Christof Koch: Consciousness
Guests: Christof Koch
Full transcript · Read the full conversation in order, with its original context.
§01Transcript · 00:00–08:03
As part of MIT course 6S099 on Artificial General Intelligence,
I got a chance to sit down with Christoph Koch,
who is one of the seminal figures in neurobiology, neuroscience,
and generally in the study of consciousness.
He is the president, the chief scientific officer
of the Allen Institute for Brain Science in Seattle.
From 1986 to 2013, he was the professor at Caltech.
Before that, he was at MIT.
He is extremely well cited, over 100,000 citations.
His research, his writing, his ideas have had big impact
on the scientific community and the general public
in the way we think about consciousness,
in the way we see ourselves as human beings.
He's the author of several books,
The Quest for Consciousness, a Neurobiological Approach,
and a more recent book,
Consciousness, Confessions of a Romantic Reductionist.
If you enjoy this conversation, this course,
subscribe, click the little bell icon
to make sure you never miss a video,
and in the comments, leave suggestions
for any people you'd like to see be part of the course
or any ideas that you would like us to explore.
Thanks very much, and I hope you enjoy.
Okay, before we delve into the beautiful mysteries
of consciousness, let's zoom out a little bit.
And let me ask, do you think there's intelligent life
out there in the universe?
- Yes, I do believe so.
We have no evidence of it,
but I think the probabilities are overwhelming
in favor of it, given universe,
where we have 10 to the 11 galaxies,
and each galaxy has between 10 to the 11, 10 to the 12 stars,
and we know most stars have one or more planets.
- So how does that make you feel?
- It still makes me feel special
because I have experiences.
I feel the world, I experience the world.
And independent of whether there are other creatures
out there, I still feel the world,
and I have access to this world
in this very strange, compelling way,
and that's the core of human existence.
- Now, you said human.
Do you think, if those intelligent creatures are out there,
do you think they experience their world?
- Yes, if they are evolved,
if they are a product of natural evolution,
as they would have to be,
they will also experience their own world.
So consciousness isn't just human, you're right,
it's much wider.
It's probably, it may be spread across all of biology.
We have, the only thing that we have special
is we can talk about it.
Of course, not all people can talk about it.
Babies and little children can talk about it,
patients who have a stroke in the left inferior frontal gyrus
can talk about it,
but most normal adult people can talk about it,
and so we think that makes us special
compared to, let's say, monkeys or dogs or cats or mice
or all the other creatures that we share the planet with.
But all the evidence seems to suggest
that they too experience the world,
and so it's overwhelmingly likely that other alien,
that aliens would also experience their world.
Of course, differently,
because they have a different sensorium,
they have different sensors,
they have a very different environment.
But the fact that I would strongly suppose
that they also have experiences.
They feel pain and pleasure
and see in some sort of spectrum
and hear and have all the other senses.
- Of course, their language,
if they have one, would be different,
so we might not be able to understand their poetry
about the experiences that they have.
- That's correct.
- Right.
So, in a talk, in a video,
I've heard you mention Sapuzo,
a Dachshund that you came up with,
that you grew up with,
was part of your family when you were young.
First of all, you're technically a Midwestern boy.
You just-- - Technically.
(laughs)
- But after that, you traveled around a bit,
hence a little bit of the accent.
You talked about Sapuzo,
the Dachshund,
having these elements of humanness,
of consciousness that you discovered.
So, I just wanted to ask,
can you look back in your childhood
and remember when was the first time
you realized you, yourself,
sort of from a third-person perspective,
are a conscious being?
This idea of stepping outside yourself
and seeing there's something special
going on here in my brain.
- I can't really actually,
it's a good question.
I'm not sure I recall a discrete moment.
I mean, you take it for granted
because that's the only world you know, right?
The only world I know, you know,
is the world of seeing and hearing voices
and touching and all the other things.
So, it's only much later,
at early in my undergraduate days
when I enrolled in physics and in philosophy
that I really thought about it
and thought, well, this is really,
fundamentally, very, very mysterious.
And there's nothing really in physics right now
that explains this transition
from the physics of the brain to feelings.
Where do the feelings come in?
So, you can look at the foundational equation
of quantum mechanics, general relativity.
You can look at the periodic table of the elements.
You can look at the endless ATGC chat in our genes
and nowhere is consciousness.
Yet, I wake up every morning to a world
where I have experiences.
And so, that's the heart
of the ancient mind-body problem.
How do experiences get into the world?
- So, what is consciousness?
- Experience.
Consciousness is any experience.
Some people call it subjective feelings.
Some people call it phenomenology.
Some people call it qualia, if they're philosopher.
But they all denote the same thing.
It feels like something,
in the famous word of the philosopher Thomas Nagel,
it feels like something to be a bat
or to be an American,
or to be angry, or to be sad,
or to be in love, or to have pain.
And that is what experience is.
Any possible experience.
Could be as mundane as just sitting in a chair.
Could be as exalted as having a mystical moment
in deep meditation.
Those are just different forms of experiences.
- Experience.
So, if you were to sit down with maybe the next,
skip a couple generations, of IBM Watson,
something that won Jeopardy.
What is the gap, I guess the question is,
between Watson, that might be much smarter than you,
than us, than any human alive,
but may not have experience.
What is the gap?
- Well, so that's a big, big question.
That's occupied people for the last,
certainly the last 50 years,
since the advent, the birth of computers.
That's a question Alan Turing tried to answer.
And of course, he did it in this indirect way,
by proposing a test, an operational test.
But that's not really,
he tried to get at what does it mean for a person to think.
And then he had this test, you lock 'em away,
and then you have a communication with them,
and then you try to guess after a while
whether that is a person or whether it's a computer system.
There's no question that now, or very soon,
Alexa or Siri or Google Now will pass this test.
And you can game it, but ultimately,
certainly in your generation,
there will be machines that will speak with complete poise,
that will remember everything you ever said,
they'll remember every email you ever had.
Like Samantha, remember in the movie "Her"?
There's no question it's gonna happen.
But of course, the key question is,
does it feel like anything to be Samantha in the movie "Her"?
Does it feel like anything to be Watson?
And there, one has to very strongly think,
there are two different concepts here that we commingle.
There is the concept of intelligence,
natural or artificial,
and there is a concept of consciousness,
of experience, natural or artificial.
Those are very, very different things.
Now, historically, we associate consciousness
with intelligence, why?
Because we live in a world, leaving aside computers,
of natural selection, where we're surrounded by creatures,
either our own kin that are less or more intelligent,
or we go across species.
§02Transcript · 08:05–15:40
Some are more adapted to a particular environment,
others are less adapted, whether it's a whale or dog
or you go talk about a paramecium or a little worm.
And we see the complexity of the nervous system
goes from one cell to specialized cells,
to a worm that has three nets,
that has 30% of its cells are nerve cells,
to a creature like us or like a blue whale
that has 100 billion, even more nerve cells.
And so, based on behavioral evidence
and based on the underlying neuroscience,
we believe that as these creatures become more complex,
they are better adapted to their particular ecological niche,
and they become more conscious,
partly because their brain grows,
and we believe consciousness, unlike the ancient,
ancient people thought most, almost every culture thought
that consciousness with intelligence
has to do with your heart.
And you still see that today, you see,
honey, I love you with all my heart.
But what you should actually say is,
you say, no, honey, I love you
with all my lateral hypothalamus.
And for Valentine's Day, you should give your sweetheart,
you know, a hypothalamic shaped piece of chocolate,
not a heart-shaped chocolate, right?
Anyway, so we still have this language,
but now we believe it's a brain.
And so we see brains of different complexity,
and we think, well,
they have different levels of consciousness.
They're capable of different experiences.
But now we confront a world where we know,
where we're beginning to engineer intelligence,
and it's radical unclear whether the intelligence
we're engineering has anything to do with consciousness
and whether it can experience anything.
'Cause fundamentally, what's the difference?
Intelligence is about function.
Intelligence, no matter exactly how you define it,
sort of adaptation to new environments,
being able to learn and quickly understand,
you know, the setup of this and what's going on
and who are the actors and what's gonna happen next,
that's all about function.
Consciousness is not about function.
Consciousness is about being.
It's in some sense much fundamental.
You can see this in several cases.
You can see it, for instance, in the case of the clinic.
When you're dealing with patients who, let's say,
had a stroke or were in a traffic accident, et cetera,
they're pretty much immobile.
Terri Schiavo, you may have heard historically,
she was a person here in the '90s in Florida.
Her heart stood still.
She was reanimated.
Then for the next 14 years,
she was what's called in a vegetative state.
So there are thousands of people in a vegetative state.
So they're like this.
Occasionally, they open their eyes
for two, three, four, five, six, eight hours
and then close their eyes.
They have sleep-wake cycle.
Occasionally, they have behavior.
They do like, you know, they're,
but there's no way that you can establish
a lawful relationship between what you say
or the doctor says or the mom says
and what the patient does.
- Correct.
- So there isn't any behavior,
yet in some of these people,
there is still experience.
You can design and build brain machine interfaces
where you can see they still experience something.
And of course, in these cases of locked-in state,
there's this famous book called
"The Diving Bell and the Butterfly"
where you had an editor, a French editor,
he had a stroke in the brainstem,
unable to move except his vertical eyes, eye movement.
He could just move his eyes up and down.
And he dictated an entire book.
And some people even lose this at the end.
And all the evidence seems to suggest
that they're still in there.
In this case, you have no behavior, you have consciousness.
Second case is tonight, like all of us,
you're gonna go to sleep.
Close your eyes, you go to sleep.
You will wake up inside your sleeping body
and you will have conscious experiences.
They are different from everyday experience.
You might fly, you might not be surprised
that you're flying.
You might meet a long-dead pet childhood dog
and you're not surprised that you're meeting them,
you know, but you have conscious experience
of love, of hate, you know, they can be very emotional.
Your body during this state, typically it's REM,
state sends an active signal to your motor neurons
to paralyze you, it's called atonia, right?
Because if you don't have that, like some patients,
what do you do?
You act out your dreams.
You get, for example, REM behavioral disorder,
which is bad juju to get, okay?
Third case is pure experience.
So I recently had this,
what some people call a mystical experience.
I went to Singapore and went into a flotation tank, right?
So this is a big tub filled with water
that's body temperature and Epsom salt.
You strip completely naked, you lie inside of it.
You close the lid. - Darkness.
- Complete darkness, soundproof.
So very quickly you become bodyless
because you're floating and you're naked.
You have no rings, no watch, no nothing.
You don't feel your body anymore.
It's no sound, soundless.
There's no photon, sightless, timeless,
because after a while, early on,
you actually hear your heart,
but then you sort of adapt to that
and then sort of the passage of time ceases.
And if you train yourself, like in a meditation,
not to think, early on you think a lot.
It's a little bit spooky.
You feel somewhat uncomfortable
or you think, well, I'm gonna get bored.
But if you try to not to think actively,
you become mindless.
There you are, bodyless, timeless,
soundless, sightless, mindless,
but you're in a conscious experience.
You're not asleep.
You're not asleep.
You are pure being.
There isn't any function.
You aren't doing any computation.
You're not remembering.
You're not projecting.
You're not planning, yet you are fully conscious.
- You're fully conscious.
There's something going on out there.
It could be just a side effect.
So what is the--
- You mean epiphenomena.
- So what's the-- - You mean a side effect.
- Meaning what is the function of you being able
to lay in this sensory-free deprivation tank
and still have a conscious experience?
- Evolutionary? - Evolutionary.
- Obviously, we didn't evolve with flotation tanks
in our environment.
I mean, so biology is notoriously bad
at asking why question, tele-anomical question.
Why do we have two eyes?
Why don't we have four eyes, like some teachers,
or three eyes or something?
Well, no, there's probably, there is a function to that,
but we're not very good at answering those questions.
We can speculate endlessly.
Biology is very, or science is very good
about mechanistic questions.
Why is there a charge in the universe, right?
We find ourselves in a universe
where there are positive and negative charges, why?
Why does quantum mechanics hold?
Why doesn't some other theory hold?
Quantum mechanics hold in our universe.
It's very unclear why.
So tele-anomical questions, why questions,
are difficult to answer.
Clearly, there's some relationship between complexity,
brain processing power, and consciousness.
But, however, in these cases, in these three examples I gave,
one is an everyday experience at night.
The other one is a trauma,
and third one is in principle,
everybody can have these sort of mystical experiences.
You have a dissociation of function,
of intelligence, from consciousness.
- You caught me asking a why question.
Let me ask a question that's not a why question.
You're giving a talk later today on the Turing test
for intelligence and consciousness,
drawing lines between the two.
So is there a scientific way to say
there's consciousness present in this entity or not?
And to anticipate your answer,
'cause there's a neurobiological answer.
So we can test the human brain,
but if you take a machine brain
that you don't know tests for yet,
how would you even begin to approach a test
§03Transcript · 15:44–23:25
if there's consciousness present in this thing?
- Okay, that's a really good question.
So let me take it in two steps.
So as you point out, for humans,
let's just stick with humans,
there's now a test called the ZAP and ZIP.
It's a procedure where you ping the brain
using transcranial magnetic stimulation.
You look at the electrical reverberations,
essentially using EG,
and then you can measure the complexity
of this brain response.
And you can do this in awake people,
in asleep, normal people,
you can do it in awake people and then anesthetize them.
You can do it in patients.
And it has 100% accuracy that in all those cases,
when you're clear, the patient or the person
is either conscious or unconscious,
the complexity is either high or low.
And then you can adopt these techniques
to similar creatures like monkeys and dogs
and mice that have very similar brains.
Now, of course, you point out that may not help you
because we don't have a cortex,
and if I send a magnetic pulse into my iPhone
or my computer, it's probably gonna break something.
So we don't have that.
So what we need ultimately,
we need a theory of consciousness.
We can't just rely on our intuition.
Our intuition is, well, yeah,
if somebody talks, they're conscious.
However, then there are all these children,
babies don't talk, right?
But we believe that babies also have conscious experiences.
And then there are all these patients I mentioned,
and they don't talk.
When you dream, you can't talk because you're paralyzed.
So what we ultimately need,
we can't just rely on our intuition.
We need a theory of conscience that tells us
what is it about a piece of matter?
What is it about a piece of highly excitable matter,
like the brain or like a computer,
that gives rise to conscious experience?
We all believe, none of us believe anymore
in the old story, it's a soul, right?
That used to be the most common explanation
that most people accept.
And still a lot of people today believe,
well, there's God and doubt,
only us with a special thing that animals don't have.
Rene Descartes famously said,
"A dog, if you hit it with your carriage,
"may yelp, may cry,
"but it doesn't have this special thing.
"It doesn't have the magic sauce."
- Soul, yeah.
- It doesn't have res cogitans, a soul.
Now we believe that isn't the case anymore.
So what is the difference between brains
and these guys, silicon?
And in particular, once their behavior matches,
so if you have Siri or Alexa in 20 years from now
that she can talk just as good as any possible human,
what grounds do you have to say she's not conscious?
In particular, if she says,
it's of course she will.
Well, of course I'm conscious.
You ask her, how are you doing?
And she'll say, well, you know,
they'll generate some way to, of course.
She'll behave like a person.
Now there's several differences.
One is, so this relates to the problem,
the very hard, why is consciousness a hard problem?
It's because it's subjective, right?
Only I have it, only I know,
I have direct experience of my own consciousness.
I don't have experience of your consciousness.
Now I assume as a sort of a Bayesian person
who believes in probability theory and all of that,
you know, I can do an abduction to the best available facts.
I deduce your brain is very similar to mine.
If I put you in a scanner,
your brain is roughly gonna behave the same way as I do.
If I gave you this muesli and asked you,
how does it taste?
You'd tell me things that I would also say, more or less.
So I infer based on all of that that you're conscious.
Now with Siri, I can't do that.
So there I really need a theory that tells me
what is it about any system, this or this,
that makes it conscious.
We have such a theory.
- Yes, so the integrated information theory.
But let me first, maybe as introduction
for people who are not familiar, Descartes.
Can you, you talk a lot about panpsychism.
Can you describe what physicalism versus dualism,
this, you mentioned the soul.
What is the history of that idea?
- The idea of panpsychism?
- Oh no, the debate, really,
out of which panpsychism can emerge
of dualism versus physicalism.
Or do you not see panpsychism as fitting into that?
- No, you can argue there's some,
well, okay, so let's step back.
So panpsychism is a very ancient belief
that's been around, I mean, Plato and Aristotle
talks about it, modern philosophers talk about it.
Of course, in Buddhism, the idea is very prevalent
that, I mean, there are different versions of it.
One version says everything is in souls, everything.
Rocks and stones and dogs and people
and forests and iPhones, all of us are all.
All matter is in soul, that's sort of one version.
Another version is that all biology,
all creatures, small or large,
from a single cell to a giant sequoia tree
feel like something.
That's one I think is somewhat more realistic.
So the different versions--
- What do you mean by feel like something?
- Have--
- Have feeling, have some kind of experience.
- It feels like something, it may well be possible
that it feels like something to be a paramecium.
I think it's pretty likely it feels like something
to be a bee or a mouse or a dog.
- Sure, so, okay.
- So that you can say that's also,
so panpsychism is very broad, right?
And you can, so some people, for example, Bertrand Russell,
tried to advocate this idea, it's called Russellian monism,
that panpsychism is really physics viewed from the inside.
So the idea is that physics is very good
at describing relationship among objects,
like charges or like gravity, right?
You know, it describes the relationship
between curvature and mass distribution, okay?
That's the relationship among things.
Physics doesn't really describe the ultimate reality itself.
It's just relationship among, you know,
quarks or all these other stuff--
- Almost from like a third person observer.
- Yes, yes, yes.
And consciousness is what physics feels from the inside.
So my conscious experience,
it's the way the physics of my brain,
particularly my cortex, feels from the inside.
And so if you are paramecium, you gotta remember,
you say paramecium, well, that's a pretty dumb creature.
It is, but it has already a billion different molecules,
probably, you know, 5,000 different proteins
assembled in a highly, highly complex system
that no single person, no computer system so far
on this planet has ever managed to accurately simulate.
Its complexity vastly escapes us.
Yes, and it may well be that that little thing
feels like a tiny bit.
Now, it doesn't have a voice in the head like me.
It doesn't have expectations.
You know, it doesn't have all that complex things,
but it may well feel like something.
- Yeah, so this is really interesting.
Can we draw some lines and maybe try to understand
the difference between life, intelligence,
and consciousness?
How do you see all of those?
If you had to define what is a living thing,
what is a conscious thing,
and what is an intelligent thing,
do those intermix for you,
or are they totally separate?
- Okay, so A, that's a question
that we don't have a full answer to.
- A lot of the stuff we're talking about today
is full of mysteries and fascinating ones, right?
- Well, for example, you can go to Aristotle,
who's probably the most important scientist
and philosopher who's ever lived
in certain Western culture.
He had this idea, it's called hylomorphism,
it's quite popular these days,
that there are different forms of soul.
The soul is really the form of something.
He says, "All biological creatures have a vegetative soul."
That's life principle.
Today, we think we understand something more
that it's biochemistry and nonlinear thermodynamics.
Then he says, "They have a sensitive soul."
Only animals and humans have also a sensitive soul,
or a petitive soul.
They can see, they can smell, and they have drives.
They wanna reproduce, they wanna eat, et cetera.
And then only humans have what he called a rational soul.
Okay? - Right.
§04Transcript · 23:25–31:26
- And that idea then made it into Christendom,
and then the rational soul is the one that lives forever.
He was very unclear.
He wasn't really, I mean, different readings
of Aristotle give different,
whether did he believe that rational soul
was immortal or not?
I probably think he didn't.
But then, of course, that made it into,
through Plato into Christianity,
and then the soul became immortal,
and then became the connection to God.
Now, so you ask me, essentially,
what is our modern conception of these three,
Aristotle would have called them different forms.
Life, we think we know something about it,
at least life on this planet, right?
Although we don't understand how it originated,
but it's been difficult to rigorously pin down.
You see this in modern definitions of death.
It's in fact, right now, there's a conference ongoing,
again, that tries to define legally and medically
what is death.
It used to be very simple.
Death is you stop breathing, your heart stops beating.
You're dead, right? - Yeah.
- Totally unconventional.
If you're unsure, you wait another 10 minutes.
If the patient doesn't breathe, you know, he's dead.
Well, now we have ventilators, we have pacemakers.
So it's much more difficult to define what death is.
Typically, death is defined as the end of life,
and life is defined before death.
- It's the thing before death.
- Okay, so we don't have really very good definitions.
Intelligence, we don't have a rigorous definition.
We know something how to measure.
It's called IQ or G-factors, right?
And we're beginning to build it in a narrow sense, right?
Like Go, AlphaGo, and Watson, and Google Cars,
and Uber cars, and all of that.
That's still narrow AI,
and some people are thinking
about artificial general intelligence.
But roughly, as we said before,
it's something to do with the ability to learn
and to adapt to new environments.
But that is, as I said also,
it's radical difference from experience.
And it's very unclear, if you build a machine that has AGI,
it's not at all, a priori, it's not at all clear
that this machine will have consciousness.
It may or may not.
- So let's ask it the other way.
Do you think, if you were to try to build
an artificial general intelligence system,
do you think figuring out how to build
artificial consciousness would help you get to an AGI?
So, or put another way,
do you think intelligent requires consciousness?
- In human, it goes hand in hand.
In human, or I think in biology,
consciousness, intelligence goes hand in hand.
Quay is illusion, because the brain evolved
to be highly complex.
Complexity, via the theory,
integrated information theory,
is sort of ultimately is what is closely tied
to consciousness.
Ultimately, it's causal power upon itself.
And so in evolved systems, they go together.
In artificial system, particularly in digital machines,
they do not go together.
And if you ask me, point blank,
is Alexa 20.0 in the year 2040,
when she can easily pass every Turing test,
is she conscious?
No.
Even if she claims she's conscious.
In fact, you could even do a more radical version
of this thought experiment.
We can build a computer simulation of the human brain.
You know what Henry Markham in the Blue Brain Project,
or the Human Brain Project in Switzerland is trying to do.
Let's grant him all the success.
So in 10 years, we have this perfect simulation
of the human brain.
Every neuron is simulated.
And it has alarmics, and it has motor neurons,
it has a broadcast area, and of course it'll talk,
and it'll say, "Hi, I just woke up, I feel great."
Okay?
Even that computer simulation that can in principle
map onto your brain will not be conscious.
Why?
Because it simulates.
It's a difference between the simulated and the real.
So it simulates the behavior associated with consciousness.
It might be, it will if it's done properly,
will have all the intelligence that that particular person
they're simulating has.
But simulating intelligence is not the same
as having conscious experiences.
And I give you a really nice metaphor
that engineers and physicists typically get.
I can write down Einstein's field equation,
nine or 10 equations that describe the link
in general relativity between curvature and mass.
I can do that, I can run this on my laptop
to predict that the center, the black hole
at the center of our galaxy will be so massive
that it'll twist space-time around it
so no light can escape.
It's a black hole, right?
But funny, have you ever wondered
why doesn't this computer simulation suck me in?
Right?
It simulates gravity, but it doesn't have
the causal power of gravity.
That's a huge difference.
So it's a difference between the real and the simulator,
just like it doesn't get wet inside a computer
when the computer runs cold, that simulates a weather storm.
And so in order to have artificial consciousness,
you have to give it the same causal power
as a human brain.
You have to build so-called a neuromorphic machine
that has hardware that is very similar to the human brain,
not a digital clocked phenomenon computer.
- So that's, just to clarify though,
you think that consciousness is not required
to create human level intelligence.
It seems to accompany in the human brain,
but for a machine, not.
- That's correct.
- So maybe just because this is AGI,
let's dig in a little bit about
what we mean by intelligence.
So one thing is the G factor,
these kind of IQ tests of intelligence.
But I think if you, maybe another way to say,
so in 2040, 2050, people will have Siri
that is just really impressive.
Do you think people will say Siri is intelligent?
- Yes.
- Intelligence is this amorphous thing.
So to be intelligent, it seems like you have to have
some kind of connections with other human beings,
in the sense that you have to impress them
with your intelligence.
And there feels, you have to somehow operate
in this world full of humans.
And for that, there feels like there has to be
something like consciousness.
So you think you can have just the world's best
natural NLP system, natural language understanding
and generation, and that will be,
that will get us happy and say,
you know what, we've created an AGI.
- I don't know happy, but yes, I do believe
we can get what we call high level functional intelligence,
in particular sort of the G, you know,
this fluid-like intelligence that we cherish,
particularly at a place like MIT, right?
In machines, I see a priori no reasons,
and I see a lot of reason to believe it's gonna happen
very, you know, over the next 50 years or 30 years.
- So for beneficial AI, for creating an AI system
that's, so you mentioned ethics,
that is exceptionally intelligent,
but also does not do, does, you know,
aligns its values with our values as humanity.
Do you think then it needs consciousness?
- Yes, I think that that is a very good argument
that if we're concerned about AI and the threat of AI,
a la Nick Bostrom, existentialist threat,
I think having an intelligence that has empathy, right?
Why do we find abusing a dog,
why do most of us find that abhorrent, abusing any animal?
Why do we find that abhorrent?
Because we have this thing called empathy,
which if you look at the Greek,
really means feeling with, I feel a path of empathy.
I have feeling with you.
I see somebody else suffer that isn't even my conspecific,
it's not a person, it's not a love,
it's not my wife or my kids, it's a dog,
but I feel naturally most of us, not all of us,
most of us will feel emphatic.
And so it may well be in the long-term interest
of survival of homo sapiens sapiens,
that if we do build AGI and it's really becomes very powerful
that it has an emphatic response
and doesn't just exterminate humanity.
- So as part of the full conscious experience,
to create a consciousness, artificial,
or in our human consciousness, do you think fear,
maybe we're gonna get into your earlier days,
with Nietzsche and so on, but do you think fear
and suffering are essential to have consciousness?
§05Transcript · 31:30–39:30
Do you have to have the full range of experience
to have a system that has experience?
Or can you have a system that only has
very particular kinds of very positive experiences?
- Look, you can have, in principle,
people have done this in the rat,
where you implant an electrode in the hypothalamus,
the pleasure center of the rat,
and the rat stimulates itself above and beyond anything else.
It doesn't care about food or natural sex
or drink anymore, it just stimulates itself
because it's such a pleasurable feeling.
I guess it's like an orgasm, just you have all day long.
And so, a priori, I see no reason why you need
a great variety.
Now, clearly, to survive, that wouldn't work, right?
But if I'd engineered artificially,
I don't think you need a great variety
of conscious experience.
You could have just pleasure or just fear.
It might be a terrible existence,
but I think that's possible,
at least on conceptual, logical ground.
Because any real creature, whether artificial or engineered,
you wanna give it fear, the fear of extinction,
that we all have, and you also wanna give it
positive, repetitive states,
states that you want the machine encouraged to do
because they give the machine positive feedback.
- So, you mentioned panpsychism,
to jump back a little bit,
everything having some kind of mental property.
How do you go from there to something
like human consciousness?
So, everything having some elements of consciousness.
Is there something special about human consciousness?
- So, just, it's not everything.
Like a spoon, there's no,
the form of panpsychism I think about
doesn't ascribe consciousness to anything,
like this, the spoon or my liver.
However, it is, the theory,
the integrated information theory,
does say that system, even ones that look
from the outside relatively simple,
at least if they have this internal causal power,
it does feel like something.
The theory, I probably doesn't say anything
what's special about human.
Biologically, we know what the one thing
that's special about human is we speak,
and we have an overblown sense of our own importance.
- Right.
- We believe we're exceptional,
and we're just God's gift to the universe.
But behaviorally, the main thing that we have,
we can plan over the long term,
we have language, and that gives us
enormous amount of power,
and that's why we are the current
dominant species on the planet.
- So you mentioned God,
you grew up a devout Roman Catholic,
in a Roman Catholic family.
So, you know, with consciousness,
you're sort of exploring some really deeply
fundamental human things that religion also touches on.
So where does religion fit into your
thinking about consciousness?
And you've grown throughout your life
and changed your views on religion,
as far as I understand.
- Yeah, I mean, I'm now much closer to,
so I'm not a Roman Catholic anymore,
I don't believe there's sort of this God,
the God I was educated to believe in,
you know, sits somewhere in the fullness of time,
I'll be united in some sort of everlasting bliss,
I just don't see any evidence for that.
Look, the world, the night is large and full of wonders.
(laughing)
There are many things that I don't understand,
I think many things that we as a cult,
look, we don't even understand more than 4%
of all the universe, right?
Dark matter, dark energy, we have no idea what it is,
maybe it's lost socks, what do I know?
So, all I can tell you is,
it's sort of my current religious or spiritual sentiment
is much closer to some form of Buddhism.
- Can you describe--
- Without the reincarnation, unfortunately,
there's no evidence for a reincarnation.
- So can you describe the way Buddhism
sees the world a little bit?
- Well, so, they talk about,
so when I spent several meetings with the Dalai Lama,
and what always impressed me about him,
he really, unlike, for example, let's say,
the Pope or some cardinal, he always emphasized
minimizing the suffering of all creatures.
So they have this, from the early beginning,
they look at suffering in all creatures,
not just in people, but in everybody, this universal,
and of course, by degrees, right?
An animal, generally, is less capable of suffering
than a well-developed, normally developed human,
and they think consciousness pervades in this universe,
and they have these techniques, you know,
you can think of them like mindfulness, et cetera,
and meditation that tries to access sort of
what they claim of this more fundamental aspect of reality.
I'm not sure it's more fundamental,
as I think about it, there's the physical,
and then there's this inside view, consciousness,
and those are the two aspects.
That's the only thing I have access to in my life,
and you've gotta remember, my conscious experience
and your conscious experience comes prior
to anything you know about physics,
comes prior to knowledge about the universe,
and atoms, and super strings, and molecules,
and all of that.
The only thing you directly are acquainted with
is this world that's populated with things,
and images, and sounds in your head,
and touches, and all of that.
- I actually have a question.
So it sounds like you kind of have a rich life.
You talk about rock climbing,
and it seems like you really love literature,
and consciousness is all about experiencing things.
So do you think that has helped your research on this topic?
- Yes, particularly if you think about it,
the various states, so for example,
when you do rock climbing,
or now I do rowing, crew rowing, and I bike every day,
you can get into this thing called the zone,
and I've always, I wonder about it,
particularly with respect to consciousness,
because it's a strangely addictive state.
You wanna, I mean, once people have it once,
they wanna keep on going back to it,
and you wonder, what is it so addicting about it?
And I think it's the experience
of almost close to pure experience,
because in this zone,
you're not conscious of your inner voice anymore.
There's always this inner voice nagging you.
You have to do this, you have to do that,
you have to pay your taxes,
you have this fight with your ex,
and all of those things are always there.
But when you're in the zone, all of that is gone,
and you're just in this wonderful state
where you're fully out in the world, right?
You're climbing, or you're rowing, or biking,
or doing soccer, or whatever you're doing,
and sort of consciousness sort of is this,
you're all action, or in this case, of pure experience.
You're not action at all,
but in both cases, you experience some aspect of,
you touch some basic part of conscious existence
that is so basic and so deeply satisfying.
I think you touch the root of being.
That's really what you're touching there.
You're getting close to the root of being,
and that's very different from intelligence.
- So what do you think about
the simulation hypothesis, simulation theory,
the idea that we all live in a computer simulation?
Have you given it a shot? - Rapture for nerds.
Raptures for nerds.
- Raptures for nerds? (laughs)
- I think it's as likely as the hypothesis
that engaged hundreds of scholars for many centuries,
are we all just existing in the mind of God?
- Right. - And this is just
a modern version of it.
It's equally plausible.
People love talking about these sorts of things.
I know they're book written about the simulation hypothesis.
If that's what people wanna do, that's fine.
Seems rather esoteric.
It's never testable.
- But it's not useful for you to think of in those terms.
So maybe connecting to the questions of free will,
which you've talked about.
I vaguely remember you saying
that the idea that there's no free will,
it makes you very uncomfortable.
So what do you think about free will
from a physics perspective, from a consciousness perspective?
What does it all fit?
- Okay, so from the physics perspective,
leaving aside quantum mechanics,
we believe we live in a fully deterministic world, right?
But then comes, of course, quantum mechanics.
So now we know that certain things
are in principle not predictable,
§06Transcript · 39:32–46:49
which, as you said, I prefer
because the idea that the initial condition of the universe
and then everything else, we're just acting out
the initial condition of the universe
that doesn't--
- It's not a romantic notion.
- Certainly not, right?
Now when it comes to consciousness,
I think we do have certain freedom.
We are much more constrained by physics, of course,
and by our past and by our own conscious desires
and what our parents told us
and what our environment tells us.
We all know that, right?
There's hundreds of experiments
that show how we can be influenced.
But finally, in the final analysis,
when you make a life,
and I'm talking really about critical decisions,
what you really think.
Should I marry?
Should I go to this school or that school?
Should I take this job or that job?
Should I cheat on my taxes or not?
(Richard laughs)
These are things where you really deliberate.
And I think under those conditions,
you are as free as you can be.
When you bring your entire being,
your entire conscious being to that question
and try to analyze it under all the various conditions,
then you make a decision.
You are as free as you can ever be.
That is, I think, what free will is.
It's not a will that's totally free to do anything it wants.
That's not possible.
- Right.
So as Jack mentioned,
you actually write a blog about books you've read.
Amazing books from, I'm Russian,
from Bulgakov to-
- Ah, my son, my Rita.
- Yeah.
Neil Gaiman, Carl Sagan, Murakami.
So what is a book that early in your life
transformed the way you saw the world?
Something that changed your life?
- Nietzsche, I guess, did.
Das Bruch der Truste,
because he talks about some of these problems.
You know, he was one of the first discoverer
of the unconscious.
This is a little bit before Freud,
when it was in the air.
He makes all these claims that people
sort of under the guise of, under the mask of charity,
actually are very non-charitable.
So he's sort of really the first discoverer
of the great land of the unconscious.
And that really struck me.
- And what do you think about the unconscious?
What do you think about Freud?
What do you think about these ideas?
What's, just like dark matter in the universe,
what's over there in that unconscious?
- A lot.
I mean, much more than we think.
This is what a lot of last hundred years of research
has shown.
So I think he was a genius, misguided towards the end,
but he started out as a neuroscientist.
He contributed.
He did studies on the Lamprey.
He contributed himself to the neuron hypothesis,
the idea that there are discrete units
that we call nerve cells now.
And then he wrote about the unconscious.
And I think it's true.
There's lots of stuff happening.
You feel this particular, when you're in a relationship
and it breaks asunder, right?
And then you have this terrible,
you can have love and hate and lust and anger
and all of it's mixed in.
And when you try to analyze yourself,
why am I so upset?
It's very, very difficult to penetrate to those basements,
those caverns in your mind,
because the prying eyes of conscience
doesn't have access to those.
But they're there in the amygdala,
or in lots of other places,
they make you upset or angry or sad or depressed.
And it's very difficult to try
to actually uncover the reason.
You can go to a shrink,
you can talk with your friend endlessly.
You construct finally a story why this happened,
why you love her or don't love her or whatever,
but you don't really know whether that actually happened,
because you simply don't have access
to those parts of the brain,
and they're very powerful.
- Do you think that's a feature or a bug of our brain?
The fact that we have this deep,
difficult to dive into subconscious?
- I think it's a feature, because otherwise,
look, we are, like any other brain
or nervous system or computer,
we are severely band-limited.
If everything I do, every emotion I feel,
every eye movement I make,
if all of that had to be
under the control of consciousness,
I wouldn't be here.
So what you do early on, your brain,
you have to be conscious when you learn things
like typing or like riding on a bike.
But then what you do, you train up,
routes I think that involve basal ganglia and striatum,
you train up different parts of your brain.
And then once you do it automatically like typing,
you can show you do it much faster
without even thinking about it,
because you've got these highly specialized,
what Franz Krick and I call zombie agents,
that are taking care of that,
while your consciousness can sort of worry
about the abstract sense of the text you wanna write.
And I think that's true for many, many things.
- But for the things like all the fights you had
with an ex-girlfriend, things that you would think
are not useful to still linger somewhere
in the subconscious.
So that seems like a bug that it would stay there.
- You think it would be better if you can analyze it
and then get it out of the system.
- And get it out of the system,
or just forget it ever happened.
You know, that seems a very buggy kind of--
- Well, yeah, in general, we don't have,
and that's probably functional,
we don't have an ability, unless it's extreme,
there are cases, clinical dissociations, right?
When people are heavily abused,
when they completely repress the memory,
but that doesn't happen in normal people.
We don't have an ability to remove traumatic memories.
And of course, we suffer from that.
On the other hand, probably if you had the ability
to constantly wipe your memory,
you probably do it to an extent that isn't useful to you.
So.
- Yeah, it's a good question.
- It's a balance.
So on the books, as Jack mentioned,
correct me if I'm wrong, but broadly speaking,
academia and the different scientific disciplines,
certainly in engineering,
reading literature seems to be a rare pursuit.
Perhaps I'm wrong on this, but that's in my experience.
Most people read much more technical texts
and do not sort of escape or seek truth in literature.
It seems like you do.
So what do you think is the value?
What do you think literature adds
to the pursuit of scientific truth?
Do you think it's good, it's useful for--
- I think it's the access to a much wider array
of human experiences.
- How valuable do you think it is?
- Well, if you wanna understand human nature
and nature in general,
then I think you have to better understand
a wide variety of experiences,
not just sitting in a lab, staring at a screen
and having a face flashed onto you
for a hundred milliseconds and pushing a button.
That's what I used to do.
That's what most psychologists do.
There's nothing wrong with that,
but you need to consider lots of other strange states.
- And literature is a shortcut for this.
- Well, yeah, because literature,
that's what literature is all about,
all sorts of interesting experiences that people have,
the contingency of it,
the fact that women experience the world different,
black people experience the world different.
And one way to experience that
is reading all these different literature
and try to find out.
You see everything is so relative, right?
You read a book 300 years ago,
they thought about certain problems
very, very differently than us today.
We today, like any culture, think we know it all.
That's common to every culture.
Every culture believes that it's heyday, they know it all.
And then you realize, well,
§07Transcript · 46:50–54:34
there's other ways of viewing the universe,
and some of them may have lots of things in their favor.
- So this is a question I wanted to ask
about time scale or scale in general.
When you, with IIT or in general,
try to think about consciousness,
try to think about these ideas,
we kind of naturally think in human time scales.
Or, and also entities that are sized close to humans.
Do you think of things that are much larger,
much smaller as containing consciousness?
And do you think of things that take, you know,
well, ages, eons, to operate
in their conscious cause effect?
- It's a very good question.
So I think a lot about small creatures
because experimentally, you know,
a lot of people work on flies and bees.
So, and most people just think they are automata.
They're just bugs for heaven's sake, right?
But if you look at their behavior,
like bees, they can recognize individual humans.
They have this very complicated way to communicate.
If you've ever been involved,
or you know your parents when they bought a house,
what sort of agonizing decision that is.
And bees have to do that once a year, right?
When they swarm in the spring.
And then they have this very elaborate way.
They have three night scouts.
They go to the individual sites.
They come back, they have this power, this dance,
literally where they dance for several days.
They try to recruit other bees.
It's very complicated decision way.
When they finally want to make a decision,
the entire swarm, the scouts warm up the entire swarm,
then go to one location.
They don't go to 50 locations.
They go to one location that the scouts
have agreed upon by themselves.
That's awesome.
If we look at the circuit complexity,
it's 10 times more denser than anything we have in our brain.
Now they only have a million neurons,
but the neurons are amazingly complex.
Complex behavior, very complicated circuitry.
So there's no question they experience something.
Their life is very different.
They're tiny.
They only live, you know, for,
for while workers live maybe for two months.
So I think, and IIT tells you this,
in principle, the substrate of consciousness
is the substrate that maximizes the cause effect power
over all possible spatial temple grains.
So when I think about, for example,
do you know the science fiction story, The Black Cloud?
Okay, it's a classic by Fred Hoyle, the astronomer.
He has this cloud intervening between the earth
and the sun and leading to some sort of,
to global cooling.
This is written in the 50s.
It turns out you can, using the radio dish,
they communicate with actually an entity.
It's actually an intelligent entity.
And they sort of, they convince it to move away.
So here you have a radical different entity.
And in principle, IIT says, well,
you can measure the integrated information
in principle at least.
And yes, if that, if the maximum of that occurs
at a timescale of months,
rather than in assets sort of fraction of a second,
yes, then they would experience life
where each moment is a month rather than,
or microsecond, right, rather than a fraction
of a second in the human case.
And so there may be forms of consciousness
that we simply don't recognize for what they are
because they are so radical different
from anything you and I are used to.
Again, that's why it's good to read
or to watch science fiction movie
in order to think about this.
Like this is, do you know Stanislav Lem,
this Polish science fiction writer?
He wrote Solaris.
It was turned into a Hollywood movie.
- Yes.
- His best novel, so it was in the '60s,
very, very engineering, he's an engineering background.
His most interesting novel is called The Victorious
where human civilization,
they have this mission to this planet
and everything is destroyed and they discover machines,
humans got killed and then these machines took over
and there was a machine evolution,
a Darwinian evolution, he talks about this very vividly.
And finally, the dominant,
the dominant machine intelligence organism that survived
were gigantic clouds of little hexagonal
universal cellular automata.
This was written in the '60s.
So typically they're all lying on the ground
individual by themselves, but in times of crisis,
they can communicate, they assemble into gigantic nets,
into clouds of trillions of these particles
and then they become hyperintelligent
and they can beat anything that humans can throw at it.
It's a very beautiful and compelling way
to have an intelligence where finally the humans
leave the planet, they simply are unable to understand
and comprehend this creature.
And they can say, well, either we can nuke
the entire planet and destroy it
or we just have to leave because fundamentally
it's an alien, it's so alien from us and our ideas
that we cannot communicate with them.
- Yeah, actually in conversation, cellular automatas,
Stephen Wolfram brought up is that there could be
his ideas that you already have
these artificial general intelligence,
like super smart or maybe conscious beings
in the cellular automata, we just don't know
how to talk to them.
So it's the language of communication
'cause you don't know what to do with it.
So that's one sort of view is consciousness
is only something you can measure.
So it's not conscious if you can't measure it.
- So you're making an ontological
and an epistemic statement.
One is it's just like seeing the multiverses,
that might be true, but I can't communicate with them.
I can't have any knowledge of them.
That's an epistemic argument.
Those are two different things.
So it may well be possible.
Look, another case that's happening right now,
people are building these mini organoids.
Do you know about this?
So you can take stem cells from under your arm,
put it in a dish, add four transcription factors
and then you can induce them to grow into large,
well large, they're a few millimeter,
they're like a half a million neurons
that look like nerve cells in a dish called mini organoids.
At Harvard, at Stanford, everywhere they're building them.
It may well be possible that they're beginning
to feel like something,
but we can't really communicate with them right now.
So people are beginning to think about the ethics of this.
So yes, he may be perfectly right,
but it's one question, are they conscious or not,
is totally separate question, how would I know?
Those are two different things.
- If you could give advice to a young researcher
sort of dreaming of understanding
or creating human level intelligence or consciousness,
what would you say?
- Just follow your dreams, read widely.
- No, I mean, I suppose what discipline,
what is the pursuit that they should take on?
Is it neuroscience,
is it computational cognitive science,
is it philosophy, is it computer science or robotics?
- No, in a sense that, okay, so the only known system
that have high level of intelligence is Homo sapiens.
So if you wanted to build it,
it's probably good to continue to study closely
what humans do, so cognitive neuroscience,
somewhere between cognitive neuroscience on the one hand,
then some philosophy of mind,
and then AI, computer science.
You can look at all the original ideas, neural networks,
they all came from neuroscience, right?
Reinforcing, whether it's Snarky, Minsky building Snarky,
or whether it's the early Schubert and Wiesel experiments
at Harvard that then gave rise to networks
and then multi-layer networks.
So it may well be possible, in fact,
some people argue that to make the next big step in AI,
once we realize the limits of deep convolutional networks,
they can do certain things,
but they can't really understand.
I can't really show 'em one image.
I can show you a single image of somebody,
a pickpocket who steals a wallet from a purse.
You immediately know that's a pickpocket.
Now, computer system would just say,
well, it's a man, it's a woman, it's a purse, right?
Unless you train this machine
on showing it 100,000 pickpockets, right?
So it doesn't have this easy understanding
that you have, right?
So some people make the argument,
in order to go to the next step,
where you really wanna build machines that understand,
in a way you and I, we have to go to psychology.
§08Transcript · 54:37–57:46
We need to understand how we do it
and how our brains enable us to do it.
And so therefore, being on the cusp,
it's also so exciting to try to understand better our nature
and then to build, to take some of those insight
and build them.
So I think the most exciting thing
is somewhere in the interface between cognitive science,
neuroscience, AI, computer science, and philosophy of mind.
- Beautiful, yeah, I'd say if there is,
from the machine learning, from the computer science,
computer vision perspective,
many of the research is kind of ignore
the way the human brain works.
- Yes. - Ignore even psychology
or literature or studying the brain.
I would hope Josh Tenenbaum talks about
bringing that in more and more.
And that's, yeah.
So you've worked on some amazing stuff
throughout your life.
What's the thing that you're really excited about?
What's the mystery that you would love to uncover
in the near term,
beyond all the mysteries that you're already surrounded by?
- Well, so there's a structure called the claustrum.
- Okay. - This is a structure,
it's underneath our cortex, it's yay big.
You have one on the left, one on the right,
underneath this part, underneath the insula.
It's very thin, it's like one millimeter.
It's embedded in wiring, in white matter,
so it's very difficult to image.
And it has connection to every cortical region.
And Francis Crick, the last paper he ever wrote,
he dictated corrections the day he died in hospital
on this paper.
He and I, we hypothesized, well,
'cause it has this unique anatomy,
it gets input from every cortical area
and projects back to every cortical area.
That the function of this structure is similar,
it's just a metaphor,
to the role of a conductor in a symphony orchestra.
You have all the different cortical players.
You have some that do motion,
some that do theory of mind,
some that infer social interaction,
and color, and hearing,
and all the different modules in cortex.
But of course, what consciousness is,
consciousness puts it all together into one package,
the binding problem, all of that.
And this is really the function,
because it has relatively few neurons compared to cortex,
but it talks, it sort of receives input from all of them,
and it projects back to all of them.
And so we are testing that right now.
We've got this beautiful neuronal reconstruction
in the mouse called crown of thorn neurons
that are in the claustrum,
that have the most widespread connection
of any neuron I've ever seen.
They're very, you have individual neurons
that sit in the claustrum, tiny,
but then they have this, single neurons
have this huge axonal tree that cover
both ipsi and contralateral cortex,
and trying to turn, using fancy tools like optogenetics,
trying to turn those neurons on or off
and study it, what happens in the mouse.
- So this thing is perhaps where the parts become the whole.
It integrates--
- Perhaps, it's one of the structures,
it's a very good way of putting it,
where the individual parts turn into the whole
of the conscious experience.
- Well, with that, thank you very much for being here today.
- Thank you very much. - It's wonderful
to have you back at MIT. - Excellent, Jack.
Thank you very much.
(upbeat music)
(upbeat music)
(upbeat music)
(upbeat music)