back to index

Karl Friston: Neuroscience and the Free Energy Principle | Lex Fridman Podcast #99


small model | large model

link |
00:00:00.000
The following is a conversation with Carl Friston,
link |
00:00:03.080
one of the greatest neuroscientists in history.
link |
00:00:06.960
Cited over 245,000 times,
link |
00:00:10.320
known for many influential ideas in brain imaging,
link |
00:00:13.580
neuroscience, and theoretical neurobiology,
link |
00:00:16.520
including especially the fascinating idea
link |
00:00:19.820
of the free energy principle for action and perception.
link |
00:00:24.200
Carl's mix of humor, brilliance, and kindness,
link |
00:00:28.040
to me, are inspiring and captivating.
link |
00:00:31.160
This was a huge honor and a pleasure.
link |
00:00:34.120
This is the Artificial Intelligence Podcast.
link |
00:00:36.800
If you enjoy it, subscribe on YouTube,
link |
00:00:38.920
review it with five stars on Apple Podcast,
link |
00:00:41.280
support it on Patreon,
link |
00:00:42.560
or simply connect with me on Twitter,
link |
00:00:44.560
at Lex Friedman, spelled F R I D M A N.
link |
00:00:48.100
As usual, I'll do a few minutes of ads now,
link |
00:00:50.480
and never any ads in the middle
link |
00:00:52.040
that can break the flow of the conversation.
link |
00:00:54.220
I hope that works for you,
link |
00:00:55.540
and doesn't hurt the listening experience.
link |
00:00:58.280
This show is presented by Cash App,
link |
00:01:00.880
the number one finance app in the App Store.
link |
00:01:03.200
When you get it, use code LEXPODCAST.
link |
00:01:06.280
Cash App lets you send money to friends by Bitcoin,
link |
00:01:09.400
and invest in the stock market with as little as $1.
link |
00:01:12.860
Since Cash App allows you to send
link |
00:01:14.480
and receive money digitally,
link |
00:01:16.200
let me mention a surprising fact related to physical money.
link |
00:01:20.200
Of all the currency in the world,
link |
00:01:22.000
roughly 8% of it is actual physical money.
link |
00:01:25.640
The other 92% of money only exists digitally.
link |
00:01:29.920
So again, if you get Cash App from the App Store,
link |
00:01:32.260
Google Play, and use the code LEXPODCAST, you get $10,
link |
00:01:37.040
and Cash App will also donate $10 to FIRST,
link |
00:01:39.880
an organization that is helping to advance robotics
link |
00:01:42.640
and STEM education for young people around the world.
link |
00:01:45.780
And now, here's my conversation with Carl Friston.
link |
00:01:50.020
How much of the human brain do we understand
link |
00:01:53.380
from the low level of neuronal communication
link |
00:01:56.580
to the functional level to the highest level,
link |
00:02:01.660
maybe the psychiatric disorder level?
link |
00:02:04.940
Well, we're certainly in a better position
link |
00:02:06.420
than we were last century.
link |
00:02:08.980
How far we've got to go, I think,
link |
00:02:10.440
is almost an unanswerable question.
link |
00:02:13.520
So you'd have to set the parameters,
link |
00:02:16.040
you know, what constitutes understanding, what level
link |
00:02:20.140
of understanding do you want?
link |
00:02:21.740
I think we've made enormous progress
link |
00:02:25.380
in terms of broad brush principles.
link |
00:02:29.140
Whether that affords a detailed cartography
link |
00:02:32.460
of the functional anatomy of the brain and what it does,
link |
00:02:35.380
right down to the microcircuitry and the neurons,
link |
00:02:38.140
that's probably out of reach at the present time.
link |
00:02:42.220
So the cartography, so mapping the brain,
link |
00:02:44.900
do you think mapping of the brain,
link |
00:02:47.440
the detailed, perfect imaging of it,
link |
00:02:50.380
does that get us closer to understanding
link |
00:02:54.100
of the mind, of the brain?
link |
00:02:56.060
So how far does it get us if we have
link |
00:02:59.100
that perfect cartography of the brain?
link |
00:03:01.580
I think there are lower bounds on that.
link |
00:03:03.060
It's a really interesting question.
link |
00:03:06.340
And it would determine the sort of scientific career
link |
00:03:09.220
you'd pursue if you believe that knowing
link |
00:03:11.640
every dendritic connection, every sort of microscopic,
link |
00:03:16.020
synaptic structure right down to the molecular level
link |
00:03:18.740
was gonna give you the right kind of information
link |
00:03:22.260
to understand the computational anatomy,
link |
00:03:25.340
then you'd choose to be a microscopist
link |
00:03:27.180
and you would study little cubic millimeters of brain
link |
00:03:32.340
for the rest of your life.
link |
00:03:33.500
If on the other hand you were interested
link |
00:03:35.300
in holistic functions and a sort of functional anatomy
link |
00:03:40.300
of the sort that a neuropsychologist would understand,
link |
00:03:44.100
you'd study brain lesions and strokes,
link |
00:03:46.380
just looking at the whole person.
link |
00:03:48.460
So again, it comes back to at what level
link |
00:03:50.900
do you want understanding?
link |
00:03:52.980
I think there are principled reasons not to go too far.
link |
00:03:57.900
If you commit to a view of the brain
link |
00:04:01.660
as a machine that's performing a form of inference
link |
00:04:06.660
and representing things, that level of understanding
link |
00:04:15.980
is necessarily cast in terms of probability densities
link |
00:04:20.540
and ensemble densities, distributions.
link |
00:04:24.180
And what that tells you is that you don't really want
link |
00:04:27.120
to look at the atoms to understand the thermodynamics
link |
00:04:30.500
of probabilistic descriptions of how the brain works.
link |
00:04:34.420
So I personally wouldn't look at the molecules
link |
00:04:38.620
or indeed the single neurons in the same way
link |
00:04:41.700
if I wanted to understand the thermodynamics
link |
00:04:44.060
of some non equilibrium steady state of a gas
link |
00:04:47.220
or an active material, I wouldn't spend my life
link |
00:04:49.980
looking at the individual molecules
link |
00:04:54.200
that constitute that ensemble.
link |
00:04:55.420
I'd look at their collective behavior.
link |
00:04:57.500
On the other hand, if you go too coarse grain,
link |
00:05:00.140
you're gonna miss some basic canonical principles
link |
00:05:03.900
of connectivity and architectures.
link |
00:05:06.300
I'm thinking here this bit colloquial,
link |
00:05:10.020
but this current excitement about high field
link |
00:05:13.540
magnetic resonance imaging at seven Tesla, why?
link |
00:05:17.240
Well, it gives us for the first time the opportunity
link |
00:05:19.460
to look at the brain in action at the level
link |
00:05:22.740
of a few millimeters that distinguish
link |
00:05:24.700
between different layers of the cortex
link |
00:05:27.260
that may be very important in terms of evincing
link |
00:05:32.260
generic principles of conical microcircuitry
link |
00:05:35.260
that are replicated throughout the brain
link |
00:05:37.620
that may tell us something fundamental
link |
00:05:39.120
about message passing in the brain
link |
00:05:41.140
and these density dynamics or neuronal
link |
00:05:44.260
or some more population dynamics
link |
00:05:46.820
that underwrite our brain function.
link |
00:05:49.420
So somewhere between a millimeter and a meter.
link |
00:05:53.920
Lingering for a bit on the big questions if you allow me,
link |
00:05:58.340
what to use the most beautiful or surprising characteristic
link |
00:06:01.940
of the human brain?
link |
00:06:03.420
I think it's hierarchical and recursive aspect.
link |
00:06:06.780
It's recurrent aspect.
link |
00:06:08.680
Of the structure or of the actual
link |
00:06:10.820
representation of power of the brain?
link |
00:06:12.920
Well, I think one speaks to the other.
link |
00:06:15.380
I was actually answering in a dull minded way
link |
00:06:18.420
from the point of view of purely its anatomy
link |
00:06:20.380
and its structural aspects.
link |
00:06:22.980
I mean, there are many marvelous organs in the body.
link |
00:06:26.740
Let's take your liver for example.
link |
00:06:28.460
Without it, you wouldn't be around for very long
link |
00:06:32.660
and it does some beautiful and delicate biochemistry
link |
00:06:35.980
and homeostasis and evolved with a finesse
link |
00:06:41.140
that would easily parallel the brain
link |
00:06:43.020
but it doesn't have a beautiful anatomy.
link |
00:06:45.140
It has a simple anatomy which is attractive
link |
00:06:47.020
in a minimalist sense but it doesn't have
link |
00:06:48.940
that crafted structure of sparse connectivity
link |
00:06:52.820
and that recurrence and that specialization
link |
00:06:55.380
that the brain has.
link |
00:06:56.380
So you said a lot of interesting terms here.
link |
00:06:58.300
So the recurrence, the sparsity,
link |
00:07:00.700
but you also started by saying hierarchical.
link |
00:07:03.540
So I've never thought of our brain as hierarchical.
link |
00:07:11.260
Sort of I always thought it's just like a giant mess,
link |
00:07:14.300
interconnected mess where it's very difficult
link |
00:07:16.740
to figure anything out.
link |
00:07:18.180
But in what sense do you see the brain as hierarchical?
link |
00:07:21.860
Well, I see it, it's not a magic soup.
link |
00:07:24.220
Which of course is what I used to think
link |
00:07:28.700
before I studied medicine and the like.
link |
00:07:34.460
So a lot of those terms imply each other.
link |
00:07:38.740
So hierarchies, if you just think about
link |
00:07:41.700
the nature of a hierarchy,
link |
00:07:43.540
how would you actually build one?
link |
00:07:46.020
And what you would have to do is basically
link |
00:07:48.300
carefully remove the right connections
link |
00:07:51.140
that destroy the completely connected soups
link |
00:07:54.740
that you might have in mind.
link |
00:07:56.100
So a hierarchy is in and of itself defined
link |
00:08:00.140
by a sparse and particular connectivity structure.
link |
00:08:04.420
I'm not committing to any particular form of hierarchy.
link |
00:08:08.380
But your sense is there is some.
link |
00:08:10.140
Oh, absolutely, yeah.
link |
00:08:11.420
In virtue of the fact that there is a sparsity
link |
00:08:14.220
of connectivity, not necessarily of a qualitative sort,
link |
00:08:19.220
but certainly of a quantitative sort.
link |
00:08:20.460
So it is demonstrably so that the further apart
link |
00:08:27.460
two parts of the brain are,
link |
00:08:29.220
the less likely they are to be wired,
link |
00:08:32.260
to possess axonal processes, neuronal processes
link |
00:08:35.820
that directly communicate one message
link |
00:08:39.100
or messages from one part of that brain
link |
00:08:41.140
to the other part of the brain.
link |
00:08:42.660
So we know there's a sparse connectivity.
link |
00:08:45.820
And furthermore, on the basis of anatomical connectivity
link |
00:08:50.300
in traces studies, we know that that sparsity
link |
00:08:57.580
underwrites a hierarchical and very structured
link |
00:09:00.980
sort of connectivity that might be best understood
link |
00:09:07.140
like a little bit like an onion.
link |
00:09:09.380
There is a concentric, sometimes referred to as centripetal
link |
00:09:15.540
by people like Marcel Masulam,
link |
00:09:17.500
hierarchical organization to the brain.
link |
00:09:19.340
So you can think of the brain as in a rough sense,
link |
00:09:23.860
like an onion, and all the sensory information
link |
00:09:28.060
and all the afferent outgoing messages
link |
00:09:31.260
that supply commands to your muscles
link |
00:09:33.500
or to your secretory organs come from the surface.
link |
00:09:37.100
So there's a massive exchange interface
link |
00:09:40.140
with the world out there on the surface.
link |
00:09:43.020
And then underneath, there's a little layer
link |
00:09:45.140
that sits and looks at the exchange on the surface.
link |
00:09:49.380
And then underneath that, there's a layer
link |
00:09:51.380
right the way down to the very center,
link |
00:09:53.380
to the deepest part of the onion.
link |
00:09:55.740
That's what I mean by a hierarchical organization.
link |
00:09:58.700
There's a discernible structure defined
link |
00:10:02.020
by the sparsity of connections
link |
00:10:04.860
that lends the architecture a hierarchical structure
link |
00:10:08.620
that tells one a lot about the kinds of representations
link |
00:10:12.780
and messages.
link |
00:10:13.620
Coming back to your earlier question,
link |
00:10:15.620
is this about the representational capacity
link |
00:10:19.140
or is it about the anatomy?
link |
00:10:20.420
Well, one underwrites the other.
link |
00:10:24.580
If one simply thinks of the brain
link |
00:10:26.380
as a message passing machine,
link |
00:10:29.100
a process that is in the service of doing something,
link |
00:10:33.460
then the circuitry and the connectivity
link |
00:10:37.500
that shape that message passing also dictate its function.
link |
00:10:42.500
So you've done a lot of amazing work
link |
00:10:44.980
in a lot of directions.
link |
00:10:46.660
So let's look at one aspect of that,
link |
00:10:49.260
of looking into the brain
link |
00:10:51.060
and trying to study this onion structure.
link |
00:10:54.980
What can we learn about the brain by imaging it?
link |
00:10:57.660
Which is one way to sort of look at the anatomy of it.
link |
00:11:00.860
Broadly speaking, what are the methods of imaging,
link |
00:11:04.700
but even bigger, what can we learn about it?
link |
00:11:07.620
Right, so well, most human neuroimaging
link |
00:11:13.820
that you might see in science journals
link |
00:11:18.980
that speaks to the way the brain works,
link |
00:11:22.540
measures brain activity over time.
link |
00:11:24.340
So that's the first thing to say,
link |
00:11:26.020
that we're effectively looking at fluctuations
link |
00:11:30.020
in neuronal responses,
link |
00:11:32.580
usually in response to some sensory input
link |
00:11:36.140
or some instruction, some task.
link |
00:11:40.660
Not necessarily, there's a lot of interest
link |
00:11:42.260
in just looking at the brain
link |
00:11:44.740
in terms of resting state, endogenous,
link |
00:11:46.860
or intrinsic activity.
link |
00:11:48.940
But crucially, at every point,
link |
00:11:50.780
looking at these fluctuations,
link |
00:11:52.460
either induced or intrinsic in the neural activity,
link |
00:11:57.340
and understanding them at two levels.
link |
00:11:59.020
So normally, people would recourse
link |
00:12:03.540
to two principles of brain organization
link |
00:12:06.260
that are complementary.
link |
00:12:07.420
One, functional specialization or segregation.
link |
00:12:10.460
So what does that mean?
link |
00:12:11.860
It simply means that there are certain parts of the brain
link |
00:12:16.300
that may be specialized for certain kinds of processing.
link |
00:12:19.580
For example, visual motion,
link |
00:12:21.900
our ability to recognize or to perceive movement
link |
00:12:26.060
in the visual world.
link |
00:12:27.780
And furthermore, that specialized processing
link |
00:12:31.100
may be spatially or anatomically segregated,
link |
00:12:34.540
leading to functional segregation.
link |
00:12:37.060
Which means that if I were to compare your brain activity
link |
00:12:40.900
during a period of viewing a static image,
link |
00:12:45.220
and then compare that to the responses of fluctuations
link |
00:12:49.260
in the brain when you were exposed to a moving image,
link |
00:12:52.660
say a flying bird,
link |
00:12:54.900
we'd expect to see
link |
00:12:56.300
restricted, segregated differences in activity.
link |
00:13:01.620
And those are basically the hotspots
link |
00:13:03.780
that you see in the statistical parametric maps
link |
00:13:06.780
that test for the significance of the responses
link |
00:13:08.900
that are circumscribed.
link |
00:13:11.060
So now, basically, we're talking about
link |
00:13:13.740
some people have perhaps unkindly called a neocartography.
link |
00:13:19.300
This is a phrenology augmented by modern day neuroimaging,
link |
00:13:24.420
basically finding blobs or bumps on the brain
link |
00:13:28.980
that do this or do that,
link |
00:13:30.860
and trying to understand the cartography
link |
00:13:33.100
of that functional specialization.
link |
00:13:35.700
So how much is there such,
link |
00:13:38.860
this is such a beautiful sort of ideal to strive for.
link |
00:13:43.220
We humans, scientists, would like this,
link |
00:13:46.860
to hope that there's a beautiful structure to this
link |
00:13:48.980
where it's, like you said, there's segregated regions
link |
00:13:51.420
that are responsible for the different function.
link |
00:13:54.100
How much hope is there to find such regions
link |
00:13:57.100
in terms of looking at the progress of studying the brain?
link |
00:14:00.420
Oh, I think enormous progress has been made
link |
00:14:02.740
in the past 20 or 30 years.
link |
00:14:06.860
So this is beyond incremental.
link |
00:14:08.940
At the advent of brain imaging,
link |
00:14:11.940
the very notion of functional segregation
link |
00:14:14.820
was just a hypothesis based upon a century,
link |
00:14:18.900
if not more, of careful neuropsychology,
link |
00:14:21.420
looking at people who had lost via insult
link |
00:14:24.660
or traumatic brain injury particular parts of the brain,
link |
00:14:29.300
and then saying, well, they can't do this
link |
00:14:30.780
or they can't do that.
link |
00:14:32.060
For example, losing the visual cortex
link |
00:14:34.100
and not being able to see,
link |
00:14:35.620
or losing particular parts of the visual cortex
link |
00:14:39.580
or regions known as V5
link |
00:14:44.060
or the middle temporal region, MT,
link |
00:14:47.980
and noticing that they selectively
link |
00:14:49.580
could not see moving things.
link |
00:14:51.460
And so that created the hypothesis
link |
00:14:55.140
that perhaps visual movement processing
link |
00:14:59.700
was located in this functionally segregated area.
link |
00:15:02.740
And you could then go and put invasive electrodes
link |
00:15:05.700
in animal models and say, yes, indeed,
link |
00:15:08.260
we can excite activity here.
link |
00:15:10.500
We can form receptive fields that are sensitive to
link |
00:15:13.980
or defined in terms of visual motion.
link |
00:15:16.220
But at no point could you exclude the possibility
link |
00:15:18.340
that everywhere else in the brain
link |
00:15:20.020
was also very interested in visual motion.
link |
00:15:23.140
By the way, I apologize to interrupt,
link |
00:15:24.900
but a tiny little tangent.
link |
00:15:26.980
You said animal models, just out of curiosity,
link |
00:15:31.180
from your perspective, how different is the human brain
link |
00:15:34.460
versus the other animals
link |
00:15:35.460
in terms of our ability to study the brain?
link |
00:15:38.980
Well, clearly, the further away you go from a human brain,
link |
00:15:43.620
the greater the differences,
link |
00:15:45.740
but not as remarkable as you might think.
link |
00:15:48.940
So people will choose their level of approximation
link |
00:15:53.180
to the human brain,
link |
00:15:54.300
depending upon the kinds of questions
link |
00:15:56.540
that they want to answer.
link |
00:15:57.380
So if you're talking about sort of canonical principles
link |
00:16:00.620
of microcircuitry, it might be perfectly okay
link |
00:16:02.780
to look at a mouse, indeed.
link |
00:16:04.540
You could even look at flies, worms.
link |
00:16:08.100
If, on the other hand, you wanted to look at
link |
00:16:09.740
the finer details of organization of visual cortex
link |
00:16:13.100
and V1, V2, these are designated patches of cortex
link |
00:16:17.420
that may do different things, indeed, do.
link |
00:16:20.660
You'd probably want to use a primate
link |
00:16:23.340
that looked a little bit more like a human,
link |
00:16:26.940
because there are lots of ethical issues
link |
00:16:28.260
in terms of the use of nonhuman primates
link |
00:16:32.780
to answer questions about human anatomy.
link |
00:16:37.500
But I think most people assume
link |
00:16:39.060
that most of the important principles are conserved
link |
00:16:43.300
in a continuous way, right from, well, yes,
link |
00:16:48.020
worms right through to you and me.
link |
00:16:53.260
So now returning to, so that was the early sort of ideas
link |
00:16:56.900
of studying the functional regions of the brain
link |
00:17:00.820
by if there's some damage to it,
link |
00:17:02.860
to try to infer that that part of the brain
link |
00:17:06.140
might be somewhat responsible for this type of function.
link |
00:17:09.100
So where does that lead us?
link |
00:17:11.140
What are the next steps beyond that?
link |
00:17:12.980
Right, well, I'll just actually just reverse a bit,
link |
00:17:16.140
come back to your sort of notion
link |
00:17:17.860
that the brain is a magic soup.
link |
00:17:19.580
That was actually a very prominent idea at one point,
link |
00:17:23.780
notions such as Lashley's law of mass action
link |
00:17:28.140
inherited from the observation that for certain animals,
link |
00:17:33.140
if you just took out spoonfuls of the brain,
link |
00:17:36.020
it didn't matter where you took these spoonfuls out,
link |
00:17:38.220
they always showed the same kinds of deficits.
link |
00:17:40.420
So it was very difficult to infer functional specialization
link |
00:17:44.540
purely on the basis of lesion deficit studies.
link |
00:17:49.260
But once we had the opportunity
link |
00:17:50.820
to look at the brain lighting up
link |
00:17:52.380
and it's literally it's sort of excitement, neuronal excitement
link |
00:17:57.740
when looking at this versus that,
link |
00:18:01.100
one was able to say, yes, indeed,
link |
00:18:03.260
these functionally specialized responses
link |
00:18:05.020
are very restricted and they're here or they're over there.
link |
00:18:08.780
If I do this, then this part of the brain lights up.
link |
00:18:11.340
And that became doable in the early 90s.
link |
00:18:16.900
In fact, shortly before with the advent
link |
00:18:19.100
of positron emission tomography.
link |
00:18:21.060
And then functional magnetic resonance imaging
link |
00:18:23.420
came along in the early 90s.
link |
00:18:26.700
And since that time, there has been an explosion
link |
00:18:29.780
of discovery, refinement, confirmation.
link |
00:18:36.300
There are people who believe that it's all in the anatomy.
link |
00:18:38.940
If you understand the anatomy,
link |
00:18:40.260
then you understand the function at some level.
link |
00:18:43.260
And many, many hypotheses were predicated
link |
00:18:45.780
on a deep understanding of the anatomy and the connectivity,
link |
00:18:49.980
but they were all confirmed
link |
00:18:51.260
and taken much further with neuroimaging.
link |
00:18:55.140
So that's what I meant by we've made an enormous amount
link |
00:18:57.580
of progress in this century indeed,
link |
00:19:01.180
and in relation to the previous century,
link |
00:19:03.940
by looking at these functionally selective responses.
link |
00:19:09.620
But that wasn't the whole story.
link |
00:19:11.100
So there's this sort of near phrenology,
link |
00:19:13.580
but finding bumps and hot spots in the brain
link |
00:19:15.620
that did this or that.
link |
00:19:17.620
The bigger question was, of course,
link |
00:19:20.020
the functional integration.
link |
00:19:22.140
How all of these regionally specific responses
link |
00:19:26.980
were orchestrated, how they were distributed,
link |
00:19:29.860
how did they relate to distributed processing
link |
00:19:32.700
and indeed representations in the brain.
link |
00:19:35.580
So then you turn to the more challenging issue
link |
00:19:39.540
of the integration, the connectivity.
link |
00:19:42.660
And then we come back to this beautiful,
link |
00:19:44.980
sparse, recurrent, hierarchical connectivity
link |
00:19:49.060
that seems characteristic of the brain
link |
00:19:51.140
and probably not many other organs.
link |
00:19:53.620
But nevertheless, we come back to this challenge
link |
00:19:58.260
of trying to figure out how everything is integrated.
link |
00:20:01.100
But what's your feeling?
link |
00:20:02.820
What's the general consensus?
link |
00:20:04.220
Have we moved away from the magic soup view of the brain?
link |
00:20:07.820
So there is a deep structure to it.
link |
00:20:11.940
And then maybe a further question.
link |
00:20:14.460
You said some people believe that the structure
link |
00:20:16.820
is most of it, that you can really get
link |
00:20:19.180
at the core of the function
link |
00:20:20.220
by just deeply understanding the structure.
link |
00:20:22.540
Where do you sit on that, do you?
link |
00:20:25.180
I think it's got some mileage to it, yes, yeah.
link |
00:20:28.260
So it's a worthy pursuit of going,
link |
00:20:31.180
of studying through imaging and all the different methods
link |
00:20:34.780
to actually study the structure.
link |
00:20:36.260
No, absolutely, yeah, yeah.
link |
00:20:38.340
Sorry, I'm just noting, you were accusing me
link |
00:20:41.180
of using lots of long words
link |
00:20:42.460
and then you introduced one there, which is deep,
link |
00:20:44.300
which is interesting.
link |
00:20:46.340
Because deep is the sort of millennial equivalent
link |
00:20:50.420
of hierarchical.
link |
00:20:51.660
So if you've put deep in front of anything,
link |
00:20:53.860
not only are you very millennial and very trending,
link |
00:20:57.620
but you're also implying a hierarchical architecture.
link |
00:21:01.540
So it is a depth, which is, for me, the beautiful thing.
link |
00:21:05.260
That's right, the word deep kind of,
link |
00:21:07.380
yeah, exactly, it implies hierarchy.
link |
00:21:10.260
I didn't even think about that.
link |
00:21:11.500
That indeed, the implicit meaning
link |
00:21:14.620
of the word deep is hierarchy.
link |
00:21:16.820
Yep. Yeah.
link |
00:21:18.340
So deep inside the onion is the center of your soul.
link |
00:21:22.380
Beautifully put.
link |
00:21:23.500
Maybe briefly, if you could paint a picture
link |
00:21:26.780
of the kind of methods of neuroimaging,
link |
00:21:30.980
maybe the history which you were a part of,
link |
00:21:33.420
from statistical parametric mapping.
link |
00:21:35.180
I mean, just what's out there that's interesting
link |
00:21:37.940
for people maybe outside the field
link |
00:21:40.540
to understand of what are the actual methodologies
link |
00:21:43.460
of looking inside the human brain?
link |
00:21:45.340
Right, well, you can answer that question
link |
00:21:47.420
from two perspectives.
link |
00:21:48.300
Basically, it's the modality.
link |
00:21:49.900
What kind of signal are you measuring?
link |
00:21:52.580
And they can range from,
link |
00:21:55.460
and let's limit ourselves
link |
00:21:56.860
to sort of imaging based noninvasive techniques.
link |
00:22:01.060
So you've essentially got brain scanners,
link |
00:22:03.020
and brain scanners can either measure
link |
00:22:05.340
the structural attributes, the amount of water,
link |
00:22:07.660
the amount of fat, or the amount of iron
link |
00:22:09.300
in different parts of the brain,
link |
00:22:10.380
and you can make lots of inferences
link |
00:22:11.620
about the structure of the organ of the sort
link |
00:22:15.460
that you might have produced from an X ray,
link |
00:22:17.780
but a very nuanced X ray that is looking
link |
00:22:21.340
at this kind of property or that kind of property.
link |
00:22:24.380
So looking at the anatomy noninvasively
link |
00:22:27.860
would be the first sort of neuroimaging
link |
00:22:30.140
that people might want to employ.
link |
00:22:32.180
Then you move on to the kinds of measurements
link |
00:22:34.940
that reflect dynamic function,
link |
00:22:38.100
and the most prevalent of those fall into two camps.
link |
00:22:42.020
You've got these metabolic, sometimes hemodynamic,
link |
00:22:46.460
blood related signals.
link |
00:22:48.940
So these metabolic and or hemodynamic signals
link |
00:22:53.460
are basically proxies for elevated activity
link |
00:22:58.380
and message passing and neuronal dynamics
link |
00:23:03.460
in particular parts of the brain.
link |
00:23:05.340
Characteristically though, the time constants
link |
00:23:07.660
of these hemodynamic or metabolic responses
link |
00:23:11.420
to neural activity are much longer
link |
00:23:14.300
than the neural activity itself.
link |
00:23:15.820
And this is referring,
link |
00:23:19.020
forgive me for the dumb questions,
link |
00:23:20.580
but this would be referring to blood,
link |
00:23:22.940
like the flow of blood.
link |
00:23:24.260
Absolutely, absolutely.
link |
00:23:25.100
So there's a ton of,
link |
00:23:26.500
it seems like there's a ton of blood vessels in the brain.
link |
00:23:29.420
Yeah.
link |
00:23:30.260
So what's the interaction between the flow of blood
link |
00:23:33.700
and the function of the neurons?
link |
00:23:35.980
Is there an interplay there or?
link |
00:23:37.420
Yup, yup, and that interplay accounts for several careers
link |
00:23:42.780
of world renowned scientists, yes, absolutely.
link |
00:23:47.220
So this is known as neurovascular coupling,
link |
00:23:49.140
is exactly what you said.
link |
00:23:50.180
It's how does the neural activity,
link |
00:23:52.300
the neuronal infrastructure, the actual message passing
link |
00:23:54.620
that we think underlies our capacity to perceive and act,
link |
00:24:01.060
how is that coupled to the vascular responses
link |
00:24:06.060
that supply the energy for that neural processing?
link |
00:24:09.900
So there's a delicate web of large vessels,
link |
00:24:13.380
arteries and veins, that gets progressively finer
link |
00:24:16.420
and finer in detail until it perfuses
link |
00:24:18.860
at a microscopic level,
link |
00:24:20.420
the machinery where little neurons lie.
link |
00:24:23.780
So coming back to this sort of onion perspective,
link |
00:24:27.420
we were talking before using the onion as a metaphor
link |
00:24:30.500
for a deep hierarchical structure,
link |
00:24:32.460
but also I think it's just anatomically quite
link |
00:24:36.220
a useful metaphor.
link |
00:24:37.780
All the action, all the heavy lifting
link |
00:24:40.140
in terms of neural computation is done
link |
00:24:41.660
on the surface of the brain,
link |
00:24:43.940
and then the interior of the brain is constituted
link |
00:24:47.380
by fatty wires, essentially, axonal processes
link |
00:24:52.980
that are enshrouded by myelin sheaths.
link |
00:24:55.940
And these, when you dissect them, they look fatty and white,
link |
00:24:59.620
and so it's called white matter,
link |
00:25:01.240
as opposed to the actual neuro pill,
link |
00:25:04.100
which does the computation constituted largely by neurons,
link |
00:25:07.220
and that's known as gray matter.
link |
00:25:08.700
So the gray matter is a surface or a skin
link |
00:25:13.260
that sits on top of this big ball,
link |
00:25:16.260
now we are talking magic soup,
link |
00:25:17.780
but a big ball of connections like spaghetti,
link |
00:25:20.860
very carefully structured with sparse connectivity
link |
00:25:23.100
that preserves this deep hierarchical structure,
link |
00:25:25.760
but all the action takes place on the surface,
link |
00:25:28.300
on the cortex of the onion, and that means
link |
00:25:34.180
that you have to supply the right amount of blood flow,
link |
00:25:38.560
the right amount of nutrient,
link |
00:25:41.100
which is rapidly absorbed and used by neural cells
link |
00:25:45.240
that don't have the same capacity
link |
00:25:46.820
that your leg muscles would have
link |
00:25:48.780
to basically spend their energy budget
link |
00:25:52.500
and then claim it back later.
link |
00:25:55.000
So one peculiar thing about cerebral metabolism,
link |
00:25:58.420
brain metabolism, is it really needs to be driven
link |
00:26:01.440
in the moment, which means you basically
link |
00:26:03.060
have to turn on the taps.
link |
00:26:04.860
So if there's lots of neural activity
link |
00:26:08.700
in one part of the brain, a little patch
link |
00:26:10.780
of a few millimeters, even less possibly,
link |
00:26:14.100
you really do have to water that piece
link |
00:26:16.060
of the garden now and quickly,
link |
00:26:18.420
and by quickly I mean within a couple of seconds.
link |
00:26:21.540
So that contains a lot of, hence the imaging
link |
00:26:26.060
could tell you a story of what's happening.
link |
00:26:28.260
Absolutely, but it is slightly compromised
link |
00:26:32.240
in terms of the resolution.
link |
00:26:33.460
So the deployment of these little microvessels
link |
00:26:37.580
that water the garden to enable the neural activity
link |
00:26:42.380
to play out, the spatial resolution
link |
00:26:45.320
is in order of a few millimeters,
link |
00:26:48.020
and crucially, the temporal resolution
link |
00:26:50.380
is the order of a few seconds.
link |
00:26:52.220
So you can't get right down and dirty
link |
00:26:54.420
into the actual spatial and temporal scale
link |
00:26:57.400
of neural activity in and of itself.
link |
00:26:59.920
To do that, you'd have to turn
link |
00:27:00.860
to the other big imaging modality,
link |
00:27:02.580
which is the recording of electromagnetic signals
link |
00:27:05.580
as they're generated in real time.
link |
00:27:07.740
So here, the temporal bandwidth, if you like,
link |
00:27:10.320
or the low limit on the temporal resolution
link |
00:27:12.960
is incredibly small, talking about milliseconds.
link |
00:27:17.960
And then you can get into the phasic fast responses
link |
00:27:23.520
that is in and of itself the neural activity,
link |
00:27:27.440
and start to see the succession or cascade
link |
00:27:32.780
of hierarchical recurrent message passing
link |
00:27:35.220
evoked by a particular stimulus.
link |
00:27:37.140
But the problem is you're looking
link |
00:27:39.400
at electromagnetic signals that have passed
link |
00:27:42.440
through an enormous amount of magic soup
link |
00:27:45.640
or spaghetti of collectivity, and through the scalp
link |
00:27:49.480
and the skull, and it's become spatially very diffuse.
link |
00:27:52.920
So it's very difficult to know where you are.
link |
00:27:54.940
So you've got this sort of catch 22.
link |
00:27:58.600
You can either use an imaging modality
link |
00:28:00.280
that tells you within millimeters
link |
00:28:02.840
which part of the brain is activated,
link |
00:28:04.440
but you don't know when,
link |
00:28:05.800
or you've got these electromagnetic EEG, MEG setups
link |
00:28:10.800
that tell you to within a few milliseconds
link |
00:28:15.840
when something has responded, but you're not aware.
link |
00:28:19.240
So you've got these two complementary measures,
link |
00:28:22.520
either indirect via the blood flow,
link |
00:28:25.840
or direct via the electromagnetic signals
link |
00:28:28.780
caused by neural activity.
link |
00:28:31.080
These are the two big imaging devices.
link |
00:28:33.360
And then the second level of responding to your question,
link |
00:28:36.960
what are the, from the outside,
link |
00:28:39.440
what are the big ways of using this technology?
link |
00:28:44.240
So once you've chosen the kind of neural imaging
link |
00:28:47.160
that you want to use to answer your set questions,
link |
00:28:50.240
and sometimes it would have to be both,
link |
00:28:53.360
then you've got a whole raft of analyses,
link |
00:28:57.540
time series analyses usually, that you can bring to bear
link |
00:29:01.920
in order to answer your questions
link |
00:29:04.360
or address your hypothesis about those data.
link |
00:29:07.000
And interestingly, they both fall
link |
00:29:08.920
into the same two camps we were talking about before,
link |
00:29:11.320
this dialectic between specialization and integration,
link |
00:29:14.800
differentiation and integration.
link |
00:29:17.120
So it's the cartography, the blobology analyses.
link |
00:29:20.880
I apologize, I probably shouldn't interrupt so much,
link |
00:29:23.220
but just heard a fun word, the blah.
link |
00:29:27.080
Blobology.
link |
00:29:27.920
Blobology.
link |
00:29:29.160
It's a neologism, which means the study of blobs.
link |
00:29:33.000
So nothing bob.
link |
00:29:34.720
Are you being witty and humorous,
link |
00:29:36.640
or does the word blobology ever appear
link |
00:29:39.640
in a textbook somewhere?
link |
00:29:40.760
It would appear in a popular book.
link |
00:29:43.320
It would not appear in a worthy specialist journal.
link |
00:29:47.960
Yeah, I thought so.
link |
00:29:48.960
It's the fond word for the study of literally little blobs
link |
00:29:53.560
on brain maps showing activations.
link |
00:29:56.160
So the kind of thing that you'd see in the newspapers
link |
00:29:59.500
on ABC or BBC reporting the latest finding
link |
00:30:04.080
from brain imaging.
link |
00:30:05.320
Interestingly though, the maths involved
link |
00:30:10.080
in that stream of analysis does actually call upon
link |
00:30:15.340
the mathematics of blobs.
link |
00:30:17.640
So seriously, they're actually called Euler characteristics
link |
00:30:21.800
and they have a lot of fancy names in mathematics.
link |
00:30:27.500
We'll talk about it, about your ideas
link |
00:30:28.920
in free energy principle.
link |
00:30:30.400
I mean, there's echoes of blobs there
link |
00:30:33.320
when you consider sort of entities,
link |
00:30:36.720
mathematically speaking.
link |
00:30:38.000
Yes, absolutely.
link |
00:30:40.800
Well, circumscribed, well defined,
link |
00:30:43.080
you entities of, well, from the free energy point of view,
link |
00:30:48.120
entities of anything, but from the point of view
link |
00:30:50.280
of the analysis, the cartography of the brain,
link |
00:30:55.800
these are the entities that constitute the evidence
link |
00:30:59.200
for this functional segregation.
link |
00:31:01.680
You have segregated this function in this blob
link |
00:31:04.500
and it is not outside of the blob.
link |
00:31:06.760
And that's basically the, if you were a map maker
link |
00:31:10.200
of America and you did not know its structure,
link |
00:31:14.080
the first thing were you doing constituting
link |
00:31:16.400
or creating a map would be to identify the cities,
link |
00:31:19.140
for example, or the mountains or the rivers.
link |
00:31:22.000
All of these uniquely spatially localizable features,
link |
00:31:26.920
possibly topological features have to be placed somewhere
link |
00:31:30.680
because that requires a mathematics of identifying
link |
00:31:33.520
what does a city look like on a satellite image
link |
00:31:36.080
or what does a river look like
link |
00:31:37.400
or what does a mountain look like?
link |
00:31:39.120
What would it, you know, what data features
link |
00:31:42.240
would evidence that particular top,
link |
00:31:46.520
you know, that particular thing
link |
00:31:48.560
that you wanted to put on the map?
link |
00:31:50.400
And they normally are characterized
link |
00:31:52.760
in terms of literally these blobs
link |
00:31:54.320
or these sort of, another way of looking at this
link |
00:31:57.000
is that a certain statistical measure
link |
00:32:01.560
of the degree of activation crosses a threshold
link |
00:32:04.000
and in crossing that threshold
link |
00:32:06.520
in the spatially restricted part of the brain,
link |
00:32:09.600
it creates a blob.
link |
00:32:11.040
And that's basically what statistical parametric mapping does.
link |
00:32:14.000
It's basically mathematically finessed blobology.
link |
00:32:19.840
Okay, so those are the,
link |
00:32:20.920
you kind of described these two methodologies for,
link |
00:32:24.000
one is temporally noisy, one is spatially noisy
link |
00:32:26.920
and you kind of have to play and figure out
link |
00:32:28.360
what can be useful.
link |
00:32:31.400
It'd be great if you can sort of comment.
link |
00:32:33.000
I got a chance recently to spend a day
link |
00:32:34.920
at a company called Neuralink
link |
00:32:37.200
that uses brain computer interfaces
link |
00:32:39.420
and their dream is to, well,
link |
00:32:42.720
there's a bunch of sort of dreams,
link |
00:32:45.200
but one of them is to understand the brain
link |
00:32:47.380
by sort of, you know, getting in there
link |
00:32:51.520
past the so called sort of factory wall,
link |
00:32:53.800
getting in there and be able to listen,
link |
00:32:55.200
communicate both directions.
link |
00:32:57.160
What are your thoughts about this,
link |
00:32:59.360
the future of this kind of technology
link |
00:33:01.040
of brain computer interfaces
link |
00:33:02.360
to be able to now have a window
link |
00:33:06.920
or direct contact within the brain
link |
00:33:08.500
to be able to measure some of the signals,
link |
00:33:10.040
to be able to sense signals,
link |
00:33:11.320
to understand some of the functionality of the brain?
link |
00:33:15.040
Ambivalent, my sense is ambivalent.
link |
00:33:17.760
So it's a mixture of good and bad
link |
00:33:19.920
and I acknowledge that freely.
link |
00:33:22.400
So the good bits, if you just look at the legacy
link |
00:33:24.760
of that kind of reciprocal but invasive
link |
00:33:29.240
your brain stimulation,
link |
00:33:31.160
I didn't paint a complete picture
link |
00:33:33.160
when I was talking about sort of the ways
link |
00:33:34.680
we understand the brain prior to neuroimaging.
link |
00:33:37.080
It wasn't just lesion deficit studies.
link |
00:33:39.680
Some of the early work, in fact,
link |
00:33:41.380
literally 100 years from where we're sitting
link |
00:33:43.500
at the institution of neurology,
link |
00:33:45.240
was done by stimulating the brain of say dogs
link |
00:33:50.000
and looking at how they responded
link |
00:33:51.880
with their muscles or with their salivation
link |
00:33:56.400
and imputing what that part of the brain must be doing.
link |
00:34:00.920
If I stimulate it and I vote this kind of response,
link |
00:34:06.000
then that tells me quite a lot
link |
00:34:07.240
about the functional specialization.
link |
00:34:09.080
So there's a long history of brain stimulation
link |
00:34:12.280
which continues to enjoy a lot of attention nowadays.
link |
00:34:16.840
Positive attention.
link |
00:34:17.720
Oh yes, absolutely.
link |
00:34:19.600
You know, deep brain stimulation for Parkinson's disease
link |
00:34:22.160
is now a standard treatment
link |
00:34:23.600
and also a wonderful vehicle
link |
00:34:25.560
to try and understand the neuronal dynamics
link |
00:34:29.000
underlying movement disorders like Parkinson's disease.
link |
00:34:33.120
Even interest in magnetic stimulation,
link |
00:34:37.920
stimulating the magnetic fields
link |
00:34:39.240
and will it work in people who are depressed, for example.
link |
00:34:43.320
Quite a crude level of understanding what you're doing,
link |
00:34:45.700
but there is historical evidence
link |
00:34:49.080
that these kinds of brute force interventions
link |
00:34:51.800
do change things.
link |
00:34:53.380
They, you know, it's a little bit like banging the TV
link |
00:34:56.000
when the valves aren't working properly,
link |
00:34:58.240
but it's still, it works.
link |
00:35:00.720
So, you know, there is a long history.
link |
00:35:04.480
Brain computer interfacing or BCI,
link |
00:35:08.700
I think is a beautiful example of that.
link |
00:35:11.000
It's sort of carved out its own niche
link |
00:35:12.720
and its own aspirations
link |
00:35:14.420
and there've been enormous advances within limits.
link |
00:35:20.800
Advances in terms of our ability to understand
link |
00:35:25.480
how the brain, the embodied brain,
link |
00:35:29.720
engages with the world.
link |
00:35:32.480
I'm thinking here of sensory substitution,
link |
00:35:34.760
the augmenting our sensory capacities
link |
00:35:37.240
by giving ourselves extra ways of sensing
link |
00:35:40.820
and sampling the world,
link |
00:35:42.260
ranging from sort of trying to replace lost visual signals
link |
00:35:48.300
through to giving people completely new signals.
link |
00:35:50.380
So, one of the, I think, most engaging examples of this
link |
00:35:57.100
is equipping people with a sense of magnetic fields.
link |
00:36:00.660
So you can actually give them magnetic sensors
link |
00:36:03.620
that enable them to feel,
link |
00:36:05.460
should we say, tactile pressure around their tummy,
link |
00:36:08.980
where they are in relation to the magnetic field of the Earth.
link |
00:36:13.100
And after a few weeks, they take it for granted.
link |
00:36:17.660
They integrate it, they embody this,
link |
00:36:19.340
simulate this new sensory information
link |
00:36:22.340
into the way that they literally feel their world,
link |
00:36:25.420
but now equipped with this sense of magnetic direction.
link |
00:36:29.300
So that tells you something
link |
00:36:31.020
about the brain's plastic potential
link |
00:36:32.980
to remodel and its plastic capacity
link |
00:36:37.980
to suddenly try to explain the sensory data at hand
link |
00:36:43.640
by augmenting the sensory sphere
link |
00:36:48.440
and the kinds of things that you can measure.
link |
00:36:51.600
Clearly, that's purely for entertainment
link |
00:36:54.720
and understanding the nature and the power of our brains.
link |
00:37:00.120
I would imagine that most BCI is pitched
link |
00:37:03.400
at solving clinical and human problems
link |
00:37:08.600
such as locked in syndrome, such as paraplegia,
link |
00:37:12.240
or replacing lost sensory capacities
link |
00:37:16.120
like blindness and deafness.
link |
00:37:18.960
So then we come to the negative part of my ambivalence,
link |
00:37:24.240
the other side of it.
link |
00:37:26.960
So I don't want to be deflationary
link |
00:37:30.960
because much of my deflationary comments
link |
00:37:33.400
is probably large out of ignorance than anything else.
link |
00:37:37.120
But generally speaking, the bandwidth
link |
00:37:42.480
and the bit rates that you get
link |
00:37:44.360
from brain computer interfaces as we currently know them,
link |
00:37:49.160
we're talking about bits per second.
link |
00:37:51.460
So that would be like me only being able to communicate
link |
00:37:55.600
with any world or with you using very, very, very slow Morse code.
link |
00:38:06.560
And it is not even within an order of magnitude
link |
00:38:13.440
near what we actually need for an inactive realization
link |
00:38:18.600
of what people aspire to when they think about
link |
00:38:21.280
sort of curing people with paraplegia or replacing sight
link |
00:38:28.760
despite heroic efforts.
link |
00:38:30.280
So one has to ask, is there a lower bound
link |
00:38:33.760
on the kinds of recurrent information exchange
link |
00:38:41.040
between a brain and some augmented or artificial interface?
link |
00:38:46.040
And then we come back to, interestingly,
link |
00:38:51.620
what I was talking about before,
link |
00:38:52.820
which is if you're talking about function
link |
00:38:56.460
in terms of inference, and I presume we'll get to that
link |
00:39:00.220
later on in terms of the free energy principle,
link |
00:39:01.900
then at the moment, there may be fundamental reasons
link |
00:39:05.140
to assume that is the case.
link |
00:39:06.400
We're talking about ensemble activity.
link |
00:39:08.540
We're talking about basically, for example,
link |
00:39:13.540
let's paint the challenge facing brain computer interfacing
link |
00:39:20.800
in terms of controlling another system
link |
00:39:24.520
that is highly and deeply structured,
link |
00:39:27.080
very relevant to our lives, very nonlinear,
link |
00:39:30.560
that rests upon the kind of nonequilibrium
link |
00:39:34.360
steady states and dynamics that the brain does,
link |
00:39:37.400
the weather, all right?
link |
00:39:39.540
So imagine you had some very aggressive satellites
link |
00:39:45.840
that could produce signals that could perturb
link |
00:39:48.280
some little parts of the weather system.
link |
00:39:53.000
And then what you're asking now is,
link |
00:39:55.140
can I meaningfully get into the weather
link |
00:39:58.320
and change it meaningfully and make the weather respond
link |
00:40:01.300
in a way that I want it to?
link |
00:40:03.360
You're talking about chaos control on a scale
link |
00:40:06.600
which is almost unimaginable.
link |
00:40:08.860
So there may be fundamental reasons
link |
00:40:11.120
why BCI, as you might read about it in a science fiction novel,
link |
00:40:18.480
aspirational BCI may never actually work
link |
00:40:22.920
in the sense that to really be integrated
link |
00:40:26.920
and be part of the system is a requirement
link |
00:40:32.160
that requires you to have evolved with that system,
link |
00:40:35.240
that you have to be part of a very delicately structured,
link |
00:40:43.320
deeply structured, dynamic, ensemble activity
link |
00:40:48.000
that is not like rewiring a broken computer
link |
00:40:51.600
or plugging in a peripheral interface adapter.
link |
00:40:54.660
It is much more like getting into the weather patterns
link |
00:40:58.140
or, come back to your magic soup,
link |
00:41:00.680
getting into the active matter
link |
00:41:02.420
and meaningfully relate that to the outside world.
link |
00:41:07.140
So I think there are enormous challenges there.
link |
00:41:09.920
So I think the example of the weather is a brilliant one.
link |
00:41:13.260
And I think you paint a really interesting picture
link |
00:41:15.300
and it wasn't as negative as I thought.
link |
00:41:17.420
It's essentially saying that it might be
link |
00:41:19.920
incredibly challenging, including the low bound
link |
00:41:22.140
of the bandwidth and so on.
link |
00:41:23.640
I kind of, so just to full disclosure,
link |
00:41:26.900
I come from the machine learning world.
link |
00:41:28.680
So my natural thought is the hardest part
link |
00:41:32.760
is the engineering challenge of controlling the weather,
link |
00:41:34.820
of getting those satellites up and running and so on.
link |
00:41:37.560
And once they are, then the rest is fundamentally
link |
00:41:42.220
the same approaches that allow you to win in the game of Go
link |
00:41:46.860
will allow you to potentially play in this soup,
link |
00:41:49.580
in this chaos.
link |
00:41:51.000
So I have a hope that sort of machine learning methods
link |
00:41:54.500
will help us play in this soup.
link |
00:41:58.820
But perhaps you're right that it is a biology
link |
00:42:04.220
and the brain is just an incredible system
link |
00:42:08.660
that may be almost impossible to get in.
link |
00:42:12.220
But for me, what seems impossible
link |
00:42:15.780
is the incredible mess of blood vessels
link |
00:42:19.780
that you also described without,
link |
00:42:22.300
we also value the brain.
link |
00:42:24.620
You can't make any mistakes, you can't damage things.
link |
00:42:27.100
So to me, that engineering challenge seems nearly impossible.
link |
00:42:31.340
One of the things I was really impressed by at Neuralink
link |
00:42:35.900
is just talking to brilliant neurosurgeons
link |
00:42:39.660
and the roboticists that made me realize
link |
00:42:43.340
that even though it seems impossible,
link |
00:42:45.860
if anyone can do it, it's some of these world class
link |
00:42:48.620
engineers that are trying to take it on.
link |
00:42:50.800
So I think the conclusion of our discussion here
link |
00:42:55.860
of this part is basically that the problem is really hard
link |
00:43:00.980
but hopefully not impossible.
link |
00:43:02.580
Absolutely.
link |
00:43:03.800
So if it's okay, let's start with the basics.
link |
00:43:07.260
So you've also formulated a fascinating principle,
link |
00:43:12.180
the free energy principle.
link |
00:43:13.500
Could we maybe start at the basics
link |
00:43:15.340
and what is the free energy principle?
link |
00:43:19.660
Well, in fact, the free energy principle
link |
00:43:23.700
inherits a lot from the building
link |
00:43:29.220
of these data analytic approaches
link |
00:43:31.240
to these very high dimensional time series
link |
00:43:34.180
you get from the brain.
link |
00:43:35.960
So I think it's interesting to acknowledge that.
link |
00:43:37.940
And in particular, the analysis tools
link |
00:43:39.980
that try to address the other side,
link |
00:43:43.100
which is a functional integration,
link |
00:43:44.360
so the connectivity analyses.
link |
00:43:46.060
So on the one hand, but I should also acknowledge
link |
00:43:51.920
it inherits an awful lot from machine learning as well.
link |
00:43:55.360
So the free energy principle is just a formal statement
link |
00:44:02.600
that the existential imperatives for any system
link |
00:44:08.500
that manages to survive in a changing world
link |
00:44:11.380
can be cast as an inference problem
link |
00:44:18.900
in the sense that you can interpret
link |
00:44:21.240
the probability of existing as the evidence that you exist.
link |
00:44:25.720
And if you can write down that problem of existence
link |
00:44:29.460
as a statistical problem,
link |
00:44:30.900
then you can use all the maths that has been developed
link |
00:44:33.940
for inference to understand and characterize
link |
00:44:38.940
the ensemble dynamics that must be in play
link |
00:44:42.960
in the service of that inference.
link |
00:44:45.560
So technically, what that means is
link |
00:44:48.240
you can always interpret anything that exists
link |
00:44:52.160
in virtue of being separate from the environment
link |
00:44:55.660
in which it exists as trying to minimize
link |
00:45:02.100
variational free energy.
link |
00:45:03.600
And if you're from the machine learning community,
link |
00:45:05.560
you will know that as a negative evidence lower bound
link |
00:45:09.200
or a negative elbow, which is the same as saying
link |
00:45:13.120
you're trying to maximize or it will look as if
link |
00:45:16.400
all your dynamics are trying to maximize
link |
00:45:19.720
the complement of that which is the marginal likelihood
link |
00:45:24.000
or the evidence for your own existence.
link |
00:45:26.480
So that's basically the free energy principle.
link |
00:45:30.120
But to even take a sort of a small step backwards,
link |
00:45:34.120
you said the existential imperative.
link |
00:45:38.360
There's a lot of beautiful poetic words here,
link |
00:45:40.120
but to put it crudely, it's a fascinating idea
link |
00:45:46.840
of basically just of trying to describe
link |
00:45:49.520
if you're looking at a blob,
link |
00:45:51.800
how do you know this thing is alive?
link |
00:45:54.320
What does it mean to be alive?
link |
00:45:55.680
What does it mean to exist?
link |
00:45:57.520
And so you can look at the brain,
link |
00:45:59.440
you can look at parts of the brain,
link |
00:46:00.720
or this is just a general principle
link |
00:46:02.840
that applies to almost any system.
link |
00:46:07.240
That's just a fascinating sort of philosophically
link |
00:46:10.160
at every level question and a methodology
link |
00:46:13.120
to try to answer that question.
link |
00:46:14.360
What does it mean to be alive?
link |
00:46:16.040
So that's a huge endeavor and it's nice
link |
00:46:21.480
that there's at least some,
link |
00:46:23.200
from some perspective, a clean answer.
link |
00:46:25.480
So maybe can you talk about that optimization view of it?
link |
00:46:30.280
So what's trying to be minimized, maximized?
link |
00:46:33.520
A system that's alive, what is it trying to minimize?
link |
00:46:36.840
Right, you've made a big move there.
link |
00:46:40.600
First of all, it's good to make big moves.
link |
00:46:45.960
But you've assumed that the thing exists
link |
00:46:52.000
in a state that could be living or nonliving.
link |
00:46:54.680
So I may ask you, what licenses you
link |
00:46:57.960
to say that something exists?
link |
00:47:00.120
That's why I use the word existential.
link |
00:47:02.280
It's beyond living, it's just existence.
link |
00:47:05.440
So if you drill down onto the definition
link |
00:47:08.000
of things that exist, then they have certain properties
link |
00:47:13.520
if you borrow the maths
link |
00:47:16.400
from nonequilibrium steady state physics
link |
00:47:19.360
that enable you to interpret their existence
link |
00:47:26.240
in terms of this optimization procedure.
link |
00:47:29.280
So it's good you introduced the word optimization.
link |
00:47:32.160
So what the free energy principle
link |
00:47:36.000
in its sort of most ambitious,
link |
00:47:39.800
but also most deflationary and simplest, says
link |
00:47:44.720
is that if something exists,
link |
00:47:47.200
then it must, by the mathematics
link |
00:47:51.440
of nonequilibrium steady state,
link |
00:47:55.120
exhibit properties that make it look
link |
00:47:59.600
as if it is optimizing a particular quantity.
link |
00:48:03.680
And it turns out that particular quantity
link |
00:48:06.120
happens to be exactly the same
link |
00:48:08.520
as the evidence lower bound in machine learning
link |
00:48:11.360
or Bayesian model evidence in Bayesian statistics.
link |
00:48:15.280
Or, and then I can list a whole other list
link |
00:48:18.800
of ways of understanding this key quantity,
link |
00:48:23.480
which is a bound on surprise or self information
link |
00:48:29.120
if you have information theory.
link |
00:48:31.000
There are a number of different perspectives
link |
00:48:34.080
on this quantity.
link |
00:48:34.920
It's just basically the log probability
link |
00:48:36.920
of being in a particular state.
link |
00:48:40.120
I'm telling this story as an honest,
link |
00:48:42.840
an attempt to answer your question.
link |
00:48:45.640
And I'm answering it as if I was pretending
link |
00:48:49.280
to be a physicist who was trying to understand
link |
00:48:52.400
the fundaments of nonequilibrium steady state.
link |
00:48:58.040
And I shouldn't really be doing that
link |
00:48:59.640
because the last time I was taught physics,
link |
00:49:02.240
I was in my 20s.
link |
00:49:03.760
What kind of systems,
link |
00:49:04.720
when you think about the free energy principle,
link |
00:49:06.400
what kind of systems are you imagining
link |
00:49:08.640
as a sort of more specific kind of case study?
link |
00:49:11.640
Yeah, I'm imagining a range of systems,
link |
00:49:15.720
but at its simplest, a single celled organism
link |
00:49:23.400
that can be identified from its eco niche
link |
00:49:26.120
or its environment.
link |
00:49:27.680
So at its simplest, that's basically
link |
00:49:31.480
what I always imagined in my head.
link |
00:49:33.840
And you may ask, well, is there any,
link |
00:49:36.680
how on earth can you even elaborate questions
link |
00:49:41.680
about the existence of a single drop of oil, for example?
link |
00:49:48.000
But there are deep questions there.
link |
00:49:49.440
Why doesn't the oil, why doesn't the thing,
link |
00:49:52.560
the interface between the drop of oil
link |
00:49:55.160
that contains an interior
link |
00:49:57.480
and the thing that is not the drop of oil,
link |
00:50:00.320
which is the solvent in which it is immersed,
link |
00:50:03.880
how does that interface persist over time?
link |
00:50:07.080
Why doesn't the oil just dissolve into solvent?
link |
00:50:10.400
So what special properties of the exchange
link |
00:50:16.160
between the surface of the oil drop
link |
00:50:18.200
and the external states in which it's immersed,
link |
00:50:22.000
if you're a physicist, say it would be the heat bath.
link |
00:50:24.040
You've got a physical system, an ensemble again,
link |
00:50:28.280
we're talking about density dynamics, ensemble dynamics,
link |
00:50:30.320
an ensemble of atoms or molecules immersed in the heat bath.
link |
00:50:35.840
But the question is, how did the heat bath get there?
link |
00:50:38.760
And why does it not dissolve?
link |
00:50:41.360
How is it maintaining itself?
link |
00:50:42.800
Exactly.
link |
00:50:43.640
What actions is it?
link |
00:50:44.480
I mean, it's such a fascinating idea of a drop of oil
link |
00:50:47.520
and I guess it would dissolve in water,
link |
00:50:50.000
it wouldn't dissolve in water.
link |
00:50:51.680
So what?
link |
00:50:52.520
Precisely, so why not?
link |
00:50:53.840
So why not?
link |
00:50:54.680
Why not?
link |
00:50:55.520
And how do you mathematically describe,
link |
00:50:57.000
I mean, it's such a beautiful idea.
link |
00:50:58.680
And also the idea of like, where does the thing,
link |
00:51:02.120
where does the drop of oil end and where does it begin?
link |
00:51:07.120
Right, so I mean, you're asking deep questions,
link |
00:51:10.400
deep in a nonmillennial sense here.
link |
00:51:12.520
In a hierarchical sense.
link |
00:51:16.280
But what you can do, so this is the deflationary part of it.
link |
00:51:21.720
Can I just qualify my answer by saying that normally
link |
00:51:23.920
when I'm asked this question,
link |
00:51:24.760
I answer from the point of view of a psychologist,
link |
00:51:26.640
we talk about predictive processing and predictive coding
link |
00:51:29.080
and the brain as an inference machine,
link |
00:51:31.640
but you haven't asked me from that perspective,
link |
00:51:33.920
I'm answering from the point of view of a physicist.
link |
00:51:36.240
So the question is not so much why,
link |
00:51:41.000
but if it exists, what properties must it display?
link |
00:51:44.440
So that's the deflationary part of the free energy principle.
link |
00:51:46.880
The free energy principle does not supply an answer
link |
00:51:50.800
as to why, it's saying if something exists,
link |
00:51:54.520
then it must display these properties.
link |
00:51:57.720
That's the sort of thing that's on offer.
link |
00:52:01.560
And it so happens that these properties it must display
link |
00:52:05.240
are actually intriguing and have this inferential gloss,
link |
00:52:10.880
this sort of self evidencing gloss that inherits on the fact
link |
00:52:14.920
that the very preservation of the boundary
link |
00:52:19.720
between the oil drop and the not oil drop
link |
00:52:22.640
requires an optimization of a particular function
link |
00:52:26.160
or a functional that defines the presence
link |
00:52:30.600
or the existence of this oil drop,
link |
00:52:33.040
which is why I started with existential imperatives.
link |
00:52:36.160
It is a necessary condition for existence
link |
00:52:39.520
that this must occur because the boundary
link |
00:52:42.880
basically defines the thing that's existing.
link |
00:52:46.040
So it is that self assembly aspect
link |
00:52:47.960
it's that you were hinting at in biology,
link |
00:52:53.080
sometimes known as autopoiesis
link |
00:52:56.880
in computational chemistry with self assembly.
link |
00:53:00.160
It's the, what does it look like?
link |
00:53:03.600
Sorry, how would you describe things
link |
00:53:06.040
that configure themselves out of nothing?
link |
00:53:08.640
The way they clearly demarcate themselves
link |
00:53:12.000
from the states or the soup in which they are immersed.
link |
00:53:18.240
So from the point of view of computational chemistry,
link |
00:53:20.920
for example, you would just understand that
link |
00:53:23.440
as a configuration of a macro molecule
link |
00:53:25.440
to minimize its free energy, its thermodynamic free energy.
link |
00:53:28.920
It's exactly the same principle that we've been talking about
link |
00:53:31.480
that thermodynamic free energy is just the negative elbow.
link |
00:53:35.040
It's the same mathematical construct.
link |
00:53:38.200
So the very emergence of existence, of structure, of form
link |
00:53:42.520
that can be distinguished from the environment
link |
00:53:45.040
or the thing that is not the thing
link |
00:53:49.240
necessitates the existence of an objective function
link |
00:53:56.120
that it looks as if it is minimizing.
link |
00:53:58.160
It's finding a free energy minima.
link |
00:54:00.280
And so just to clarify, I'm trying to wrap my head around.
link |
00:54:04.880
So the free energy principle says that if something exists,
link |
00:54:09.560
these are the properties it should display.
link |
00:54:11.520
Yeah.
link |
00:54:12.520
So what that means is we can't just look,
link |
00:54:17.520
we can't just go into a soup and there's no mechanism.
link |
00:54:21.320
Free energy principle doesn't give us a mechanism
link |
00:54:23.480
to find the things that exist.
link |
00:54:25.680
Is that what's implying, is being implied
link |
00:54:28.680
that you can kind of use it to reason,
link |
00:54:33.200
to think about like, study a particular system
link |
00:54:36.040
and say, does this exhibit these qualities?
link |
00:54:40.600
That's an excellent question.
link |
00:54:42.400
But to answer that, I'd have to return
link |
00:54:44.680
to your previous question about what's the difference
link |
00:54:46.240
between living and nonliving things.
link |
00:54:48.520
Yes, well, actually, sorry.
link |
00:54:51.880
So yeah, maybe we can go there.
link |
00:54:54.080
Maybe we can go there, you kind of drew a line
link |
00:54:57.160
and forgive me for the stupid questions,
link |
00:54:58.960
but you kind of drew a line between living and existing.
link |
00:55:02.560
Is there an interesting sort of distinction?
link |
00:55:07.200
Yeah, I think there is.
link |
00:55:09.600
So things do exist, grains of sand,
link |
00:55:15.240
rocks on the moon, trees, you.
link |
00:55:19.480
So all of these things can be separated from the environment
link |
00:55:24.960
in which they are immersed.
link |
00:55:26.320
And therefore, they must at some level
link |
00:55:28.200
be optimizing their free energy,
link |
00:55:32.320
taking this sort of model evidence interpretation
link |
00:55:36.200
of this quantity that basically means
link |
00:55:38.160
they're self evidencing.
link |
00:55:39.640
Another nice little twist of phrase here
link |
00:55:42.840
is that you are your own existence proof,
link |
00:55:45.520
statistically speaking, which I don't think
link |
00:55:48.880
I said that, somebody did, but I love that phrase.
link |
00:55:53.480
You are your own existence proof.
link |
00:55:55.600
Yeah, so it's so existential, isn't it?
link |
00:55:59.280
I'm gonna have to think about that for a few days.
link |
00:56:01.440
That's a beautiful line.
link |
00:56:06.080
So the step through to answer your question
link |
00:56:09.760
about what's it good for,
link |
00:56:13.840
we'll go along the following lines.
link |
00:56:15.760
First of all, you have to define what it means
link |
00:56:18.720
to exist, which now, as you've rightly pointed out,
link |
00:56:22.040
you have to define what probabilistic properties
link |
00:56:25.000
must the states of something possess
link |
00:56:27.440
so it knows where it finishes.
link |
00:56:30.600
And then you write that down in terms
link |
00:56:32.720
of statistical dependencies, again, sparsity.
link |
00:56:36.000
Again, it's not what's connected or what's correlated
link |
00:56:39.680
or what depends upon, it's what's not correlated
link |
00:56:43.720
and what doesn't depend upon something.
link |
00:56:45.960
Again, it comes down to the deep structures,
link |
00:56:49.680
not in this instance, hierarchical,
link |
00:56:50.920
but the structures that emerge
link |
00:56:54.040
from removing connectivity and dependency.
link |
00:56:56.920
And in this instance, basically being able
link |
00:57:00.160
to identify the surface of the oil drop
link |
00:57:02.720
from the water in which it is immersed.
link |
00:57:06.480
And when you do that, you start to realize,
link |
00:57:09.120
well, there are actually four kinds of states
link |
00:57:12.720
in any given universe that contains anything.
link |
00:57:15.680
The things that are internal to the surface,
link |
00:57:18.840
the things that are external to the surface
link |
00:57:20.680
and the surface in and of itself,
link |
00:57:22.560
which is why I use a metaphor,
link |
00:57:24.040
a little single celled organism
link |
00:57:25.440
that has an interior and exterior
link |
00:57:27.080
and then the surface of the cell.
link |
00:57:29.520
And that's mathematically a Markov blanket.
link |
00:57:32.720
Just to pause, I'm in awe of this concept
link |
00:57:34.960
that there's the stuff outside the surface,
link |
00:57:36.560
stuff inside the surface and the surface itself,
link |
00:57:38.960
the Markov blanket.
link |
00:57:40.240
It's just the most beautiful kind of notion
link |
00:57:43.680
about trying to explore what it means
link |
00:57:46.160
to exist mathematically.
link |
00:57:48.240
I apologize, it's just a beautiful idea.
link |
00:57:50.720
But it came out of California, so that's.
link |
00:57:53.080
I changed my mind.
link |
00:57:54.000
I take it all back.
link |
00:57:55.080
So anyway, so you were just talking
link |
00:57:59.440
about the surface, about the Markov blanket.
link |
00:58:01.280
So this surface or these blanket states
link |
00:58:04.680
that are the, because they are now defined
link |
00:58:09.680
in relation to these independencies
link |
00:58:17.560
and what different states internal blanket
link |
00:58:21.560
or external states can,
link |
00:58:23.840
which ones can influence each other
link |
00:58:25.280
and which cannot influence each other.
link |
00:58:27.760
You can now apply standard results
link |
00:58:30.960
that you would find in non equilibrium physics
link |
00:58:33.600
or steady state or thermodynamics or hydrodynamics,
link |
00:58:37.880
usually out of equilibrium solutions
link |
00:58:41.640
and apply them to this partition.
link |
00:58:43.200
And what it looks like is if all the normal gradient flows
link |
00:58:48.120
that you would associate with any non equilibrium system
link |
00:58:52.080
apply in such a way that part of the Markov blanket
link |
00:58:57.600
and the internal states seem to be hill climbing
link |
00:59:01.560
or doing a gradient descent on the same quantity.
link |
00:59:05.840
And that means that you can now describe
link |
00:59:09.360
the very existence of this oil drop.
link |
00:59:13.120
You can write down the existence of this oil drop
link |
00:59:16.000
in terms of flows, dynamics, equations of motion,
link |
00:59:20.600
where the blanket states or part of them,
link |
00:59:24.160
we call them active states and the internal states
link |
00:59:28.040
now seem to be and must be trying to look
link |
00:59:32.680
as if they're minimizing the same function,
link |
00:59:35.480
which is a low probability of occupying these states.
link |
00:59:39.280
Interesting thing is that what would they be called
link |
00:59:44.040
if you were trying to describe these things?
link |
00:59:45.680
So what we're talking about are internal states,
link |
00:59:50.080
external states and blanket states.
link |
00:59:52.040
Now let's carve the blanket states
link |
00:59:54.080
into two sensory states and active states.
link |
00:59:57.200
Operationally, it has to be the case
link |
00:59:59.520
that in order for this carving up
link |
01:00:01.720
into different sets of states to exist,
link |
01:00:04.440
the active states, the Markov blanket
link |
01:00:06.800
cannot be influenced by the external states.
link |
01:00:09.800
And we already know that the internal states
link |
01:00:11.560
can't be influenced by the external states
link |
01:00:13.640
because the blanket separates them.
link |
01:00:15.720
So what does that mean?
link |
01:00:16.560
Well, it means the active states, the internal states
link |
01:00:19.280
are now jointly not influenced by external states.
link |
01:00:23.440
They only have autonomous dynamics.
link |
01:00:26.120
So now you've got a picture of an oil drop
link |
01:00:30.000
that has autonomy, it has autonomous states,
link |
01:00:34.040
it has autonomous states in the sense
link |
01:00:35.440
that there must be some parts of the surface of the oil drop
link |
01:00:38.320
that are not influenced by the external states
link |
01:00:40.400
and all the interior.
link |
01:00:41.880
And together, those two states endow
link |
01:00:44.520
even a little oil drop with autonomous states
link |
01:00:47.640
that look as if they are optimizing
link |
01:00:51.360
their variational free energy or their negative elbow,
link |
01:00:56.200
their moral evidence.
link |
01:00:59.400
And that would be an interesting intellectual exercise.
link |
01:01:03.240
And you could say, you could even go into the realms
link |
01:01:05.600
of panpsychism, that everything that exists
link |
01:01:08.160
is implicitly making inferences on self evidencing.
link |
01:01:13.000
Now we make the next move, but what about living things?
link |
01:01:17.040
I mean, so let me ask you,
link |
01:01:19.200
what's the difference between an oil drop
link |
01:01:21.600
and a little tadpole or a little lava or a plankton?
link |
01:01:27.200
The picture was just painted of an oil drop.
link |
01:01:30.720
Just immediately in a matter of minutes
link |
01:01:32.840
took me into the world of panpsychism,
link |
01:01:35.200
where you've just convinced me,
link |
01:01:38.040
made me feel like an oil drop is a living,
link |
01:01:41.240
it's certainly an autonomous system,
link |
01:01:43.400
but almost a living system.
link |
01:01:44.720
So it has sensory capabilities and acting capabilities
link |
01:01:48.960
and it maintains something.
link |
01:01:50.600
So what is the difference between that
link |
01:01:53.920
and something that we traditionally
link |
01:01:56.120
think of as a living system?
link |
01:01:59.880
That it could die or it can't,
link |
01:02:02.200
I mean, yeah, mortality, I'm not exactly sure.
link |
01:02:05.240
I'm not sure what the right answer there is
link |
01:02:08.640
because they can move,
link |
01:02:09.640
like movement seems like an essential element
link |
01:02:11.840
to being able to act in the environment,
link |
01:02:13.520
but the oil drop is doing that.
link |
01:02:15.800
So I don't know.
link |
01:02:16.640
Is it?
link |
01:02:18.120
The oil drop will be moved,
link |
01:02:19.720
but does it in and of itself move autonomously?
link |
01:02:22.640
Well, the surface is performing actions
link |
01:02:26.880
that maintain its structure.
link |
01:02:29.680
Yeah, you're being too clever.
link |
01:02:30.800
I was, I had in mind a passive little oil drop
link |
01:02:34.560
that's sitting there at the bottom
link |
01:02:37.680
on the top of a glass of water.
link |
01:02:39.240
Sure, I guess.
link |
01:02:40.600
What I'm trying to say is you're absolutely right.
link |
01:02:42.200
You've nailed it.
link |
01:02:44.400
It's movement.
link |
01:02:45.880
So where does that movement come from?
link |
01:02:47.200
If it comes from the inside,
link |
01:02:49.400
then you've got, I think, something that's living.
link |
01:02:53.320
What do you mean from the inside?
link |
01:02:54.680
What I mean is that the internal states
link |
01:02:58.840
that can influence the active states,
link |
01:03:01.040
where the active states can influence,
link |
01:03:02.600
but they're not influenced by the external states,
link |
01:03:05.120
can cause movement.
link |
01:03:07.160
So there are two types of oil drops, if you like.
link |
01:03:10.440
There are oil drops where the internal states
link |
01:03:12.800
are so random that they average themselves away,
link |
01:03:20.320
and the thing cannot, on average,
link |
01:03:23.800
when you do the averaging, move.
link |
01:03:25.960
So a nice example of that would be the Sun.
link |
01:03:29.400
The Sun certainly has internal states.
link |
01:03:31.160
There's lots of intrinsic autonomous activity going on,
link |
01:03:34.360
but because it's not coordinated,
link |
01:03:35.840
because it doesn't have the deep, in the millennial sense,
link |
01:03:38.080
the hierarchical structure that the brain does,
link |
01:03:40.960
there is no overall mode or pattern or organization
link |
01:03:45.800
that expresses itself on the surface
link |
01:03:48.160
that allows it to actually swim.
link |
01:03:51.400
It can certainly have a very active surface,
link |
01:03:54.080
but en masse, at the scale of the actual surface of the Sun,
link |
01:03:58.280
the average position of that surface cannot, in itself, move,
link |
01:04:02.920
because the internal dynamics are more like a hot gas.
link |
01:04:06.680
They are literally like a hot gas,
link |
01:04:08.480
whereas your internal dynamics are much more structured
link |
01:04:11.440
and deeply structured,
link |
01:04:12.880
and now you can express on your active states
link |
01:04:16.440
with your muscles and your secretory organs,
link |
01:04:19.720
your autonomic nervous system and its effectors,
link |
01:04:22.920
you can actually move, and that's all you can do.
link |
01:04:26.760
And that's something which,
link |
01:04:28.240
if you haven't thought of it like this before,
link |
01:04:30.440
I think it's nice to just realize
link |
01:04:32.440
there is no other way that you can change the universe
link |
01:04:37.040
other than simply moving.
link |
01:04:39.240
Whether that moving is articulating with my voice box
link |
01:04:43.840
or walking around or squeezing juices
link |
01:04:46.600
out of my secretory organs,
link |
01:04:48.720
there's only one way you can change the universe.
link |
01:04:51.960
It's moving.
link |
01:04:53.240
And the fact that you do so nonrandomly makes you alive.
link |
01:04:58.600
Yeah, so it's that nonrandomness.
link |
01:05:00.600
And that would be manifested,
link |
01:05:04.840
we realize, in terms of essentially swimming,
link |
01:05:07.800
essentially moving, changing one's shape,
link |
01:05:10.480
a morphogenesis that is dynamic and possibly adaptive.
link |
01:05:15.760
So that's what I was trying to get at
link |
01:05:17.920
between the difference between the oil drop
link |
01:05:19.160
and the little tadpole.
link |
01:05:21.000
The tadpole is moving around.
link |
01:05:23.000
Its active states are actually changing the external states.
link |
01:05:26.920
And there's now a cycle,
link |
01:05:28.480
an action perception cycle, if you like,
link |
01:05:30.400
a recurrent dynamic that's going on
link |
01:05:34.040
that depends upon this deeply structured autonomous behavior
link |
01:05:39.360
that rests upon internal dynamics
link |
01:05:44.600
that are not only modeling
link |
01:05:48.880
the data impressed upon their surface or the blanket states,
link |
01:05:53.800
but they are actively resampling those data by moving.
link |
01:05:58.800
They're moving towards chemical gradients and chemotaxis.
link |
01:06:03.920
So they've gone beyond just being good little models
link |
01:06:08.280
of the kind of world they live in.
link |
01:06:11.120
For example, an oil droplet could, in a panpsychic sense,
link |
01:06:15.960
be construed as a little being
link |
01:06:18.440
that has now perfectly inferred.
link |
01:06:20.720
It's a passive, nonliving oil drop
link |
01:06:23.800
living in a bowl of water.
link |
01:06:25.600
No problem.
link |
01:06:27.960
But to now equip that oil drop with the ability to go out
link |
01:06:31.040
and test that hypothesis about different states of beings.
link |
01:06:34.040
So it can actually push its surface over there, over there,
link |
01:06:36.920
and test for chemical gradients,
link |
01:06:38.680
or then you start to move to a much more lifelike form.
link |
01:06:42.880
This is all fun, theoretically interesting,
link |
01:06:44.920
but it actually is quite important
link |
01:06:47.000
in terms of reflecting what I have seen
link |
01:06:49.800
since the turn of the millennium,
link |
01:06:53.120
which is this move towards an inactive
link |
01:06:56.600
and embodied understanding of intelligence.
link |
01:07:00.760
And you say you're from machine learning.
link |
01:07:03.760
So what that means,
link |
01:07:05.720
the central importance of movement,
link |
01:07:10.680
I think has yet to really hit machine learning.
link |
01:07:14.000
It certainly has now diffused itself throughout robotics.
link |
01:07:20.400
And perhaps you could say certain problems in active vision
link |
01:07:23.200
where you actually have to move the camera
link |
01:07:25.360
to sample this and that.
link |
01:07:27.240
But machine learning of the data mining deep learning sort
link |
01:07:31.600
simply hasn't contended with this issue.
link |
01:07:34.000
What it's done, instead of dealing with the movement problem
link |
01:07:37.240
and the active sampling of data,
link |
01:07:39.160
it's just said, we don't need to worry about,
link |
01:07:40.720
we can see all the data because we've got big data.
link |
01:07:43.120
So we can ignore movement.
link |
01:07:45.120
So that for me is an important omission
link |
01:07:50.920
in current machine learning.
link |
01:07:52.200
The current machine learning is much more like the oil drop.
link |
01:07:54.800
Yes.
link |
01:07:55.960
But an oil drop that enjoys exposure
link |
01:07:59.480
to nearly all the data that it will ever need to be exposed to,
link |
01:08:03.600
as opposed to the tadpoles swimming out
link |
01:08:05.760
to find the right data.
link |
01:08:07.360
For example, it likes food.
link |
01:08:10.320
That's a good hypothesis.
link |
01:08:11.240
Let's test it out.
link |
01:08:12.080
Let's go and move and ingest food, for example,
link |
01:08:15.720
and see is that evidence that I'm the kind of thing
link |
01:08:18.680
that likes this kind of food.
link |
01:08:20.400
So the next natural question, and forgive this question,
link |
01:08:24.000
but if we think of sort of even artificial intelligence
link |
01:08:27.120
systems, which I just painted a beautiful picture
link |
01:08:29.440
of existence and life.
link |
01:08:32.840
So do you ascribe, do you find within this framework
link |
01:08:39.920
a possibility of defining consciousness
link |
01:08:45.240
or exploring the idea of consciousness?
link |
01:08:47.360
Like what, you know, self awareness
link |
01:08:52.760
and expand it to consciousness?
link |
01:08:55.160
Yeah.
link |
01:08:56.160
How can we start to think about consciousness
link |
01:08:58.840
within this framework?
link |
01:08:59.720
Is it possible?
link |
01:09:00.920
Well, yeah, I think it's possible to think about it,
link |
01:09:03.160
whether you'll get it.
link |
01:09:04.320
Get anywhere is another question.
link |
01:09:06.400
And again, I'm not sure that I'm licensed
link |
01:09:10.320
to answer that question.
link |
01:09:12.760
I think you'd have to speak to a qualified philosopher
link |
01:09:15.120
to get a definitive answer to that.
link |
01:09:17.320
But certainly, there's a lot of interest
link |
01:09:19.680
in using not just these ideas, but related ideas
link |
01:09:23.400
from information theory to try and tie down
link |
01:09:27.960
the maths and the calculus and the geometry of consciousness,
link |
01:09:34.080
either in terms of sort of a minimal consciousness,
link |
01:09:39.720
even less than a minimal selfhood.
link |
01:09:42.440
And what I'm talking about is the ability, effectively,
link |
01:09:48.320
to plan, to have agency.
link |
01:09:52.440
So you could argue that a virus does have a form of agency
link |
01:09:57.760
in virtue of the way that it selectively
link |
01:10:00.360
finds hosts and cells to live in and moves around.
link |
01:10:05.880
But you wouldn't endow it with the capacity
link |
01:10:09.280
to think about planning and moving in a purposeful way
link |
01:10:14.880
where it countenances the future.
link |
01:10:17.280
Whereas you might an ant.
link |
01:10:18.680
You might think an ant's not quite as unconscious
link |
01:10:22.040
as a virus.
link |
01:10:24.280
It certainly seems to have a purpose.
link |
01:10:26.160
It talks to its friends en route during its foraging.
link |
01:10:29.680
It has a different kind of autonomy, which is biotic,
link |
01:10:37.160
but beyond a virus.
link |
01:10:38.720
So there's something about, so there's
link |
01:10:41.400
some line that has to do with the complexity of planning
link |
01:10:45.680
that may contain an answer.
link |
01:10:48.000
I mean, it would be beautiful if we
link |
01:10:49.960
can find a line beyond which we could say a being is conscious.
link |
01:10:55.480
Yes, it will be.
link |
01:10:56.560
These are wonderful lines that we've drawn with existence,
link |
01:11:00.800
life, and consciousness.
link |
01:11:02.680
Yes, it will be very nice.
link |
01:11:05.320
One little wrinkle there, and this
link |
01:11:07.280
is something I've only learned in the past few months,
link |
01:11:09.400
is the philosophical notion of vagueness.
link |
01:11:12.400
So you're saying it would be wonderful to draw a line.
link |
01:11:14.840
I had always assumed that that line at some point
link |
01:11:18.080
would be drawn until about four months ago,
link |
01:11:22.720
and the philosopher taught me about vagueness.
link |
01:11:24.880
So I don't know if you've come across this,
link |
01:11:26.720
but it's a technical concept.
link |
01:11:28.440
And I think most revealingly illustrated with,
link |
01:11:33.480
at what point does a pile of sand become a pile?
link |
01:11:37.120
Is it one grain, two grains, three grains, or four grains?
link |
01:11:41.620
So at what point would you draw the line
link |
01:11:44.200
between being a pile of sand and a collection of grains of sand?
link |
01:11:51.320
In the same way, is it right to ask,
link |
01:11:53.520
where would I draw the line between conscious
link |
01:11:55.560
and unconscious?
link |
01:11:56.740
And it might be a vague concept.
link |
01:11:59.680
Having said that, I agree with you entirely.
link |
01:12:02.920
Systems that have the ability to plan.
link |
01:12:06.480
So just technically, what that means
link |
01:12:08.440
is your inferential self evidencing,
link |
01:12:13.880
by which I simply mean the thermodynamics and gradient
link |
01:12:19.880
flows that underwrite the preservation of your oil
link |
01:12:22.960
droplet like form, can be described
link |
01:12:29.080
as an optimization of log Bayesian model
link |
01:12:32.560
evidence, your elbow.
link |
01:12:36.680
That self evidencing must be evidence
link |
01:12:39.760
for a model of what's causing the sensory impressions
link |
01:12:44.000
on the sensory part of your surface or your Markov
link |
01:12:47.040
blanket.
link |
01:12:48.520
If that model is capable of planning,
link |
01:12:51.160
it must include a model of the future consequences
link |
01:12:53.800
of your active states or your action, just planning.
link |
01:12:56.800
So we're now in the game of planning as inference.
link |
01:12:59.200
Now notice what we've made, though.
link |
01:13:00.620
We've made quite a big move away from big data and machine
link |
01:13:04.400
learning, because again, it's the consequences of moving.
link |
01:13:08.520
It's the consequences of selecting those data or those
link |
01:13:11.600
data or looking over there.
link |
01:13:14.240
And that tells you immediately that even
link |
01:13:17.420
to be a contender for a conscious artifact or a strong
link |
01:13:22.760
AI or generalized, I don't know what that's called nowadays,
link |
01:13:26.760
then you've got to have movement in the game.
link |
01:13:29.280
And furthermore, you've got to have a generative model
link |
01:13:32.560
of the sort you might find in, say, a variational auto
link |
01:13:34.920
encoder that is thinking about the future conditioned
link |
01:13:39.840
upon different courses of action.
link |
01:13:41.900
Now that brings a number of things to the table, which
link |
01:13:44.400
now you start to think, well, those
link |
01:13:46.280
have got all the right ingredients
link |
01:13:47.400
to talk about consciousness.
link |
01:13:48.640
I've now got to select among a number of different courses
link |
01:13:51.600
of action into the future as part of planning.
link |
01:13:54.920
I've now got free will.
link |
01:13:56.440
The act of selecting this course of action or that policy
link |
01:13:59.440
or that policy or that action suddenly
link |
01:14:02.480
makes me into an inference machine,
link |
01:14:04.720
a self evidencing artifact that now
link |
01:14:09.480
looks as if it's selecting amongst different alternative
link |
01:14:12.480
ways forward, as I actively swim here or swim there
link |
01:14:15.480
or look over here, look over there.
link |
01:14:17.920
So I think you've now got to a situation
link |
01:14:19.920
if there is planning in the mix.
link |
01:14:22.240
You're now getting much closer to that line
link |
01:14:25.200
if that line were ever to exist.
link |
01:14:27.360
I don't think it gets you quite as far as self aware, though.
link |
01:14:32.400
And then you have to, I think, grapple with the question,
link |
01:14:39.680
how would formally write down a calculus or a maths
link |
01:14:43.160
of self awareness?
link |
01:14:44.620
I don't think it's impossible to do.
link |
01:14:47.840
But I think there would be pressure on you
link |
01:14:51.400
to actually commit to a formal definition of what
link |
01:14:53.360
you mean by self awareness.
link |
01:14:55.920
I think most people that I know would probably
link |
01:15:00.480
say that a goldfish, a pet fish, was not self aware.
link |
01:15:07.000
They would probably argue about their favorite cat,
link |
01:15:10.600
but would be quite happy to say that their mom was self aware.
link |
01:15:14.000
So.
link |
01:15:15.480
I mean, but that might very well connect
link |
01:15:17.400
to some level of complexity with planning.
link |
01:15:20.960
It seems like self awareness is essential for complex planning.
link |
01:15:26.520
Yeah.
link |
01:15:27.240
Do you want to take that further?
link |
01:15:28.080
Because I think you're absolutely right.
link |
01:15:29.880
Again, the line is unclear, but it
link |
01:15:31.400
seems like integrating yourself into the world,
link |
01:15:36.400
into your planning, is essential for constructing complex plans.
link |
01:15:42.320
Yes.
link |
01:15:43.080
Yeah.
link |
01:15:43.720
So mathematically describing that in the same elegant way
link |
01:15:47.440
as you have with the free energy principle might be difficult.
link |
01:15:51.120
Well, yes and no.
link |
01:15:53.360
I don't think that, well, perhaps we should just,
link |
01:15:55.360
can we just go back?
link |
01:15:57.000
That's a very important answer you gave.
link |
01:15:58.720
And I think if I just unpacked it,
link |
01:16:01.880
you'd see the truisms that you've just exposed for us.
link |
01:16:06.480
But let me, sorry, I'm mindful that I didn't answer
link |
01:16:10.400
your question before.
link |
01:16:11.360
Well, what's the free energy principle good for?
link |
01:16:13.880
Is it just a pretty theoretical exercise
link |
01:16:15.680
to explain nonequilibrium steady states?
link |
01:16:17.880
Yes, it is.
link |
01:16:19.400
It does nothing more for you than that.
link |
01:16:21.360
It can be regarded, it's going to sound very arrogant,
link |
01:16:24.040
but it is of the sort of theory of natural selection,
link |
01:16:27.880
or a hypothesis of natural selection.
link |
01:16:32.800
Beautiful, undeniably true, but tells you
link |
01:16:37.200
absolutely nothing about why you have legs and eyes.
link |
01:16:42.040
It tells you nothing about the actual phenotype,
link |
01:16:44.720
and it wouldn't allow you to build something.
link |
01:16:48.080
So the free energy principle by itself
link |
01:16:51.120
is as vacuous as most tautological theories.
link |
01:16:54.880
And by tautological, of course,
link |
01:16:56.200
I'm talking to the theory of natural,
link |
01:16:58.800
the survival of the fittest.
link |
01:17:00.080
What's the fittest of those that survive?
link |
01:17:01.760
Why do they cycle?
link |
01:17:02.600
It's the fitter.
link |
01:17:03.440
It just goes around in circles.
link |
01:17:05.640
In a sense, the free energy principle has that same
link |
01:17:08.560
deflationary tautology under the hood.
link |
01:17:15.000
It's a characteristic of things that exist.
link |
01:17:17.680
Why do they exist?
link |
01:17:18.520
Because they minimize their free energy.
link |
01:17:19.720
Why do they minimize their free energy?
link |
01:17:21.400
Because they exist.
link |
01:17:22.440
And you just keep on going round and round and round.
link |
01:17:24.720
But the practical thing,
link |
01:17:28.080
which you don't get from natural selection,
link |
01:17:32.680
but you could say has now manifest in things
link |
01:17:35.680
like differential evolution or genetic algorithms
link |
01:17:38.160
and MCMC, for example, in machine learning.
link |
01:17:41.320
The practical thing you can get is,
link |
01:17:43.240
if it looks as if things that exist
link |
01:17:45.440
are trying to have density dynamics
link |
01:17:49.400
and look as though they're optimizing
link |
01:17:51.560
a variational free energy,
link |
01:17:53.320
and a variational free energy has to be
link |
01:17:55.200
a functional of a generative model,
link |
01:17:57.280
a probabilistic description of causes and consequences,
link |
01:18:01.720
causes out there, consequences in the sensorium
link |
01:18:04.560
on the sensory parts of the Markov blanket,
link |
01:18:07.080
then it should, in theory, be possible
link |
01:18:08.680
to write down the generative model,
link |
01:18:10.400
work out the gradients,
link |
01:18:11.840
and then cause it to autonomously self evidence.
link |
01:18:15.920
So you should be able to write down oil droplets.
link |
01:18:18.160
You should be able to create artifacts
link |
01:18:20.080
where you have supplied the objective function
link |
01:18:24.240
that supplies the gradients,
link |
01:18:25.560
that supplies the self organizing dynamics
link |
01:18:28.320
to non equilibrium steady state.
link |
01:18:30.280
So there is actually a practical application
link |
01:18:32.680
of the free energy principle
link |
01:18:34.120
when you can write down your required evidence
link |
01:18:37.800
in terms of, well, when you can write down
link |
01:18:40.400
the generative model that is the thing
link |
01:18:43.720
that has the evidence.
link |
01:18:44.800
The probability of these sensory data
link |
01:18:46.760
or this data, given that model,
link |
01:18:50.880
is effectively the thing that the elbow
link |
01:18:54.200
or the variational free energy bounds or approximates.
link |
01:18:57.400
That means that you can actually write down the model
link |
01:19:01.720
and the kind of thing that you want to engineer,
link |
01:19:04.640
the kind of AGI or artificial general intelligence
link |
01:19:10.440
that you want to manifest probabilistically,
link |
01:19:14.480
and then you engineer, a lot of hard work,
link |
01:19:16.720
but you would engineer a robot and a computer
link |
01:19:19.760
to perform a gradient descent on that objective function.
link |
01:19:23.400
So it does have a practical implication.
link |
01:19:26.160
Now, why am I wittering on about that?
link |
01:19:27.440
It did seem relevant to, yes.
link |
01:19:28.880
So what kinds of, so the answer to,
link |
01:19:32.920
would it be easier or would it be hard?
link |
01:19:34.240
Well, mathematically, it's easy.
link |
01:19:36.200
I've just told you all you need to do
link |
01:19:38.000
is write down your perfect artifact,
link |
01:19:43.880
probabilistically, in the form
link |
01:19:45.280
of a probabilistic generative model,
link |
01:19:46.480
a probability distribution over the causes
link |
01:19:48.800
and consequences of the world
link |
01:19:52.520
in which this thing is immersed.
link |
01:19:54.680
And then you just engineer a computer and a robot
link |
01:19:58.040
to perform a gradient descent on that objective function.
link |
01:20:00.800
No problem.
link |
01:20:02.720
But of course, the big problem
link |
01:20:04.160
is writing down the generative model.
link |
01:20:05.960
So that's where the heavy lifting comes in.
link |
01:20:08.040
So it's the form and the structure of that generative model
link |
01:20:12.160
which basically defines the artifact that you will create
link |
01:20:15.640
or, indeed, the kind of artifact that has self awareness.
link |
01:20:19.600
So that's where all the hard work comes,
link |
01:20:22.080
very much like natural selection doesn't tell you
link |
01:20:24.640
in the slightest why you have eyes.
link |
01:20:27.000
So you have to drill down on the actual phenotype,
link |
01:20:29.480
the actual generative model.
link |
01:20:31.480
So with that in mind, what did you tell me
link |
01:20:36.360
that tells me immediately the kinds of generative models
link |
01:20:40.720
I would have to write down in order to have self awareness?
link |
01:20:43.480
What you said to me was I have to have a model
link |
01:20:48.200
that is effectively fit for purpose
link |
01:20:50.440
for this kind of world in which I operate.
link |
01:20:53.680
And if I now make the observation
link |
01:20:55.880
that this kind of world is effectively largely populated
link |
01:20:59.120
by other things like me, i.e. you,
link |
01:21:02.360
then it makes enormous sense
link |
01:21:04.200
that if I can develop a hypothesis
link |
01:21:07.280
that we are similar kinds of creatures,
link |
01:21:11.520
in fact, the same kind of creature,
link |
01:21:13.640
but I am me and you are you,
link |
01:21:16.320
then it becomes, again, mandated to have a sense of self.
link |
01:21:21.440
So if I live in a world
link |
01:21:23.640
that is constituted by things like me,
link |
01:21:26.560
basically a social world, a community,
link |
01:21:29.520
then it becomes necessary now for me to infer
link |
01:21:32.320
that it's me talking and not you talking.
link |
01:21:34.400
I wouldn't need that if I was on Mars by myself
link |
01:21:37.280
or if I was in the jungle as a feral child.
link |
01:21:40.040
If there was nothing like me around,
link |
01:21:43.120
there would be no need to have an inference
link |
01:21:46.200
at a hypothesis, oh yes, it is me
link |
01:21:49.160
that is experiencing or causing these sounds
link |
01:21:51.240
and it is not you.
link |
01:21:52.360
It's only when there's ambiguity in play
link |
01:21:54.680
induced by the fact that there are others in that world.
link |
01:21:58.240
So I think that the special thing about self aware artifacts
link |
01:22:03.520
is that they have learned to, or they have acquired,
link |
01:22:08.280
or at least are equipped with, possibly by evolution,
link |
01:22:11.680
generative models that allow for the fact
link |
01:22:14.560
there are lots of copies of things like them around,
link |
01:22:17.360
and therefore they have to work out it's you and not me.
link |
01:22:20.600
That's brilliant.
link |
01:22:23.280
I've never thought of that.
link |
01:22:24.560
I never thought of that, that the purpose
link |
01:22:28.160
of the really usefulness of consciousness
link |
01:22:31.360
or self awareness in the context of planning
link |
01:22:34.200
existing in the world is so you can operate
link |
01:22:36.920
with other things like you, and like you could,
link |
01:22:39.240
it doesn't have to necessarily be human.
link |
01:22:40.800
It could be other kind of similar creatures.
link |
01:22:43.480
Absolutely, well, we view a lot of our attributes
link |
01:22:46.080
into our pets, don't we?
link |
01:22:47.880
Or we try to make our robots humanoid.
link |
01:22:49.920
And I think there's a deep reason for that,
link |
01:22:51.840
that it's just much easier to read the world
link |
01:22:54.680
if you can make the simplifying assumption
link |
01:22:56.240
that basically you're me, and it's just your turn to talk.
link |
01:23:00.040
I mean, when we talk about planning,
link |
01:23:01.520
when you talk specifically about planning,
link |
01:23:04.200
the highest, if you like, manifestation or realization
link |
01:23:07.800
of that planning is what we're doing now.
link |
01:23:09.600
I mean, the human condition doesn't get any higher
link |
01:23:12.560
than this talking about the philosophy of existence
link |
01:23:16.800
and the conversation.
link |
01:23:17.920
But in that conversation, there is a beautiful art
link |
01:23:23.760
of turn taking and mutual inference, theory of mind.
link |
01:23:28.120
I have to know when you wanna listen.
link |
01:23:29.680
I have to know when you want to interrupt.
link |
01:23:31.120
I have to make sure that you're online.
link |
01:23:32.520
I have to have a model in my head
link |
01:23:34.360
of your model in your head.
link |
01:23:35.800
That's the highest, the most sophisticated form
link |
01:23:38.320
of generative model, where the generative model
link |
01:23:40.160
actually has a generative model
link |
01:23:41.360
of somebody else's generative model.
link |
01:23:42.800
And I think that, and what we are doing now evinces
link |
01:23:47.080
the kinds of generative models
link |
01:23:49.160
that would support self awareness,
link |
01:23:51.280
because without that, we'd both be talking over each other,
link |
01:23:54.640
or we'd be singing together in a choir.
link |
01:23:58.320
That's not a brilliant analogy for what I'm trying to say,
link |
01:24:01.280
but yeah, we wouldn't have this discourse.
link |
01:24:05.160
We wouldn't have this.
link |
01:24:06.000
Yeah, the dance of it.
link |
01:24:06.840
Yeah, that's right.
link |
01:24:07.680
As I interrupt, I mean, that's beautifully put.
link |
01:24:12.680
I'll re listen to this conversation many times.
link |
01:24:17.360
There's so much poetry in this, and mathematics.
link |
01:24:21.600
Let me ask the silliest, or perhaps the biggest question
link |
01:24:26.240
as a last kind of question.
link |
01:24:29.840
We've talked about living in existence
link |
01:24:33.360
and the objective function under which
link |
01:24:35.120
these objects would operate.
link |
01:24:37.480
What do you think is the objective function
link |
01:24:39.720
of our existence?
link |
01:24:41.560
What's the meaning of life?
link |
01:24:44.160
What do you think is the, for you, perhaps,
link |
01:24:47.160
the purpose, the source of fulfillment,
link |
01:24:50.200
the source of meaning for your existence,
link |
01:24:53.080
as one blob in this soup?
link |
01:24:57.640
I'm tempted to answer that, again, as a physicist,
link |
01:25:00.480
until it's the free energy I expect
link |
01:25:03.040
consequent upon my behavior.
link |
01:25:05.480
So technically, we could get a really interesting
link |
01:25:08.400
conversation about what that comprises
link |
01:25:10.720
in terms of searching for information,
link |
01:25:13.160
resolving uncertainty about the kind of thing that I am.
link |
01:25:16.560
But I suspect that you want a slightly more personal
link |
01:25:20.240
and fun answer, but which can be consistent with that.
link |
01:25:25.120
And I think it's reassuringly simple
link |
01:25:30.320
and hops back to what you were taught as a child,
link |
01:25:36.520
that you have certain beliefs about the kind of creature
link |
01:25:39.520
and the kind of person you are.
link |
01:25:41.840
And all that self evidencing means,
link |
01:25:44.840
all that minimizing variational free energy
link |
01:25:46.840
in an inactive and embodied way,
link |
01:25:50.000
means is fulfilling the beliefs about
link |
01:25:53.640
what kind of thing you are.
link |
01:25:55.720
And of course, we're all given those scripts,
link |
01:25:58.040
those narratives, at a very early age,
link |
01:26:01.240
usually in the form of bedtime stories or fairy stories
link |
01:26:04.280
that I'm a princess and I'm gonna meet a beast
link |
01:26:07.160
who's gonna transform and he's gonna be a prince.
link |
01:26:09.360
And so the narratives are all around you
link |
01:26:11.880
from your parents to the friends
link |
01:26:14.560
to the society feeds these stories.
link |
01:26:17.680
And then your objective function is to fulfill.
link |
01:26:21.000
Exactly, that narrative that has been encultured
link |
01:26:24.240
by your immediate family, but as you say,
link |
01:26:27.160
also the sort of the culture in which you grew up
link |
01:26:29.600
and you create for yourself.
link |
01:26:30.880
I mean, again, because of this active inference,
link |
01:26:33.440
this inactive aspect of self evidencing,
link |
01:26:36.680
not only am I modeling my environment,
link |
01:26:40.960
my eco niche, my external states out there,
link |
01:26:44.040
but I'm actively changing them all the time
link |
01:26:46.520
and doing the same back, we're doing it together.
link |
01:26:49.840
So there's a synchrony that means that I'm creating
link |
01:26:53.680
my own culture over different timescales.
link |
01:26:56.800
So the question now is for me being very selfish,
link |
01:27:00.800
what scripts were I given?
link |
01:27:02.240
It basically was a mixture between Einstein and shark homes.
link |
01:27:06.160
So I smoke as heavily as possible,
link |
01:27:09.880
try to avoid too much interpersonal contact,
link |
01:27:15.280
enjoy the fantasy that you're a popular scientist
link |
01:27:21.000
who's gonna make a difference in a slightly quirky way.
link |
01:27:23.400
So that's what I grew up on.
link |
01:27:25.200
My father was an engineer and loved science
link |
01:27:28.320
and he loved sort of things like Sir Arthur Edmonds,
link |
01:27:33.320
Spacetime and Gravitation, which was the first
link |
01:27:37.440
understandable version of general relativity.
link |
01:27:41.800
So all the fairy stories I was told as I was growing up
link |
01:27:45.840
were all about these characters.
link |
01:27:48.640
I'm keeping the Hobbit out of this
link |
01:27:50.600
because that doesn't quite fit my narrative.
link |
01:27:53.280
There's a journey of exploration, I suppose, of sorts.
link |
01:27:56.240
So yeah, I've just grown up to be what I imagine
link |
01:28:01.240
a mild mannered Sherlock Holmes slash Albert Einstein
link |
01:28:05.720
would do in my shoes.
link |
01:28:07.960
And you did it elegantly and beautifully.
link |
01:28:10.080
Carl was a huge honor talking today, it was fun.
link |
01:28:12.440
Thank you so much for your time.
link |
01:28:13.560
No, thank you. Appreciate it.
link |
01:28:15.560
Thank you for listening to this conversation
link |
01:28:17.240
with Carl Friston and thank you
link |
01:28:19.280
to our presenting sponsor, Cash App.
link |
01:28:21.320
Please consider supporting the podcast
link |
01:28:23.080
by downloading Cash App and using code LexPodcast.
link |
01:28:27.080
If you enjoy this podcast, subscribe on YouTube,
link |
01:28:29.720
review it with five stars on Apple Podcast,
link |
01:28:32.040
support on Patreon, or simply connect with me on Twitter
link |
01:28:35.440
at LexFriedman.
link |
01:28:37.480
And now let me leave you with some words from Carl Friston.
link |
01:28:41.520
Your arm moves because you predict it will
link |
01:28:44.600
and your motor system seeks to minimize prediction error.
link |
01:28:48.040
Thank you for listening and hope to see you next time.