,
link |
let's say you have a Hooke's law,
link |
and you put one kilogram on the spring,
link |
and the spring is one meter,
link |
and you say, had this weight been two kilogram,
link |
the spring would have been twice as long.
link |
It's no problem for physicists to say that,
link |
except that mathematics is only in the form of equation,
link |
okay, equating the weight,
link |
proportionality constant, and the length of the string.
link |
So you don't have the asymmetry
link |
in the equation of physics,
link |
although every physicist thinks counterfactually.
link |
Ask the high school kids,
link |
had the weight been three kilograms,
link |
what would be the length of the spring?
link |
They can answer it immediately,
link |
because they do the counterfactual processing in their mind,
link |
and then they put it into equation,
link |
algebraic equation, and they solve it, okay?
link |
But a robot cannot do that.
link |
How do you make a robot learn these relationships?
link |
Well, why you would learn?
link |
Suppose you tell him, can you do it?
link |
So before you go learning,
link |
you have to ask yourself,
link |
suppose I give you all the information, okay?
link |
Can the robot perform the task that I ask him to perform?
link |
Can he reason and say, no, it wasn't the aspirin.
link |
It was the good news you received on the phone.
link |
Right, because, well, unless the robot had a model,
link |
a causal model of the world.
link |
I'm sorry I have to linger on this.
link |
But now we have to linger and we have to say,
link |
How do we build it?
link |
How do we build a causal model
link |
without a team of human experts running around?
link |
Why don't you go to learning right away?
link |
You're too much involved with learning.
link |
Because I like babies.
link |
Babies learn fast.
link |
I'm trying to figure out how they do it.
link |
So that's another question.
link |
How do the babies come out with a counterfactual
link |
model of the world?
link |
And babies do that.
link |
They know how to play in the crib.
link |
They know which balls hit another one.
link |
And they learn it by playful manipulation of the world.
link |
The simple world involves only toys and balls and chimes.
link |
But if you think about it, it's a complex world.
link |
We take for granted how complicated.
link |
And kids do it by playful manipulation
link |
plus parents guidance, peer wisdom, and hearsay.
link |
They meet each other and they say,
link |
you shouldn't have taken my toy.
link |
And these multiple sources of information,
link |
they're able to integrate.
link |
So the challenge is about how to integrate,
link |
how to form these causal relationships
link |
from different sources of data.
link |
So how much information is it to play,
link |
how much causal information is required
link |
to be able to play in the crib with different objects?
link |
I haven't experimented with the crib.
link |
Picking up, manipulating physical objects
link |
on this very, opening the pages of a book,
link |
all the tasks, the physical manipulation tasks.
link |
Do you have a sense?
link |
Because my sense is the world is extremely complicated.
link |
It's extremely complicated.
link |
I agree, and I don't know how to organize it
link |
because I've been spoiled by easy problems
link |
such as cancer and death, okay?
link |
And I'm a, but she's a.
link |
First we have to start trying to.
link |
No, but it's easy.
link |
There is in a sense that you have only 20 variables.
link |
And they are just variables and not mechanics.
link |
You just put them on the graph and they speak to you.
link |
Yeah, and you're providing a methodology
link |
for letting them speak.
link |
I'm working only in the abstract.
link |
The abstract was knowledge in, knowledge out,
link |
Now, can we take a leap to trying to learn
link |
in this very, when it's not 20 variables,
link |
but 20 million variables, trying to learn causation
link |
Not learn, but somehow construct models.
link |
I mean, it seems like you would only have to be able
link |
to learn because constructing it manually
link |
would be too difficult.
link |
Do you have ideas of?
link |
I think it's a matter of combining simple models
link |
from many, many sources, from many, many disciplines,
link |
and many metaphors.
link |
Metaphors are the basics of human intelligence, basis.
link |
Yeah, so how do you think of about a metaphor
link |
in terms of its use in human intelligence?
link |
Metaphors is an expert system.
link |
An expert, it's mapping problem
link |
from a problem with which you are not familiar
link |
to a problem with which you are familiar.
link |
Like, I'll give you a good example.
link |
The Greek believed that the sky is an opaque shell.
link |
It's not really infinite space.
link |
It's an opaque shell, and the stars are holes
link |
poked in the shells through which you see
link |
the eternal light.
link |
That was a metaphor.
link |
Because they understand how you poke holes in the shells.
link |
They were not familiar with infinite space.
link |
And we are walking on a shell of a turtle,
link |
and if you get too close to the edge,
link |
you're gonna fall down to Hades or wherever.
link |
That's a metaphor.
link |
But this kind of metaphor enabled Aristoteles
link |
to measure the radius of the Earth,
link |
because he said, Kamal, if we are walking on a turtle shell,
link |
then the ray of light coming to this place
link |
will be a different angle than coming to this place.
link |
I know the distance, I'll measure the two angles,
link |
and then I have the radius of the shell of the turtle.
link |
And he did, and he found his measurement
link |
very close to the measurements we have today,
link |
through the, what, 6,700 kilometers of the Earth.
link |
That's something that would not occur
link |
to Babylonian astronomer,
link |
even though the Babylonian experiment
link |
were the machine learning people of the time.
link |
They fit curves, and they could predict
link |
the eclipse of the moon much more accurately
link |
than the Greek, because they fit curve.
link |
That's a different metaphor.
link |
Something that you're familiar with,
link |
a game, a turtle shell.
link |
What does it mean if you are familiar?
link |
Familiar means that answers to certain questions
link |
You don't have to derive them.
link |
And they were made explicit because somewhere in the past
link |
you've constructed a model of that.
link |
Yeah, you're familiar with,
link |
so the child is familiar with billiard balls.
link |
So the child could predict that if you let loose
link |
of one ball, the other one will bounce off.
link |
You obtain that by familiarity.
link |
Familiarity is answering questions,
link |
and you store the answer explicitly.
link |
You don't have to derive them.
link |
So this is the idea of a metaphor.
link |
All our life, all our intelligence
link |
is built around metaphors,
link |
mapping from the unfamiliar to the familiar.
link |
But the marriage between the two is a tough thing,
link |
which we haven't yet been able to algorithmatize.
link |
So you think of that process of using metaphor
link |
to leap from one place to another,
link |
we can call it reasoning?
link |
Is it a kind of reasoning?
link |
It is reasoning by metaphor, metaphorical reasoning.
link |
Do you think of that as learning?
link |
So learning is a popular terminology today
link |
in a narrow sense.
link |
It is, it is, it is definitely a form.
link |
So you may not, okay, right.
link |
It's one of the most important learnings,
link |
taking something which theoretically is derivable
link |
and store it in accessible format.
link |
I'll give you an example, chess, okay?
link |
Finding the winning starting move in chess is hard.
link |
It is hard, but there is an answer.
link |
Either there is a winning move for white
link |
or there isn't, or there is a draw, okay?
link |
So it is, the answer to that
link |
is available through the rule of the games.
link |
But we don't know the answer.
link |
So what does a chess master have that we don't have?
link |
He has stored explicitly an evaluation
link |
of certain complex pattern of the board.
link |
Ordinary people like me, I don't know about you,
link |
I'm not a chess master.
link |
So for me, I have to derive things that for him is explicit.
link |
He has seen it before, or he has seen the pattern before,
link |
or similar pattern, you see metaphor, yeah?
link |
And he generalize and said, don't move, it's a dangerous move.
link |
It's just that not in the game of chess,
link |
but in the game of billiard balls,
link |
we humans are able to initially derive very effectively
link |
and then reason by metaphor very effectively
link |
and make it look so easy that it makes one wonder
link |
how hard is it to build it in a machine.
link |
So in your sense, how far away are we
link |
to be able to construct?
link |
I don't know, I'm not a futurist.
link |
All I can tell you is that we are making tremendous progress
link |
in the causal reasoning domain.
link |
Something that I even dare to call it revolution,
link |
the code of revolution, because what we have achieved
link |
in the past three decades is something that dwarf
link |
everything that was derived in the entire history.
link |
So there's an excitement about
link |
current machine learning methodologies,
link |
and there's really important good work you're doing
link |
in causal inference.
link |
Where does the future, where do these worlds collide
link |
and what does that look like?
link |
First, they're gonna work without collision.
link |
It's gonna work in harmony.
link |
Harmony, it's not collision.
link |
The human is going to jumpstart the exercise
link |
by providing qualitative, noncommitting models
link |
of how the universe works, how in reality
link |
the domain of discourse works.
link |
The machine is gonna take over from that point of view
link |
and derive whatever the calculus says can be derived.
link |
Namely, quantitative answer to our questions.
link |
Now, these are complex questions.
link |
I'll give you some example of complex questions
link |
that will bug your mind if you think about it.
link |
You take result of studies in diverse population
link |
under diverse condition, and you infer the cause effect
link |
of a new population which doesn't even resemble
link |
any of the ones studied, and you do that by do calculus.
link |
You do that by generalizing from one study to another.
link |
See, what's common with Berto?
link |
What is different?
link |
Let's ignore the differences and pull out the commonality,
link |
and you do it over maybe 100 hospitals around the world.
link |
From that, you can get really mileage from big data.
link |
It's not only that you have many samples,
link |
you have many sources of data.
link |
So that's a really powerful thing, I think,
link |
especially for medical applications.
link |
I mean, cure cancer, right?
link |
That's how from data you can cure cancer.
link |
So we're talking about causation,
link |
which is the temporal relationships between things.
link |
Not only temporal, it's both structural and temporal.
link |
Temporal enough, temporal precedence by itself
link |
cannot replace causation.
link |
Is temporal precedence the arrow of time in physics?
link |
It's important, necessary.
link |
It's efficient, yes.
link |
Yes, I never seen cause propagate backward.
link |
But if we use the word cause,
link |
but there's relationships that are timeless.
link |
I suppose that's still forward in the arrow of time.
link |
But are there relationships, logical relationships,
link |
that fit into the structure?
link |
Sure, the whole do calculus is logical relationship.
link |
That doesn't require a temporal.
link |
It has just the condition that
link |
you're not traveling back in time.
link |
So it's really a generalization of,
link |
a powerful generalization of what?
link |
Yeah, Boolean logic.
link |
That is sort of simply put,
link |
and allows us to reason about the order of events,
link |
Not about, between, we're not deriving the order of events.
link |
We are given cause effects relationship, okay?
link |
They ought to be obeying the time presidents relationship.
link |
And now that we ask questions about
link |
other causes of relationship,
link |
that could be derived from the initial ones,
link |
but were not given to us explicitly.
link |
Like the case of the firing squad I gave you
link |
in the first chapter.
link |
And I ask, what if rifleman A declined to shoot?
link |
Would the prisoner still be dead?
link |
To decline to shoot, it means that he disobey order.
link |
And the rule of the games were that he is a
link |
obedient marksman, okay?
link |
That's how you start.
link |
That's the initial order.
link |
But now you ask question about breaking the rules.
link |
What if he decided not to pull the trigger?
link |
He just became a pacifist.
link |
And you and I can answer that.
link |
The other rifleman would have killed him, okay?
link |
I want the machine to do that.
link |
Is it so hard to ask a machine to do that?
link |
It's such a simple task.
link |
You have to have a calculus for that.
link |
But the curiosity, the natural curiosity for me is
link |
that yes, you're absolutely correct and important.
link |
And it's hard to believe that we haven't done this
link |
seriously extensively already a long time ago.
link |
So this is really important work.
link |
But I also wanna know, maybe you can philosophize
link |
about how hard is it to learn.
link |
Okay, let's assume we're learning.
link |
We wanna learn it, okay?
link |
We put a learning machine that watches execution trials
link |
in many countries and many locations, okay?
link |
All the machine can learn is to see shut or not shut.
link |
A court issued an order or didn't, okay?
link |
For the fact you don't know who listens to whom.
link |
You don't know that the condemned person
link |
listened to the bullets,
link |
that the bullets are listening to the captain, okay?
link |
All we hear is one command, two shots, dead, okay?
link |
A triple of variable.
link |
Okay, that you can learn who listens to whom
link |
and you can answer the question, no.
link |
But don't you think you can start proposing ideas
link |
for humans to review?
link |
You want machine to learn, you want a robot.
link |
So robot is watching trials like that, 200 trials,
link |
and then he has to answer the question,
link |
what if rifleman A refrain from shooting?
link |
That's exactly my point.
link |
It's looking at the facts,
link |
don't give you the strings behind the facts.
link |
Absolutely, but do you think of machine learning
link |
as it's currently defined as only something
link |
that looks at the facts and tries to do?
link |
Right now, they only look at the facts, yeah.
link |
So is there a way to modify, in your sense?
link |
Playful manipulation.
link |
Playful manipulation.
link |
Yes, once in a while.
link |
Doing the interventionist kind of thing, intervention.
link |
But it could be at random.
link |
For instance, the rifleman is sick that day
link |
or he just vomits or whatever.
link |
So machine can observe this unexpected event
link |
which introduce noise.
link |
The noise still have to be random
link |
to be able to relate it to randomized experiment.
link |
And then you have observational studies
link |
from which to infer the strings behind the facts.
link |
It's doable to a certain extent.
link |
But now that we are expert in what you can do
link |
once you have a model, we can reason back and say,
link |
what kind of data you need to build a model.
link |
Got it, so I know you're not a futurist,
link |
but are you excited?
link |
Have you, when you look back at your life,
link |
longed for the idea of creating
link |
a human level intelligence system?
link |
Yeah, I'm driven by that.
link |
All my life, I'm driven just by one thing.
link |
I go from what I know to the next step incrementally.
link |
So without imagining what the end goal looks like.
link |
Do you imagine what an eight?
link |
The end goal is gonna be a machine
link |
that can answer sophisticated questions,
link |
counterfactuals of regret, compassion,
link |
responsibility, and free will.
link |
So what is a good test?
link |
Is a Turing test a reasonable test?
link |
A test of free will doesn't exist yet.
link |
How would you test free will?
link |
So far, we know only one thing.
link |
If robots can communicate with reward and punishment
link |
among themselves and hitting each other on the wrist
link |
and say, you shouldn't have done that, okay?
link |
Playing better soccer because they can do that.
link |
What do you mean, because they can do that?
link |
Because they can communicate among themselves.
link |
Because of the communication they can do.
link |
Because they communicate like us.
link |
Reward and punishment, yes.
link |
You didn't pass the ball at the right time,
link |
and so therefore you're gonna sit on the bench
link |
If they start communicating like that,
link |
the question is, will they play better soccer?
link |
As opposed to what?
link |
As opposed to what they do now?
link |
Without this ability to reason about reward and punishment.
link |
So far, I can only think about communication.
link |
Communication is, and not necessarily natural language,
link |
but just communication.
link |
Just communication.
link |
And that's important to have a quick and effective means
link |
of communicating knowledge.
link |
If the coach tells you you should have passed the ball,
link |
pink, he conveys so much knowledge to you
link |
as opposed to what?
link |
Go down and change your software.
link |
That's the alternative.
link |
But the coach doesn't know your software.
link |
So how can the coach tell you
link |
you should have passed the ball?
link |
But our language is very effective.
link |
You should have passed the ball.
link |
You know your software.
link |
You tweak the right module, and next time you don't do it.
link |
Now that's for playing soccer,
link |
the rules are well defined.
link |
No, no, no, no, they're not well defined.
link |
When you should pass the ball.
link |
Is not well defined.
link |
No, it's very soft, very noisy.
link |
Yes, you have to do it under pressure.
link |
But in terms of aligning values
link |
between computers and humans,
link |
do you think this cause and effect type of thinking
link |
is important to align the values,
link |
values, morals, ethics under which the machines
link |
make decisions, is the cause effect
link |
where the two can come together?
link |
Cause and effect is necessary component
link |
to build an ethical machine.
link |
Because the machine has to empathize
link |
to understand what's good for you,
link |
to build a model of you as a recipient,
link |
which should be very much, what is compassion?
link |
They imagine that you suffer pain as much as me.
link |
I do have already a model of myself, right?
link |
So it's very easy for me to map you to mine.
link |
I don't have to rebuild the model.
link |
It's much easier to say, oh, you're like me.
link |
Okay, therefore I would not hate you.
link |
And the machine has to imagine,
link |
has to try to fake to be human essentially
link |
so you can imagine that you're like me, right?
link |
And moreover, who is me?
link |
That's the first, that's consciousness.
link |
They have a model of yourself.
link |
Where do you get this model?
link |
You look at yourself as if you are a part
link |
of the environment.
link |
If you build a model of yourself
link |
versus the environment, then you can say,
link |
I need to have a model of myself.
link |
I have abilities, I have desires and so forth, okay?
link |
I have a blueprint of myself though.
link |
Not the full detail because I cannot
link |
get the whole thing problem.
link |
But I have a blueprint.
link |
So on that level of a blueprint, I can modify things.
link |
I can look at myself in the mirror and say,
link |
hmm, if I change this model, tweak this model,
link |
I'm gonna perform differently.
link |
That is what we mean by free will.
link |
And consciousness.
link |
And consciousness.
link |
What do you think is consciousness?
link |
Is it simply self awareness?
link |
So including yourself into the model of the world?
link |
Some people tell me, no, this is only part of consciousness.
link |
And then they start telling me what they really mean
link |
by consciousness, and I lose them.
link |
For me, consciousness is having a blueprint
link |
Do you have concerns about the future of AI?
link |
All the different trajectories of all of our research?
link |
Where's your hope, where the movement has,
link |
where are your concerns?
link |
I'm concerned, because I know we are building a new species
link |
that has a capability of exceeding our, exceeding us,
link |
exceeding our capabilities, and can breed itself
link |
and take over the world.
link |
It's a new species that is uncontrolled.
link |
We don't know the degree to which we control it.
link |
We don't even understand what it means
link |
to be able to control this new species.
link |
I don't have anything to add to that,
link |
because it's such a gray area, it's unknown.
link |
It never happened in history.
link |
The only time it happened in history
link |
was evolution with human beings.
link |
It wasn't very successful, was it?
link |
Some people say it was a great success.
link |
For us it was, but a few people along the way,
link |
a few creatures along the way would not agree.
link |
So it's just because it's such a gray area,
link |
there's nothing else to say.
link |
We have a sample of one.
link |
But some people would look at you and say,
link |
yeah, but we were looking to you to help us
link |
make sure that the sample two works out okay.
link |
We have more than a sample of one.
link |
We have theories, and that's a good.
link |
We don't need to be statisticians.
link |
So sample of one doesn't mean poverty of knowledge.
link |
Sample of one plus theory, conjectural theory,
link |
of what could happen.
link |
But I really feel helpless in contributing
link |
to this argument, because I know so little,
link |
and my imagination is limited,
link |
and I know how much I don't know,
link |
and I, but I'm concerned.
link |
You were born and raised in Israel.
link |
Born and raised in Israel, yes.
link |
And later served in Israel military,
link |
In the Israel Defense Force.
link |
What did you learn from that experience?
link |
From this experience?
link |
There's a kibbutz in there as well.
link |
Yes, because I was in the nachal,
link |
which is a combination of agricultural work
link |
and military service.
link |
We were supposed, I was really idealist.
link |
I wanted to be a member of the kibbutz throughout my life,
link |
and to live a communal life,
link |
and so I prepared myself for that.
link |
Slowly, slowly, I wanted a greater challenge.
link |
So that's a far world away, both.
link |
What I learned from that, what I can add,
link |
It was a miracle that I served in the 1950s.
link |
I don't know how we survived.
link |
The country was under austerity.
link |
It tripled its population from 600,000 to a million point eight
link |
when I finished college.
link |
No one went hungry.
link |
And austerity, yes.
link |
When you wanted to make an omelet in a restaurant,
link |
you had to bring your own egg.
link |
And they imprisoned people from bringing the food
link |
from farming here, from the villages, to the city.
link |
But no one went hungry.
link |
And I always add to it,
link |
and higher education did not suffer any budget cut.
link |
They still invested in me, in my wife, in our generation
link |
to get the best education that they could, okay?
link |
So I'm really grateful for the opportunity,
link |
and I'm trying to pay back now, okay?
link |
It's a miracle that we survived the war of 1948.
link |
We were so close to a second genocide.
link |
It was all planned.
link |
But we survived it by miracle,
link |
and then the second miracle
link |
that not many people talk about, the next phase.
link |
How no one went hungry,
link |
and the country managed to triple its population.
link |
You know what it means to triple?
link |
Imagine United States going from what, 350 million
link |
to a trillion, unbelievable.
link |
So it's a really tense part of the world.
link |
It's a complicated part of the world,
link |
Israel and all around.
link |
Religion is at the core of that complexity.
link |
One of the components.
link |
Religion is a strong motivating cause
link |
to many, many people in the Middle East, yes.
link |
In your view, looking back, is religion good for society?
link |
That's a good question for robotic, you know?
link |
There's echoes of that question.
link |
Equip robot with religious belief.
link |
Suppose we find out, or we agree
link |
that religion is good to you, to keep you in line, okay?
link |
Should we give the robot the metaphor of a god?
link |
As a matter of fact, the robot will get it without us also.
link |
The robot will reason by metaphor.
link |
And what is the most primitive metaphor
link |
a child grows with?
link |
Mother smile, father teaching,
link |
father image and mother image, that's god.
link |
So, whether you want it or not,
link |
the robot will, well, assuming that the robot
link |
is gonna have a mother and a father,
link |
it may only have a programmer,
link |
which doesn't supply warmth and discipline.
link |
Well, discipline it does.
link |
So the robot will have a model of the trainer,
link |
and everything that happens in the world,
link |
cosmology and so on, is going to be mapped
link |
into the programmer, it's god.
link |
Man, the thing that represents the origin
link |
of everything for that robot.
link |
It's the most primitive relationship.
link |
So it's gonna arrive there by metaphor.
link |
And so the question is if overall
link |
that metaphor has served us well as humans.
link |
I really don't know.
link |
I think it did, but as long as you keep
link |
in mind it's only a metaphor.
link |
So, if you think we can, can we talk about your son?
link |
Can you tell his story?
link |
His story is known, he was abducted
link |
in Pakistan by Al Qaeda driven sect,
link |
and under various pretenses.
link |
I don't even pay attention to what the pretence was.
link |
Originally they wanted to have the United States
link |
deliver some promised airplanes.
link |
It was all made up, and all these demands were bogus.
link |
Bogus, I don't know really, but eventually
link |
he was executed in front of a camera.
link |
At the core of that is hate and intolerance.
link |
At the core, yes, absolutely, yes.
link |
We don't really appreciate the depth of the hate
link |
at which billions of peoples are educated.
link |
We don't understand it.
link |
I just listened recently to what they teach you
link |
Okay, okay, when the water stopped in the tap,
link |
we knew exactly who did it, the Jews.
link |
We didn't know how, but we knew who did it.
link |
We don't appreciate what it means to us.
link |
The depth is unbelievable.
link |
Do you think all of us are capable of evil?
link |
And the education, the indoctrination
link |
is really what creates evil.
link |
Absolutely we are capable of evil.
link |
If you're indoctrinated sufficiently long and in depth,
link |
you're capable of ISIS, you're capable of Nazism.
link |
Yes, we are, but the question is whether we,
link |
after we have gone through some Western education
link |
and we learn that everything is really relative.
link |
It is not absolute God.
link |
It's only a belief in God.
link |
Whether we are capable now of being transformed
link |
under certain circumstances to become brutal.
link |
I'm worried about it because some people say yes,
link |
given the right circumstances,
link |
given bad economical crisis,
link |
you are capable of doing it too.
link |
I want to believe it, I'm not capable.
link |
So seven years after Daniel's death,
link |
you wrote an article at the Wall Street Journal
link |
titled Daniel Pearl and the Normalization of Evil.
link |
What was your message back then
link |
and how did it change today over the years?
link |
What was the message?
link |
The message was that we are not treating terrorism
link |
We are treating it as a bargaining device that is accepted.
link |
People have grievance and they go and bomb restaurants.
link |
Look, you're even not surprised when I tell you that.
link |
20 years ago you say, what?
link |
For grievance you go and blow a restaurant?
link |
Today it's becoming normalized.
link |
The banalization of evil.
link |
And we have created that to ourselves by normalizing,
link |
by making it part of political life.
link |
It's a political debate.
link |
Every terrorist yesterday becomes a freedom fighter today
link |
and tomorrow it becomes terrorist again.
link |
Right, and so we should call out evil when there's evil.
link |
If we don't want to be part of it.
link |
Yeah, if we want to separate good from evil,
link |
that's one of the first things that,
link |
what was it, in the Garden of Eden,
link |
remember the first thing that God told him was,
link |
hey, you want some knowledge, here's a tree of good and evil.
link |
Yeah, so this evil touched your life personally.
link |
Does your heart have anger, sadness, or is it hope?
link |
Look, I see some beautiful people coming from Pakistan.
link |
I see beautiful people everywhere.
link |
But I see horrible propagation of evil in this country too.
link |
It shows you how populistic slogans
link |
can catch the mind of the best intellectuals.
link |
Today is Father's Day.
link |
I didn't know that.
link |
Yeah, what's a fond memory you have of Daniel?
link |
What's a fond memory you have of Daniel?
link |
Oh, very good memories, immense.
link |
He had a sense of balance that I didn't have.
link |
He saw the beauty in every person.
link |
He was not as emotional as I am,
link |
the more looking things in perspective.
link |
He really liked every person.
link |
He really grew up with the idea that a foreigner
link |
is a reason for curiosity, not for fear.
link |
That one time we went in Berkeley,
link |
and a homeless came out from some dark alley,
link |
and said, hey, man, can you spare a dime?
link |
I retreated back, two feet back,
link |
and then I just hugged him and say,
link |
here's a dime, enjoy yourself.
link |
Maybe you want some money to take a bus or whatever.
link |
Where did you get it?
link |
Do you have advice for young minds today,
link |
dreaming about creating as you have dreamt,
link |
creating intelligent systems?
link |
What is the best way to arrive at new breakthrough ideas
link |
and carry them through the fire of criticism
link |
and past conventional ideas?
link |
Ask your questions freely.
link |
Your questions are never dumb.
link |
And solve them your own way.
link |
And don't take no for an answer.
link |
Look, if they are really dumb,
link |
you will find out quickly by trying an arrow
link |
to see that they're not leading any place.
link |
But follow them and try to understand things your way.
link |
That is my advice.
link |
I don't know if it's gonna help anyone.
link |
Not as brilliantly.
link |
There is a lot of inertia in science, in academia.
link |
It is slowing down science.
link |
Yeah, those two words, your way, that's a powerful thing.
link |
It's against inertia, potentially, against the flow.
link |
Against your professor.
link |
Against your professor.
link |
I wrote the Book of Why in order to democratize
link |
In order to instill rebellious spirit in students
link |
so they wouldn't wait until the professor get things right.
link |
So you wrote the manifesto of the rebellion
link |
against the professor.
link |
Against the professor, yes.
link |
So looking back at your life of research,
link |
what ideas do you hope ripple through the next many decades?
link |
What do you hope your legacy will be?
link |
I already have a tombstone carved.
link |
The fundamental law of counterfactuals.
link |
That's what, it's a simple equation.
link |
Counterfactual in terms of a model surgery.
link |
That's it, because everything follows from that.
link |
If you get that, all the rest, I can die in peace.
link |
And my student can derive all my knowledge
link |
by mathematical means.
link |
Thank you so much for talking today.
link |
I really appreciate it.
link |
Thank you for being so attentive and instigating.
link |
The coffee helped.
link |
Thanks for listening to this conversation with Judea Pearl.
link |
And thank you to our presenting sponsor, Cash App.
link |
Download it, use code LexPodcast, you'll get $10,
link |
and $10 will go to FIRST, a STEM education nonprofit
link |
that inspires hundreds of thousands of young minds
link |
to learn and to dream of engineering our future.
link |
If you enjoy this podcast, subscribe on YouTube,
link |
give it five stars on Apple Podcast,
link |
support on Patreon, or simply connect with me on Twitter.
link |
And now, let me leave you with some words of wisdom
link |
You cannot answer a question that you cannot ask,
link |
and you cannot ask a question that you have no words for.
link |
Thank you for listening, and hope to see you next time.