back to indexKarl Friston: Neuroscience and the Free Energy Principle | Lex Fridman Podcast #99
link |
The following is a conversation with Carl Friston, one of the greatest neuroscientists in history,
link |
cited over 245,000 times, known for many influential ideas in brain imaging,
link |
neuroscience, and theoretical neurobiology, including especially
link |
the fascinating idea of the free energy principle for action and perception.
link |
Carl's mix of humor, brilliance, and kindness to me are inspiring and captivating. This was a
link |
huge honor and a pleasure. This is the Artificial Intelligence Podcast. If you enjoy it, subscribe
link |
on YouTube, review it with 5 stars on Apple Podcasts, support on Patreon, or simply connect with me
link |
on Twitter. Alex Friedman, spelled F R I D M A N. As usual, I'll do a few minutes of ads now,
link |
and never any ads in the middle that can break the flow of the conversation.
link |
I hope that works for you and doesn't hurt the listening experience.
link |
This show is presented by Cash App, the number one finance app in the App Store. When you get it,
link |
use code LEX Podcast. Cash App lets you send money to friends by Bitcoin and invest in the
link |
stock market with as little as $1. Since Cash App allows you to send and receive money digitally,
link |
let me mention a surprising fact related to physical money. Of all the currency in the world,
link |
roughly 8% of it is actual physical money. The other 92% of money only exists digitally.
link |
So again, if you get Cash App from the App Store, Google Play, and use the code LEX Podcast,
link |
you get $10, and Cash App will also donate $10 the first, an organization that is helping to
link |
advance robotics and STEM education for young people around the world. And now, here's my
link |
conversation with Carl Friston. How much of the human brain do we understand from
link |
the low level of neuronal communication to the functional level to the highest level,
link |
maybe the psychiatric disorder level? Well, we're certainly in a better position than we were last
link |
century. How far we've got to go, I think, is almost an answerable question. So you'd have to
link |
set the parameters, what constitutes understanding, what level of understanding do you want.
link |
I think we've made enormous progress in terms of broad brush principles, whether that affords a
link |
detailed cartography of the functional anatomy of the brain and what it does, and write down to
link |
the microcircuitry and the neurons, that's probably out of reach at the present time.
link |
So the cartography, so mapping the brain. Do you think mapping of the brain,
link |
the detailed, perfect imaging of it, does that get us closer to understanding of the mind
link |
of the brain? So how far does it get us if we have that perfect cartography of the brain?
link |
I think there are lower bounds on that. It's a really interesting question. It would determine
link |
this sort of scientific career you'd pursue if you believe that knowing every dendritic connection,
link |
every sort of microscopic synaptic structure, right down to the molecular level, was going to
link |
give you the right kind of information to understand the computational anatomy. Then you'd
link |
choose to be a microscopist and you would study little cubic millimeters of brain for the rest
link |
of your life. If on the other hand you were interested in holistic functions and a sort
link |
of functional anatomy of the sort that a neuropsychologist would understand, you'd study
link |
brain lesions and strokes, just looking at the whole person. So again, it comes back to at what
link |
level do you want understanding. I think there are principled reasons not to go too far. If you
link |
commit to a view of the brain as a machine that's performing a form of inference and representing
link |
things, that level of understanding is necessarily cast in terms of probability densities and ensemble
link |
densities, distributions. What that tells you is that you don't really want to look at the atoms
link |
to understand the thermodynamics of probabilistic descriptions for how the brain works.
link |
So I personally wouldn't look at the molecules or indeed the single neurons in the same way if I
link |
wanted to understand the thermodynamics of some nonequilibrium steady state of a gas or an active
link |
material. I wouldn't spend my life looking at the individual molecules that constitute that ensemble.
link |
I'd look at that collective behavior. On the other hand, if you go to coarse grain, you're going to
link |
miss some basic canonical principles of connectivity and architectures. I'm thinking here
link |
this bit colloquial, but there's current excitement about high field magnetic resonance imaging at
link |
7 Tesla. Why? Well, it gives us for the first time the opportunity to look at the brain in action
link |
at the level of a few millimeters that distinguish between different layers of the cortex that may
link |
be very important in terms of evincing generic principles of canonical microcircuitry that
link |
are replicated throughout the brain that may tell us something fundamental about message passing in
link |
the brain and these density dynamics or neuronal ensemble population dynamics that underwrite
link |
our brain function. So somewhere between a millimeter and a meter.
link |
Yeah. Lingering for a bit on the big questions, if you allow me. What to use the most beautiful
link |
or surprising characteristic of the human brain? I think it's hierarchical and recursive aspect,
link |
it's recurrent aspect. Of the structure or of the actual representation or power of the brain?
link |
Well, I think one speaks to the other. I was actually answering in a delminded way from the
link |
point of view of purely its anatomy and its structural aspects. I mean, there are many
link |
marvelous organs in the body. Let's take your liver, for example. Without it, you wouldn't
link |
be around for very long and it does some beautiful and delicate biochemistry and homeostasis and
link |
evolved with a finesse that would easily parallel the brain, but it doesn't have a beautiful anatomy.
link |
It has a simple anatomy, which is attractive in a minimalist sense, but it doesn't have that
link |
crafted structure of sparse connectivity and that recurrence and that specialization
link |
that the brain has. So you said a lot of interesting terms here. So the recurrence,
link |
the sparsity, but you also started by saying hierarchical. So I've never thought of our
link |
brain as hierarchical. I always thought it's just like a giant, interconnected mess where
link |
it's very difficult to figure anything out. But in what sense do you see the brain as hierarchical?
link |
Well, I see it. It's not a magic soup. Of course, it's what I used to think when I was
link |
before I studied medicine and the like. So a lot of those terms imply each other.
link |
So hierarchies, if you just think about the nature of a hierarchy, how would you actually
link |
build one? And what you would have to do is basically carefully remove the right connections
link |
that destroy the completely connected soups that you might have in mind. So a hierarchy is
link |
in and of itself defined by a sparse and particular connectivity structure.
link |
And I'm not committing to any particular form of hierarchy.
link |
But your sense says there is some. Oh, absolutely. Yeah. In virtue of the fact that there is a
link |
sparsity of connectivity, not necessarily of a qualitative sort, but certainly of a quantitative
link |
sort. So it is demonstrably so that the further apart two parts of the brain are, the less likely
link |
that they are to be wired to possess axonal processes, neuronal processes that directly
link |
communicate one message or messages from one part of that brain to the other part of the brain.
link |
So we know there's a sparse connectivity. And furthermore, on the basis of anatomical
link |
connectivity and tracer studies, we know that that has that sparsity underwrites a hierarchical
link |
and very structured sort of connectivity that might be best understood like a little bit like
link |
an onion. There is a concentric, sometimes referred to as centripetal by people like
link |
Marcel Mesulam, hierarchical organisation to the brain. So you can think of the brain as in a rough
link |
sense like an onion and all the sensory information and all the afferent outgoing messages that supply
link |
commands to your muscles or to your secretory organs come from the surface. So there's a massive
link |
exchange interface with the world out there on the surface. And then underneath, there's a little
link |
layer that sits and looks at the exchange on the surface. And then underneath that, there's a layer
link |
right the way down to the very centre through the deepest part of the onion. That's what I mean by
link |
a hierarchical organisation. There's a discernible structure defined by the sparsity of connections
link |
that lends the architecture, a hierarchical structure that tells one a lot about the kinds
link |
of representations and messages. So going back to your earlier question, is this about the
link |
representational capacity? Or is it about the anatomy? Well, one underwrites the other.
link |
If one simply thinks of the brain as a message passing machine, a process that is in the service
link |
of doing something, then the circuitry and the connectivity that shape that message passing
link |
also dictate its function. So you've done a lot of amazing work in a lot of directions.
link |
So let's look at one aspect of that, of looking into the brain and trying to study this onion
link |
structure. What can we learn about the brain by imaging it? Which is one way to sort of look
link |
at the anatomy of it, broadly speaking. What are the methods of imaging, but even bigger? What can
link |
we learn about it? Right. So well, most human neuroimaging that you might see in science journals
link |
that speaks to the way the brain works, measures brain activity over time. So that's the first
link |
thing to say that we're effectively looking at fluctuations in neuronal responses,
link |
usually in response to some sensory input or some instruction, some task. Not necessarily,
link |
there's a lot of interest in just looking at the brain in terms of resting state and dodgeness or
link |
intrinsic activity. But crucially, at every point, looking at these fluctuations either induced or
link |
intrinsic in the neural activity and understanding them at two levels. So normally, people would
link |
recourse to two principles of brain organization that are complementary. One, functional specialization
link |
or segregation. So what does that mean? It simply means that there are certain parts of the brain
link |
that may be specialized for certain kinds of processing, for example, visual motion. Our
link |
ability to recognize or to perceive movement in the visual world. And furthermore, that
link |
specialized processing may be spatially or anatomically segregated, leading to functional
link |
segregation, which means that if I were to compare your brain activity during a period of
link |
viewing a static image, and then compare that to the responses of fluctuations in the brain
link |
when you were exposed to a moving image, so a flying bird, we would expect to see restricted,
link |
segregated differences in activity. And those are basically the hotspots that you see in the
link |
in statistical parametric maps that test for the significance of the responses that are circumscribed.
link |
So now, basically, we're talking about some people of perhaps unkindly called a neo cartography.
link |
This is a phrenology augmented by modern day neuroimaging, basically finding
link |
blobs or bumps on the brain that do this or do that. I'm trying to understand the cartography
link |
of that functional specialization. So how much is there such a beautiful sort of ideal to strive
link |
for? We humans scientists would like this to hope that there's a beautiful structure to this,
link |
whereas, like you said, there are segregated regions that are responsible for the different
link |
function. How much hope is there to find such regions in terms of looking at the progress
link |
of studying the brain? Oh, I think enormous progress has been made in the past 20 or 30 years.
link |
So this is beyond incremental. At the advent of brain imaging, the very notion of functional
link |
segregation was just a hypothesis based upon a century, if not more, of careful neuropsychology,
link |
looking at people who had lost via insult or traumatic brain injury, particular parts of
link |
the brain, and then saying, well, they can't do this or they can't do that. For example,
link |
losing the visual cortex and not being able to see or usually losing particular parts of the
link |
visual cortex or regions known as V5 or the middle temporal region,
link |
MT, and noticing that they selectively could not see moving things. And so that created the
link |
hypothesis that perhaps movement processing, visual movement processing, was located in
link |
this functionally segregated area. And you could then go and put invasive electrodes in animal
link |
models and say, yes, indeed, we can excite activity here, we can form receptive fields that
link |
are sensitive to or defined in terms of visual motion. But at no point could you exclude the
link |
possibility that everywhere else in the brain was also very interested in visual motion.
link |
By the way, I apologize to Interoppa's a tiny little tangent. He said animal models
link |
are just out of curiosity from your perspective. How different is the human brain versus the
link |
other animals in terms of our ability to study the brain?
link |
Well, clearly, the further away you go from a human brain, the greater the differences,
link |
but not as remarkable as you might think. So people will choose their level of
link |
approximation to the human brain, depending upon the kinds of questions that they want to answer.
link |
So if you're talking about canonical principles of microcircuitry, it might be perfectly okay
link |
to look at a mouse, indeed, you could even look at flies, worms. If on the other hand,
link |
you wanted to look at the finer details of organization of visual cortex and V1, V2,
link |
these are designated patches of cortex that may do different things, indeed do.
link |
You probably want to use a primate that looked a little bit more like a human,
link |
because there are lots of ethical issues in terms of the use of nonhuman primates to
link |
transfer questions about human anatomy. But I think most people assume that most of the
link |
important principles are conserved in a continuous way, right from worms right through to you and me.
link |
So now returning to this, so that was the early, the ideas of studying the functional
link |
regions of the brain by, if there's some damage to it to try to infer that there's
link |
that part of the brain might be somewhat responsible for this type of function.
link |
So what, where does that lead us? What are the next steps beyond that?
link |
Right, well, this actually is a reverse a bit, come back to your sort of notion that the brain
link |
is a magic zoo, but that was actually a very prominent idea at one point to notions such as
link |
Lashley's law of mass action inherited from the observation that for certain animals,
link |
if you just took out spoonfuls of the brain, it didn't matter where you took these spoonfuls
link |
out, they always show the same kinds of deficits. So, you know, it was very difficult to infer
link |
functional specialization pure on the basis of lesion deficit studies. But once we had the
link |
opportunity to look at the brain lighting up in its, it's literally it's sort of excitement,
link |
neuronal excitement. When looking at this versus that, one was able to say, yes, indeed, these
link |
functionally specialized responses are very restricted and they're here or they're over there.
link |
If I do this, then this part of the brain lights up. And that became doable in the early 90s.
link |
In fact, you know, shortly before with the advent of positron emission tomography,
link |
and then functional magnetic resonance imaging came along in the early 90s. And since that time,
link |
there has been an explosion of discovery, refinement, confirmation. You know, there are
link |
people who believe that it's all in the anatomy. If you understand the anatomy, then you understand
link |
the function at some level. And many, many hypotheses were predicated on a deep understanding
link |
of the anatomy and the connectivity. But they were all confirmed and taken much further
link |
with neuroimaging. So that's what I meant by we've made an enormous amount of progress
link |
in this century, indeed, and in relation to the previous century by looking at these
link |
functionally selective responses. But that wasn't the whole story. So there's this sort of near
link |
phrenology, but finding bumps and hops spots in the brain that did this or that. The bigger
link |
question was, of course, the functional integration, how all of these regionally specific responses
link |
were orchestrated, how they were distributed, how did they relate to distributed processing and,
link |
indeed, representations in the brain. So then you turn to the more challenging issue of the
link |
integration of the connectivity. And then we come back to this beautiful sparse, recurrent,
link |
hierarchical connectivity that seems characteristic of the brain and probably not many other organs.
link |
And, but nevertheless, we come back to this challenge of trying to figure out how everything
link |
is integrated. But what's your feeling? What's the general consensus? Have we moved away from
link |
the magic soup view of the brain? So there is a deep structure to it. And then maybe
link |
a further question, you said some people believe that the structure is most of it,
link |
that you could really get at the core of the function by just deeply understanding the structure.
link |
Where do you sit on that? Do you? I think it's got some mileage to it. Yes. Yeah.
link |
So it's a worthy pursuit of going, of studying through imaging and all the different methods
link |
to actually study the structure. No, absolutely. Yeah. Yeah. Sorry, I'm just noting you were accusing
link |
me of using lots of long words. And then you introduced one there, which is deep, which is
link |
interesting. Because deep is the sort of millennial equivalent of hierarchical. So if you put deep
link |
in front of anything, not only are you very millennial and very trending, but you're also
link |
implying a hierarchical architecture. So it is a depth, which is for me the beautiful thing.
link |
That's right. The word deep kind of, yeah, exactly. It implies hierarchy. I didn't even think about
link |
that. That indeed the implicit meaning of the word deep is a hierarchy. Yeah. Yeah. So deep
link |
inside the onion is the center of your soul. Let's see. Beautifully put. Maybe briefly, if you
link |
can paint a picture of the kind of methods of neuroimaging, maybe the history, which you were
link |
a part of from statistical parametric mapping. I mean, just what's out there that's interesting
link |
for people maybe outside the field to understand of what are the actual methodologies of looking
link |
inside the human brain? Right. Well, you can answer that question from two perspectives. Basically,
link |
it's the modality. What kind of signal are you measuring? And they can range from,
link |
let's limit ourselves to sort of imaging based noninvasive techniques. So you've essentially
link |
got brain scanners. And brain scanners can either measure the structural attributes,
link |
the amount of water, the amount of fat or the amount of iron in different parts of the brain.
link |
And you can make lots of inferences about the structure of the organ of the sort that you
link |
might abduce from an x ray, but a very nuanced x ray that is looking at this kind of property
link |
and that kind of property. So looking at the anatomy noninvasively is would be the first
link |
sort of neuroimaging that people might want to employ. Then you move on to the kinds of measurements
link |
that reflect dynamic function. The most prevalent of those fall into two camps, you've got these
link |
metabolic sometimes hemodynamic blood related signals. So these metabolic and or hemodynamic
link |
signals are basically proxies for elevated activity and message passing and neuronal dynamics,
link |
in particular parts of the brain. Characteristically though, the time constants of these
link |
hemodynamic or metabolic responses to neural activity are much longer than the neural activity
link |
itself. And this is referring, forgive me for the dumb questions, but this would be referring to
link |
blood like the flow of blood. Absolutely. So there's a ton of, it seems like there's a ton
link |
of blood vessels in the brain. So what's the interaction between the flow of blood and the
link |
function of the neurons? Is there an interplay there? And that interplay accounts for several
link |
careers of world renowned scientists. Yes, absolutely. So this is known as neurovascular
link |
coupling is exactly what you said. It's how does the neural activity, the neural infrastructure,
link |
the actual message passing that we think underlies our capacity to perceive and act?
link |
How is that coupled to the vascular responses that supply the energy for that neural processing?
link |
So there's a delicate web of large vessels, arteries and veins, that gets progressively
link |
finer and finer in detail until it perfuses at a microscopic level, the machinery where
link |
little neurons lie. So coming back to this sort of onion perspective,
link |
we were talking before using the onion as a metaphor for a deep hierarchical structure,
link |
but also I think it's just anatomical, anatomically quite a useful metaphor. All the action,
link |
all the heavy lifting in terms of neural computation is done on the surface of the brain.
link |
And then the interior of the brain is constituted by fatty wires, essentially external processes
link |
that are enshrouded by myelin sheaths. And these give the, when you dissect them, they look fatty
link |
and white. And so it's called white matter, as opposed to the actual neuro pill, which does the
link |
computation constituted largely by neurons. And that's known as gray matter. So the gray matter
link |
is a surface or a skin that sits on top of this big ball, now we are talking magic soup, but it's
link |
big ball of connections like spaghetti, very carefully structured with sparse connectivity
link |
that preserves this deep hierarchical structure. But all the action takes place on the surface,
link |
on the cortex of the onion. And that means that you have to supply the right amount of blood flow,
link |
the right amount of nutrient, which is rapidly absorbed and used by neural cells that don't
link |
have the same capacity that your leg muscles would have to basically spend their energy budget and
link |
then claim it back later. So one peculiar thing about cerebral metabolism, brain metabolism,
link |
is it really needs to be driven in the moment, which means you basically have to turn on the taps.
link |
So if there's lots of neural activity in one part of the brain, a little patch of a few millimeters,
link |
even less possibly, you really do have to water that piece of the garden now and quickly. And
link |
that by quickly, I mean within a couple of seconds. So that contains a lot of that.
link |
Hence the imaging could tell you a story of what's happening. Absolutely. But it is slightly
link |
compromised in terms of the resolution. So the deployment of these little micro vessels that
link |
water the garden to enable the activity to the neural activity to play out. The spatial resolution
link |
is in order of a few millimeters. And crucially, the temporal resolution is the order of a few
link |
seconds. So you can't get right down and dirty into the actual spatial and temporal scale of
link |
neural activity in and of itself. To do that, you'd have to turn to the other big imaging
link |
modality, which is the recording of electromagnetic signals as they're generated in real time.
link |
So here the temporal bandwidth, if you like, or the low limit on the temporal resolution
link |
is incredibly small. You're talking about, you know, Nala seconds, milliseconds. And then you
link |
can get into the phasic fast responses that is in of itself the neural activity and start to
link |
see the succession or cascade of hierarchical recurrent message passing evoked by a particular
link |
stimulus. But the problem is you're looking at electromagnetic signals that have passed through
link |
an enormous amount of magic soup or spaghetti of collectivity and through the scalp and the skull.
link |
And it's become spatially very diffuse. It's very difficult to know where you are.
link |
So you've got this sort of catch 22. You can either use an imaging modality. It tells you
link |
within millimeters, which part of the brain is activated, we don't know when. Or you've got these
link |
electromagnetic EEG MEG setups that tell you to within a few milliseconds when something has
link |
responded, be aware. So you've got these two complementary measures, either indirect via
link |
the blood flow or direct via the electromagnetic signals caused by neural activity. These are
link |
the two big imaging devices. And then the second level of responding to your question, what are
link |
the, you know, from the outside, what are the big ways of using this technology? So once you've
link |
chosen the kind of neuroimaging they want to use to answer your set questions, and sometimes it would
link |
have to be both. Then you've got a whole raft of analyses, time series analyses, usually that you
link |
can bring to bear in order to answer your questions or address your hypotheses about those data.
link |
And interestingly, they've both fallen to the same two camps we're talking about before,
link |
you know, this dialectic between specialization and integration, differentiation and integration.
link |
So it's the cartography, the blobology analyses.
link |
I apologize. I probably shouldn't interrupt so much, but just heard a fun word.
link |
The blobology. It's a neologism, which means the study of blobs.
link |
Are you being woody and humorous? Or is there an actual, does the word blobology ever appear
link |
in a textbook somewhere? It would appear in a popular book. It would not appear in a worthy
link |
specialist journal. But it's the fond word for the study of literally little blobs on
link |
brain maps showing activations. So the kind of thing that you'd see in the newspapers on
link |
ABC or BBC reporting the latest finding from brain imaging. Interestingly, though, the maths
link |
involved in that stream of analysis does actually call upon the mathematics of blobs.
link |
So seriously, they're actually called Euler characteristics and they have a lot of fancy
link |
names in mathematics.
link |
We'll talk about it, but your ideas in free energy principle, I mean,
link |
there's a echoes of blobs there when you consider sort of entities, so mathematically speaking.
link |
Yes, absolutely. Well, circumscribed, well defined. You entities of,
link |
well, from the free energy point of view, entities of anything, but from the point of view of
link |
the analysis, the cartography of the brain. These are the entities that constitute the
link |
evidence for this functional segregation. You have segregated this function in this blob,
link |
and it is not outside of the blob. And that's basically the, if you were a map maker of America
link |
and you did not know its structure, the first thing you're doing constituting or creating a map
link |
would be to identify the cities, for example, or the mountains or the rivers. All of these
link |
uniquely spatially localizable features, possibly topological features, have to be placed somewhere
link |
because that requires a mathematics of identifying what does a city look like on the satellite image
link |
or what does a river look like or what does a mountain look like. What would it, what data
link |
features would it, would evidence that particular top, that particular thing that you wanted to
link |
put on the map. And they normally are characterized in terms of literally these blobs or these sort of,
link |
now the way of looking at this is that a certain statistical measure of the degree of activation
link |
crosses a threshold and in crossing that threshold in the spatially restricted part of the brain,
link |
it creates a blob. And that's basically what statistical parametric mapping does. It's basically
link |
mathematically finessed blobology. Okay, so those, you kind of describe these two methodologies for
link |
one is temporally noisy, one is spatially noisy, and you kind of have to play and figure out what
link |
can be useful. It'd be great if you can sort of comment, I got a chance recently to spend a day
link |
at a company called Neuralink that uses brain computer interfaces. And their dream is to,
link |
well, there's a bunch of sort of dreams, but one of them is to understand the brain by sort of,
link |
you know, getting in there past the so called sort of factory while getting in there and be
link |
able to listen, communicate both directions. What are your thoughts about this, the future of this
link |
kind of technology of brain computer interfaces to be able to now have a, have a window or direct
link |
contact within the brain to be able to measure some of the signals to be able to send signals to
link |
understand some of the functionality of the brain. I'm bivalent. My sense is ambivalent. So it's a
link |
mixture of good and bad. And I acknowledge that freely. So the good bits, if you just look at
link |
the legacy of that kind of reciprocal but invasive brain stimulation, I didn't paint a complete picture
link |
when I was talking about some of the ways we understand the brain prior to neuroimaging.
link |
It wasn't just lesion deficit studies. Some of the early work, in fact, literally 100 yards from
link |
where we're sitting at the Institute of Neurology was done by stimulating the brain of, say, dogs
link |
and looking at how they responded with their muscles or with their salivation. And imputing
link |
what that part of the brain must be doing that if I stimulate it and I vote this kind of response,
link |
then that tells me quite a lot about the functional specialization. So there's a long
link |
history of brain stimulation, which continues to enjoy a lot of attention nowadays.
link |
Oh, yes, absolutely. Deep brain stimulation for Parkinson's disease is now a standard treatment
link |
and also a wonderful vehicle to try and understand the neuronal dynamics underlie movement disorders
link |
like Parkinson's disease. Even interest in magnetic stimulation, stimulating the magnetic
link |
fields and will it work in people who are depressed, for example? Quite a crude level
link |
of understanding what you're doing. But there is historical evidence that these kinds of
link |
brute thought interventions do change things. A little bit like banging the TV when the valves
link |
aren't working properly, but it still works. So there is a long history. Brain computer
link |
interfacing or BCI, I think is a beautiful example of that. It's sort of carved out its own
link |
niche and its own aspirations and there have been enormous advances within limits.
link |
Advances in terms of our ability to understand how the brain, the embodied brain,
link |
engages with the world. I'm thinking here of sensory substitution, the augmenting our sensory
link |
capacities by giving ourselves extra ways of sensing and sampling the world, ranging from
link |
sort of trying to replace lost visual signals through to giving people completely new signals.
link |
One of the, I think, most engaging examples of this is equipping people with a sense of
link |
magnetic fields. So you can actually give them magnetic sensors that enable them to feel,
link |
should we say, tactile pressure around their tummy, where they are in relation to the magnetic
link |
field of the earth. And after a few weeks, they take it for granted, they integrate it,
link |
they embody this, simulate this new sensory information into the way that they literally
link |
feel their world, but now equipped with this sense of magnetic direction. So that tells you
link |
something about the brain's plastic potential to remodel and its plastic capacity to suddenly
link |
try to explain the sensory data at hand by augmenting the sensory sphere and the kinds of
link |
things that you can measure. Clearly, that's purely for entertainment and understanding the
link |
other nature and the power of our brains. I would imagine that most BCI is pitched at solving
link |
clinical and human problems such as locked in syndrome, such as paraplegia, or replacing lost
link |
sensory capacities like blindness and deafness. So then we come to the more negative part of my
link |
ambivalence, the other side of it. So I don't want to be deflationary because
link |
much of my deflationary comments are probably large out of ignorance than anything else. But
link |
generally speaking, the bandwidth and the bit rates that you get from
link |
brain computer interfaces as we currently know them, we're talking about bits per second.
link |
So that would be like me only being able to communicate with any world or with you using very,
link |
very, very slow Morse code. And it is not even within an order of magnitude near what we actually
link |
need for an inactive realization of what people aspire to when they think about sort of curing
link |
people with paraplegia or replacing sight, despite heroic efforts. So one has to ask,
link |
is there a lower bound on the kinds of recurrent information exchange between a brain and some
link |
augmented or artificial interface? And then we come back to, interestingly, what I was talking
link |
about before, which is if you're talking about function in terms of inference, and I presume
link |
we'll get to that later on in terms of the free energy principle, then the moment there may be
link |
fundamental reasons to assume that is the case, we're talking about ensemble activity, we're
link |
talking about basically, for example, let's paint the challenge facing brain computer
link |
interfacing in terms of controlling another system that is highly and deeply structured,
link |
very relevant to our lives, very nonlinear, that rests upon the kind of non equilibrium
link |
steady states and dynamics that the brain does, the weather. So good example, yeah.
link |
Imagine you had some very aggressive satellites that could produce signals that could perturb
link |
some little parts of the of the weather system. And then what you're asking now is,
link |
can I meaningfully get into the weather and change it meaningfully and make the weather
link |
respond in a way that I want it to, you're talking about chaos control on a scale, which is almost
link |
unimaginable. So there may be fundamental reasons why BCI, as you might read about it in a science
link |
fiction novel, aspirational BCI may never actually work in the sense that to really be integrated
link |
and be part of the system is a requirement that requires you to have evolved with that system.
link |
You have to be part of a very delicately structured, deeply structured dynamic ensemble
link |
activity that is not like rewiring a broken computer or plugging in a peripheral interface
link |
adapter. It is much more like getting into the weather patterns or a come back to your magic
link |
soup is getting into the active matter and meaningfully relate that to the outside world.
link |
So I think there are enormous challenges there. So I think the example of the weather is a brilliant
link |
one. And I think you paint a really interesting picture. And it wasn't as negative as I thought.
link |
It's essentially saying that it's it might be incredibly challenging, including the low bound
link |
of the bandwidth and so on. So just to full disclosure, I come from the machine learning world.
link |
So my natural thought is the hardest part is the engineering challenge of controlling the
link |
weather, of getting those satellites up and running and so on. And once they are, then the rest
link |
is fundamentally the same approaches that allow you to win in the game of go will allow you to
link |
potentially play in this soup in this chaos. So I have I have a hope that sort of machine
link |
learning methods will will help us play in this soup. But perhaps you're right that it is a biology
link |
in the brain is just an incredible, incredible system that may be almost impossible to get in.
link |
But for me, what seems impossible is is the incredible mess of blood vessels that you also
link |
described without, you know, we also value the brain. You can't make any mistakes. You can't
link |
damage things. So to me, that engineering challenge seems nearly impossible. One of the things I was
link |
really impressed by at Neuralink is just just talking to brilliant neurosurgeons and the roboticists
link |
that made me realize that even though it seems impossible, if anyone can do it, it's some of
link |
these world class engineers that are trying to take it on. So so I think the conclusion of our
link |
discussion here is of this part is basically that the problem is really hard, but hopefully not
link |
impossible. Absolutely. So if it's okay, let's start with the basics. So you've also formulated a
link |
fascinating principle, the free energy principle. Can we maybe start at the basics and what is
link |
the free energy principle? Well, in fact, the free energy principle
link |
inherits a lot from the building of these data analytic approaches to these very high dimensional
link |
time series you get get from the brain. So I think it's interesting to acknowledge that. And in
link |
particular, the analysis tools that try to address the other side, which is the functional
link |
integrations and the connectivity analysis. On the one hand, but I should also acknowledge it
link |
inherits an awful lot from machine learning as well. So the free energy principle is just a
link |
formal statement that the existential imperatives are any system that manages to survive in a
link |
changing world is can be cast as an inference problem, in the sense that you can interpret
link |
the probability of existing as the evidence that you exist. And if you can write down that problem
link |
of existence as a statistical problem, then you can use all the maths that has been developed
link |
for inference to understand and characterize the ensemble dynamics that must be in play in the
link |
service of that inference. So technically, what that means is you can always interpret anything
link |
that exists in virtue of being separate from the environment in which it exists as trying to
link |
minimize variational free energy. And if you're from the machine learning community, you will
link |
know that as a negative evidence lower bound or a negative elbow, which is the same as saying
link |
you're trying to maximize or it will look as if all your dynamics are trying to maximize the
link |
complement of that which is the marginal likelihood or the evidence for your own existence. So that's
link |
basically the free energy principle. But to even take a sort of a small step backwards,
link |
you said the existential imperative. There's a lot of beautiful poetic words here, but to put
link |
it crudely, it's a fascinating idea of basically just of trying to describe if you're looking at a
link |
blob, how do you know this thing is alive? What does it mean to be alive? What does it mean to
link |
be to exist? And so you can look at the brain, you can look at parts of the brain or you this is
link |
just a general principle that applies to almost any system. That's just a fascinating sort of
link |
philosophically at every level question and a methodology to try to answer that question.
link |
What does it mean to be alive? Yes. So that's a huge endeavor and it's nice that there's at least
link |
some from some perspective a clean answer. So maybe can you talk about that optimization
link |
view of it? So what's trying to be minimized to maximize a system that's alive? What is it trying
link |
to minimize? Right, you've made a big move there. First of all, it's good to make big moves.
link |
But you've assumed that the things, the thing exists in a state that could be living or non
link |
living. So I may ask you, what licenses you to say that something exists? That's why I use the
link |
word existential. It's beyond living, it's just existence. So if you drill down onto the definition
link |
of things that exist, then they have certain properties. If you borrow the maths from non
link |
equilibrium steady state physics that enable you to interpret their existence in terms of this
link |
optimization procedure. So it's good you introduce the word optimization. So what the free energy
link |
principle in its sort of most ambitious but also most deflationary and simplest says is that if
link |
something exists, then it must buy the mathematics of non equilibrium steady state exhibit properties
link |
that make it look as if it is optimizing a particular quantity. And it turns out that particular
link |
quantity happens to be exactly the same as the evidence lower bound in machine learning,
link |
or Bayesian model evidence in Bayesian statistics, or I can list a whole other list of ways of
link |
understanding this key quantity, which is abound on surprises, self information, if you
link |
know information theory, there are a number of different perspectives on this quantity.
link |
It's just basically the log probability of being in a particular state. I'm telling this story
link |
as an honest attempt to answer your question. And I'm answering it as if I was pretending
link |
to be a physicist who was trying to understand the fundamentals of non equilibrium steady state.
link |
And I shouldn't really be doing that because the last time I was taught physics, I was in my 20s.
link |
What kind of systems when you think about the free energy principle, what kind of systems
link |
are you imagining as a sort of more specific kind of case study?
link |
Yeah, I'm imagining a range of systems, but at its simplest, a single celled organism
link |
that can be identified from its eco niche or its environment. So at its simplest,
link |
that's basically what I always imagined in my head. And you may ask, well, how on earth
link |
can you even elaborate questions about the existence of a single drop of oil, for example?
link |
But there are deep questions there. Why doesn't the thing, the interface between the drop of oil
link |
that contains an interior and the thing that is not the drop of oil, which is the solvent in
link |
which it is immersed? How does that interface persist over time? Why doesn't the oil just dissolve
link |
into solvent? So what's special properties of the exchange between the surface of the oil drop
link |
and the external states in which it's immersed? If you're a physicist, say it would be the heat
link |
path. You've got a physical system, an ensemble again, we're talking about density dynamics,
link |
ensemble dynamics, an ensemble of atoms or molecules immersed in the heat path. But the
link |
question is, how did the heat path get there? And why is it not dissolved?
link |
How's it maintaining itself? Exactly. What actions is it? I mean, it's such a fascinating idea of a
link |
drop of oil and I guess it would dissolve in water. It wouldn't dissolve in water. So what?
link |
Precisely. So why not? So why not? Why not? And how do you mathematically describe it? I mean,
link |
it's such a beautiful idea and also the idea of where does the drop of oil end and where does it
link |
begin? Right. So I mean, you're asking deep questions, deep in a non millennial sense here.
link |
But what you can do, so this is a deflationary part of it. Can I just qualify my answer by
link |
saying that normally when I'm asked this question, I answer from the point of view of a psychologist
link |
when we talk about predictive processing and predictive coding and the brain as an inference
link |
machine. But you have asked me from that perspective, I'm answering from the point of view of a
link |
physicist. So the question is not so much why, but if it exists, what properties must it display?
link |
So that's the deflationary part of the free energy principle. The free energy principle
link |
does not supply an answer as to why. It's saying, if something exists, then it must display these
link |
properties. That's the thing that's on offer. And it so happens that these properties, it must
link |
display are actually intriguing and have this inferential gloss, this sort of self evidencing
link |
gloss that inherits on the fact that the very preservation of the boundary between the oil
link |
drop and the not oil drop requires an optimization of a particular function or a functional
link |
that defines the presence of the existence of this oil drop, which is why I started with
link |
existential imperatives. It is a necessary condition for existence that this must occur,
link |
because the boundary basically defines the thing that's existing. So it is that self assembly
link |
aspect that you were hinting at in biology, sometimes known as auto poesis in computational
link |
chemistry with self assembly. What does it look like? Sorry, how would you describe things that
link |
configure themselves out of nothing? The way they clearly demarcate themselves from the states or the
link |
soup in which they are immersed. So from the point of view of computational chemistry, for example,
link |
you would just understand that as a configuration of a macromolecule to minimize its free energy,
link |
its thermodynamic free energy. It's exactly the same principle that we've been talking about,
link |
that thermodynamic free energy is just the negative elbow. It's the same mathematical construct.
link |
So the very emergence of existence, of structure, of form that can be distinguished from the environment
link |
or the thing that is not the thing necessitates the existence of an objective function that it
link |
looks as if it is minimizing. It's finding a free energy minima. And so just to clarify,
link |
I'm trying to wrap my head around, so the free energy principle says that if something exists,
link |
these are the properties it should display. So what that means is we can't just go into a soup
link |
and there's no mechanism. A free energy principle doesn't give us a mechanism to find the things
link |
that exist. Is that what's implying is being applied that you can use it to reason, to think about
link |
study a particular system and say, does this exhibit these qualities?
link |
That's an excellent question. But to answer that, I'd have to return to your previous question
link |
about what's the difference between living and nonliving things.
link |
Yes, actually, sorry. So yeah, maybe we can go there. You drew a line, and forgive me for
link |
the stupid questions, but you drew a line between living and existing. Is there an
link |
interesting distinction? I think there is. So things do exist, grains of sand, rocks on the
link |
moon, trees, you. So all of these things can be separated from the environment in which
link |
they are immersed. And therefore, they must at some level be optimizing their free energy,
link |
taking this sort of model evidence interpretation of this quantity that basically means they're
link |
self evidencing. Another nice little twist of phrase here is that you are your own existence
link |
proof. And statistically speaking, which I don't think I said that, somebody did, but I love that
link |
phrase. You are your own existence proof. Yeah. So it's so existential, isn't it?
link |
I'm going to have to think about that for a few days. That's a beautiful line.
link |
So the step through to answer your question about what's it good for,
link |
well, we go along the following lines. First of all, you have to define what it means to exist,
link |
which down as you've rightly pointed out, you have to define what probabilistic properties
link |
must the states of something possess so that it has so it knows where it finishes. And then you
link |
write down that down in terms of statistical independence is again sparsity. Again, it's not
link |
what's connected or what's correlated or what depends upon what it's what's not
link |
correlated and what doesn't depend upon something. Again, it comes down to the deep structures,
link |
not in this instance hierarchical, but the certainly the structures that emerge from removing
link |
connectivity and dependency. And in this instance, basically being able to identify the surface of
link |
the oil drop from the water in which it is immersed. And when you do that, you start to realise, well,
link |
there are actually four kinds of states in any given universe that contains anything,
link |
the things that are internal to the surface, the things that are external to the surface and
link |
the surface in and of itself, which is why I use a metaphor, a little single celled organism that
link |
has an interior and exterior and then the surface of the cell. And that's mathematically a Markov
link |
blanket. Just to pause, I'm in awe of this concept that there's the stuff outside the surface stuff
link |
inside the surface and the surface itself, the Markov blanket. It's just the most beautiful
link |
kind of notion about trying to explore what it means to exist, mathematically. I apologize,
link |
this is the beautiful idea. I came out of California. I changed my mind. I take it all back. So
link |
you were just talking about the surface, about the Markov blanket.
link |
So this surface or this blanket, these blanket states that are the
link |
because they are now defined in relation to these dependencies and what different states internal
link |
or blanket or external states can, which ones can influence each other and which cannot influence
link |
each other, you can now apply standard results that you would find in non equilibrium physics
link |
or steady state or thermodynamics or hydrodynamics, usually out of equilibrium solutions and apply
link |
them to this partition. And what it looks like is if all the normal, normal gradient flows that
link |
you would associate with any non equilibrium system, apply in such a way that two part of the
link |
Markov blanket and the internal states seem to be hill climbing or doing a gradient descent on
link |
the same quantity. And that means that you can now describe the very existence of this oil drop.
link |
You can write down the existence of this oil drop in terms of flows, dynamics, equations of motion
link |
where the blanket states or part of them, we call them active states and the internal states
link |
now seem to be and must be trying to look as if they're minimizing the same function,
link |
which is a lot of probability of occupying these states. The interesting thing is that
link |
what would they be called if you were trying to describe these things? So
link |
what we're talking about are internal states, external states and blanket states. Now let's
link |
carve the blanket states into two sensory states and active states. Operationally, it has to be
link |
the case that in order for this carving up into different sets of states to exist, the active
link |
states, the Markov blanket cannot be influenced by the external states. And we already know
link |
that the internal states can't be influenced by the external states because the blanket separates
link |
them. So what does that mean? Well, it means the active states, the internal states are now
link |
jointly not influenced by external states. They only have autonomous dynamics. So now you've got
link |
a picture of an oil drop that has autonomy. It has autonomous states. It has autonomous states
link |
in the sense that there must be some parts of the surface of the oil drop that are not influenced
link |
by the external states and all the interior. And together, those two states endow even a
link |
little oil drop with autonomous states that look as if they are optimizing their variational free
link |
energy or their negative elbow, their model evidence. And that would be an interesting
link |
intellectual exercise. And you could say you could even go into the realms of panpsychism,
link |
that everything that exists is implicitly making inferences on self evidencing.
link |
Now, we make the next move. But what about living things? I mean, so let me ask you,
link |
what's the difference between an oil drop and a little tadpole or a little lava or plankton?
link |
The picture was just painted of an oil drop. Just immediately, in a matter of minutes,
link |
took me into the world of panpsychism, where you just convinced me, made me feel like an
link |
oil drop is a living, certainly an autonomous system, but almost a living system. So it has
link |
a sensor capabilities and acting capabilities and it maintains something. So what is the
link |
difference between that and something that we traditionally think of as a living system?
link |
That it could die? Or it can't, I mean, yeah, mortality? I'm not exactly sure. I'm not sure
link |
what the right answer there is, because it can move, like movement seems like an essential
link |
element to being able to act in the environment, but the oil drop is doing that. So I don't know.
link |
Is it? The oil drop will be moved, but does it in and of itself move autonomously?
link |
Well, the surface is performing actions that maintain its structure.
link |
Yeah, you're being too clever. I had in mind a passive little oil drop that's sitting there.
link |
Yeah. At the bottom of the top of a glass of water.
link |
What I'm trying to say is you're absolutely right. You've nailed it. It's movement.
link |
So where does that movement come from? If it comes from the inside,
link |
then then you've got, I think, something that's living.
link |
What do you mean from the inside?
link |
What I mean is that the internal states that can influence the active states,
link |
where the active states can influence, but they're not influenced by the external states,
link |
can cause movement. So there are two types of oil drops, if you like.
link |
There are oil drops where the internal states are so random that they average themselves away,
link |
and the thing cannot balance on average when you do the averaging move.
link |
So a nice example of that would be the sun. The sun certainly has internal states.
link |
There's lots of intrinsic autonomous activity going on, but because it's not coordinated,
link |
because it doesn't have the deep in the millennial sense, a hierarchical structure that the brain
link |
does, there is no overall mode or pattern or organization that expresses itself on the surface
link |
that allows it to actually swim. It can certainly have a very active surface, but on mass at the
link |
scale of the actual surface of the sun, the average position of that surface cannot in itself move,
link |
because the internal dynamics are more like a hot gas. They are literally like a hot gas,
link |
whereas your internal dynamics are much more structured and deeply structured,
link |
and now you can express on your mark of, on your active states with your muscles and
link |
your secretory organs, your autonomic nervous system, and its effectors.
link |
You can actually move, and that's all you can do. And that's something which, you know,
link |
if you haven't thought of it like this before, I think it's nice to just realize there is no other
link |
way that you can change the universe other than simply moving. Whether that moving is
link |
articulating my, with my voice box, or walking around, or squeezing juices out of my secretory
link |
organs, there's only one way you can change the universe. It's moving. And the fact that you do
link |
so non randomly makes you alive. Yeah. So it's that non randomness. And that would be manifested,
link |
we realize in terms of essentially swimming, essentially moving, changing one shape, a morphogenesis
link |
that is dynamic, and possibly adaptive. So that that's what I was trying to get up between the
link |
difference from the oil drop and the little tadpole, that the tadpole is moving around.
link |
Its active states are actually changing the external states. And there's now a cycle,
link |
an action perception cycle, if you like, a recurrent dynamic that's going on that depends upon this
link |
deeply structured autonomous behavior that rests upon internal dynamics that are not only
link |
modeling, the data impressed upon their surface or the blanket states, but they are actively
link |
resampling those data by moving, they're moving towards chemical gradients and chemotaxis.
link |
So they've gone beyond just being good little models of the kind of world they live in. For
link |
example, an oil droplet could, in a panpsychic sense, be construed as a little being that has
link |
now perfectly inferred. It's a passive non living oil drop living in a bowl of water. No problem.
link |
But to now equip that oil drop with the ability to go out and test that hypothesis about different
link |
states of beings, so you can actually push its surface over there, over there, and test for
link |
chemical gradients, or then you start to move to much more lifelike form. This is all fun,
link |
theoretically interesting, but it actually is quite important in terms of reflecting what I
link |
have seen since the turn of the millennium, which is this move towards an inactive and embodied
link |
understanding of intelligence. And you say you're from machine learning. So what that means,
link |
the central importance of movement, I think has yet to really hit machine learning. It certainly
link |
has now diffused itself throughout robotics. And perhaps you could say certain problems in
link |
active vision where you actually have to move the camera to sample this and that. But machine
link |
learning of the data mining deep learning sort simply hasn't contended with this issue. What
link |
it's done instead of dealing with the movement problem and the active sampling of data, it's
link |
just said, we don't need to worry about we can see all the data because we've got big data.
link |
So we need to ignore movement. So that, for me, is an important omission in current machine
link |
learning. The current machine learning is much more like the oil drop. Yes. But an oil drop that
link |
enjoys exposure to nearly all the data it will ever need to be exposed to, as opposed to the
link |
tapels swimming out to find the right data. For example, it likes food. That's a good hypothesis.
link |
Let's test it out. Let's go and move and ingest food, for example, and see what that is that
link |
evidence that I'm the kind of thing that likes this kind of food. So the next natural question,
link |
and forgive this question, but if we think of sort of even artificial intelligence systems,
link |
which has just painted a beautiful picture of existence and life. So do you ascribe,
link |
but do you find within this framework a possibility of defining consciousness or exploring the idea
link |
of consciousness, like what self awareness and expanded to consciousness? How can we start
link |
to think about consciousness within this framework? Is it possible?
link |
Well, I think it's possible to think about it, whether you'll get anywhere. It's another question.
link |
And again, I'm not sure that I'm licensed to answer that question. I think you'd have to
link |
speak to a qualified philosopher to get a definitive answer there. But certainly,
link |
there's a lot of interest in using not just these ideas, but related ideas from information theory
link |
to try and tie down the maths and the calculus and the geometry of consciousness,
link |
either in terms of sort of a minimal consciousness, even less than a minimal
link |
selfhood. And what I'm talking about is the ability effectively to plan, to have agency.
link |
So you could argue that a virus does have a form of agency in virtue of the way that it selectively
link |
finds hosts and cells to live in and moves around. But you wouldn't endow it with the capacity to think
link |
about planning and moving in a purposeful way where it countens as the future. Whereas you might
link |
an ant, you might think an ant's not quite as unconscious as a virus. It certainly seems to
link |
have a purpose. It talks to its friends on route during its foraging. It has a different kind of
link |
autonomy, which is biotic, but beyond a virus.
link |
So there's something about, so there's some line that has to do with the complexity of planning
link |
that may contain an answer. I mean, it would be beautiful if we can find a line beyond which
link |
we can say a being is conscious. These are wonderful lines that we've drawn with existence,
link |
life and consciousness. It will be very nice. One little wrinkle there, and this is something
link |
I've only learned in the past few months, is the philosophical notion of vagueness.
link |
So you're saying it would be wonderful to draw a line. I had always assumed that that line at some
link |
point would be drawn until about four months ago, and the philosopher taught me about vagueness.
link |
So I don't know if you've come across this, but it's a technical concept and I think most
link |
revealingly illustrated with at what point does a pile of sand become a pile?
link |
Is it one grain, two grains, three grains or four grains? So at what point would you draw
link |
the line between being a pile of sand and a collection of grains of sand? In the same way,
link |
is it right to ask, where would I draw the line between conscious and unconscious?
link |
And it might be a vague concept. Having said that, I agree with you entirely.
link |
So systems that have the ability to plan. So just technically what that means is
link |
your inferential self evidencing by which I simply mean the dynamics, literally the
link |
thermodynamics and gradient flows that underwrite the preservation of your oil droplet like form
link |
are described as an optimization of log Bayesian model evidence, your elbow.
link |
That self evidencing must be evidence for a model of what's causing the sensory impressions on the
link |
sensory part of your surface or your Markov blanket. If that model is capable of planning,
link |
it must include a model of the future consequences of your active states or your action just planning.
link |
So we're now in the game of planning as inference. Now notice what we've made though.
link |
We've made quite a big move away from big data and machine learning because again,
link |
it's the consequences of moving. It's the consequences of selecting those data or those
link |
data or looking over there. And that tells you immediately that even to be a contender for
link |
a conscious artifact or a strong AI or general, then you've got to have movement in the game.
link |
And furthermore, you've got to have a generative model of the sort you might find in say a variation
link |
autoencoder that is thinking about the future conditioned upon different courses of action.
link |
Now that brings a number of things to the table, which which now you start to think,
link |
well, those have got all the right ingredients to talk about consciousness. I've now got to
link |
select among a number of different courses of action into the future as part of planning.
link |
I've now got free will. The act of selecting this course of action or that policy or that
link |
policy or that action suddenly makes me into an inference machine, a self evidencing artifact
link |
that now looks as if it's selecting amongst different alternative ways forward as I actively
link |
swim here or swim there or look over here, look over there. So I think you've now got to a situation
link |
if there is planning in the mix, you're now getting much closer to that line if that line
link |
wherever to exist. I don't think it gets you quite as far as self aware, though. I think
link |
and then you're you have to I think grapple with the question.
link |
How would formally write down a calculus or a maths of self awareness? I don't think it's
link |
impossible to do. But I think you would be pressure on you to actually commit to a formal
link |
definition of what you mean by self awareness. I think most people that I know would probably
link |
say that a goldfish, a pet fish was not self aware. They would probably argue about their
link |
favorite cat, but would be quite happy to say that their mum was self aware.
link |
But that might very well connect to some level of complexity with planning. It seems like
link |
self awareness is essential for complex planning.
link |
Yeah. Do you want to take that further? I think you're absolutely right.
link |
Again, the line is unclear, but it seems like integrating yourself into the world,
link |
into your planning, is essential for constructing complex plans.
link |
So mathematically describing that in the same elegant way as you have with the free energy
link |
principle might be difficult.
link |
Well, yes and no. I don't think that perhaps we should just, can we just go back? That's
link |
a very important answer you gave. And I think if I just unpacked it, you'd see the truisms
link |
that you've just exposed for us. But let me, sorry, I'm mindful that I didn't answer your
link |
question before. What's the free energy principle good for? Is it just a pretty
link |
theoretical exercise to explain nonequilibrium in steady states? Yes, it is. It does nothing
link |
more for you than that. It can be regarded, it's going to sound very arrogant, but it is of the
link |
sort of theory of natural selection or a hypothesis of natural selection.
link |
Beautiful, undeniably true, but tells you absolutely nothing about why you have legs
link |
and eyes. It tells you nothing about the actual phenotype and it wouldn't allow you to build
link |
something. So the free energy principle by itself is as vacuous as most
link |
tautological theories. And by tautological, of course, I'm talking to the theory of
link |
natural survival of the fittest. What's the fittest survival? Why are the cycles the fitter?
link |
It discorrected circles. In a sense, the free energy principle has that same deflationary
link |
tautology under the hood. It's a characteristic of things that exist. Why do they exist? Because
link |
they minimise their free energy. Why do they minimise their free energy? Because they exist.
link |
And you just keep on going round and round and round. But the practical thing which you
link |
don't get from natural selection, but you could say has now manifest in things like differential
link |
evolution or genetic algorithms and MCMC, for example, in machine learning. The practical
link |
thing you can get is if it looks as if things that exist are trying to have density dynamics
link |
and look as though they're optimising a variational free energy. And a variational free energy has
link |
to be a functional of a generative model, probabilistic description of causes and
link |
consequences, causes out there, consequences in the sensorium on the sensory parts of the
link |
Markov Planckin. Then it should, in theory, impossible to write down the generative model,
link |
work out the gradients, and then cause it to autonomously self evidence. So you should be
link |
able to write down oil droplets. You should be able to create artefacts where you have supplied
link |
the objective function that supplies the gradients that supplies the self organising dynamics to
link |
non equilibrium steady state. So there is actually a practical application, the free energy principle,
link |
when you can write down your required evidence in terms of, well, when you can write down the
link |
generative model, that is the thing that has the evidence, the probability of these sensory data
link |
or this data, given that given that model is effectively the thing that the elbow of the
link |
variational free energy bounds or approximates. That means that you can actually write down the
link |
model. And the kind of thing that you want to engineer the kind of AGI artificial general
link |
intelligence that you want to manifest probabilistically. And then you engineer a robot and a computer
link |
to perform a gradient descent on that objective function. So it does have a practical implication.
link |
Now, why am I wittering on about that? It did seem relevant to, yes. So what kinds of, so the
link |
answer to, would it be easy or would it be hard? Well, mathematically, it's easy. I've just told
link |
you all you need to do is write down your perfect artefact probabilistically in the form of a
link |
probabilistic generative model, a probability distribution over the causes and consequences
link |
of the world in which this thing is immersed. And then you just engineer a computer and a robot
link |
to form a gradient descent on that objective function. No problem. But of course, the big
link |
problem is writing down the generative model. So that's where the heavy lifting comes in.
link |
So it's the form and the structure of that generative model, which basically defines the
link |
artefact that you will create or indeed the kind of artefact that has self awareness.
link |
So that's where all the hard work comes in very much like natural selection doesn't tell you in
link |
the slightest why you have eyes. So you have to drill down on the actual phenotype, the actual
link |
generative model. So with that in mind, what did you tell me that tells me immediately the kinds
link |
of generative models I would have to write down in order to have self awareness? What you said to me
link |
was I have to have a model that is effectively fit for purpose for this kind of world in which I
link |
operate. And if I now make the observation that this kind of world is effectively largely populated
link |
by other things like me, i.e. you, then it makes enormous sense that if I can develop a hypothesis
link |
that we are similar kinds of creatures, in fact, the same kind of creature, but I am me and you
link |
are you, then it becomes again mandated to have a sense of self. So if I live in a world
link |
that is constituted by things like me, basically a social world or community,
link |
then it becomes necessary now for me to infer that it's me talking and not you talking. I wouldn't
link |
need that if it was on Mars by myself or if I was in the jungle as a feral child. If there was
link |
nothing like me around, there would be no need to have an inference that a hypothesis, oh yes,
link |
it is me that is experiencing or causing these sounds and it is not you. It's only when there's
link |
ambiguity in play induced by the fact that there are others in that world. So I think that the
link |
special thing about self aware artifacts is that they have learned to or they have acquired,
link |
or at least not equipped with, possibly by evolution, generative models that allow for the
link |
fat. There are lots of copies of things like them around and therefore they have to work out it's
link |
you and not me. That's brilliant. I've never thought of that. I never thought of that,
link |
that the purpose of the really usefulness of consciousness or self awareness in the context
link |
of planning existing in the world is so you can operate with other things like you. It doesn't
link |
have to necessarily be human, it could be other kind of similar creatures. Absolutely. Well,
link |
we view a lot of our attributes into our pets, don't we? Or we try to make our robots humanoid.
link |
And I think there's a deep reason for that, that it's just much easier to read the world
link |
if you can make the simplifying assumption that basically you're me and it's just your turn to
link |
talk. I mean, when we talk about planning, when you talk specifically about planning, the highest
link |
if like manifestation or realization of that planning is what we're doing now. I mean,
link |
the human condition doesn't get any higher than this talking about the philosophy of existence
link |
and the conversation. But in that conversation, there is a beautiful art of turn taking
link |
and mutual inference, theory of mind. I have to know when you want to listen,
link |
I have to know when you want to interrupt, I have to make sure that you're online,
link |
I have to have a model in my head of your model in your head. That's the highest,
link |
the most sophisticated form of generative model, where the generative model actually has a
link |
generative model, somebody else's generative model. And I think that and what we are doing now
link |
evinces the kinds of generative models that would support self awareness. Because without that,
link |
we'd both be talking over each other, or we'd be singing together in a choir. That's not a
link |
brilliant analogy for what I'm trying to say, but we wouldn't have this discourse.
link |
Yeah, the dance of it. Yeah, that's right. You get to have as I interrupt. I mean, that's
link |
beautifully put. I'll re listen to this conversation many times. There's so much
link |
poetry in this and mathematics. Let me ask the silliest or perhaps the biggest question
link |
as a last kind of question. We've talked about living in existence and the objective function
link |
under which these objects would operate. What do you think is the objective function of our
link |
existence? What's the meaning of life? What do you think is the for you perhaps, the purpose, the
link |
source of fulfillment, the source of meaning for your existence as one blob in the soup?
link |
I'm tempted to answer that again as a physicist. Free energy, I expect consequent upon my behavior.
link |
So technically, we can get a really interesting conversation about
link |
what that comprises in terms of searching for information, resolving uncertainty about the
link |
kind of thing that I am. But I suspect that you want a slightly more personal and fun answer.
link |
But which is can be consistent with that. And I think it's reassuringly simple and
link |
harps back to what you were taught as a child, that you have certain beliefs about the kind of
link |
creature and the kind of person you are. And all that self evidencing means, all that minimizing
link |
variational free energy in an inactive and embodied way means is fulfilling the beliefs
link |
about what kind of thing you are. And of course, we're all given those scripts, those narratives
link |
very early age, usually in the former bedtime stories or fairy stories that I'm a princess
link |
and I'm going to make a beast who's going to transform. The narratives are all around you
link |
from your parents to the friends, to the society feeds these stories. And then your objective
link |
function is to fulfill. Exactly. That narrative that has been encultured by your immediate family,
link |
but as you say also, the sort of the culture in which you grew up. And you create for yourself.
link |
I mean, again, because of this active inference, this inactive aspect of self evidencing, you know,
link |
not only am I modeling my environment, my economy, my my external states out there,
link |
but I'm actively changing them all the time. And external states are doing the same back,
link |
we're doing it together. So there's a synchrony that means that I'm creating my own culture
link |
over different timescales. So the question now is for me being very selfish, what scripts were I
link |
given? It basically was a mixture between Einstein and shark homes. So I smoke as heavily as possible,
link |
try to avoid too much interpersonal contact. Yeah, enjoy the fantasy that you're a popular
link |
scientist who's going to make a difference in a slightly quirky way. So that's what I grew up on.
link |
My father was an engineer and loved science and he loved, you know, sort of things like Sir Arthur
link |
Edwards, Space, Time and Gravitation, which was the first understandable version of general
link |
relativity. And so all the fairy stories I was told as I was growing up were all about these
link |
characters. I'm keeping the hobbit out of this because that was quite fit my narrative. There's
link |
a journey of exploration, I suppose, of sorts. So yeah, I've just grown up to be what I imagine
link |
a mild mannered Sherlock Holmes slash Albert Einstein would would do in my shoes.
link |
And you did it elegantly and beautifully. Carl was a huge honor talking today. It was fun.
link |
Thank you so much for your time. Thank you, Shane. Thank you for listening to this conversation
link |
with Carl Friston. And thank you to our presenting sponsor, Cash App. Please consider supporting
link |
the podcast by downloading Cash App and using code Lex podcast. If you enjoy this podcast,
link |
subscribe on YouTube, review it with five stars and Apple podcast, support on Patreon,
link |
or simply connect with me on Twitter at Lex Friedman. And now let me leave you with some words
link |
from Carl Friston. Your arm moves because you predict it will. And your motor system seeks
link |
to minimize prediction error. Thank you for listening. I hope to see you next time.