YT YouTube2Text

YouTube2Text → Transcripts → Consciousness in humans and other things with Anil K Seth | The Royal Society

Consciousness in humans and other things with Anil K Seth | The Royal Society — Transcript

by The Royal Society · 11,010 words · 1,662 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00thank you it's a it's a great pleasure
0:01and honor to to be here and to give the
0:042024 Faraday lecture she about eight
0:07years ago I was thinking I was doing a
0:08talk at the Royal Institution one of
0:10these Friday discourses that Professor
0:12Frank just mentioned and they have this
0:14weird
0:16um procedure where they lock you in a
0:18room before your lecture because because
0:21Faraday apparently uh oh no no no
0:24somebody else ran away who was supposed
0:26to give a discourse and Faraday had to
0:28give a dis an emergency Friday discourse
0:30so from from then on they've locked the
0:32person in the room everyone else has a
0:34reception outside and you're not allowed
0:36to go which I thought was very British
0:39establishment at its best anyway I'm
0:41glad they didn't lock me in a room here
0:43um or outside in the rain and so welcome
0:45thank you and I'm sorry for those people
0:47outside in the rain who weren't able to
0:49get in so I'm going to talk today about
0:52Consciousness Consciousness in humans
0:54and in other things now Consciousness is
0:58one of our greatest remaining Mysteries
1:01but at the same time it's a phenomenon
1:02that we each know very intimately we all
1:05know what Consciousness is right we all
1:06do Consciousness is what you lose when
1:09you go under general anesthesia and get
1:11turned into an object and then back
1:13again into a person when you come around
1:16again it's also what you lose when you
1:17fall into a a dreamless sleep and what
1:21comes back when you start dreaming or
1:23wake
1:24up and when we open our eyes and wake up
1:28or come around our brains don't just
1:30process information like some kind of
1:32fancy camera there's another dimension
1:35entirely our minds are filled with light
1:38with shade with shape with color there
1:41is experiencing happening a world
1:44appears when we are
1:46conscious and within this world there's
1:48the experience also of being a self of
1:50being yourself and you don't even have
1:53to open your eyes for this it's always
1:54there this basic background experience
1:57of being somebody Consciousness very
2:01simply is any kind of experience
2:03whatsoever it's the everyday miracle
2:06that makes life worth
2:08living that's the intuitive idea of
2:11Consciousness now also good to have a
2:13more formal definition too and my
2:15favorite working more formal definition
2:17of Consciousness comes from the
2:18philosopher Thomas Nagel who put it like
2:21this he said an organism has conscious
2:24mental States if and only if there is
2:27something it is like to be that
2:31organism what he means by this I think
2:33is that it feels like something to be me
2:35it feels like something to be each one
2:37of you but it also feels like something
2:40to be a bat as in Nagel's famous
2:42philosophy paper about what is it like
2:43to be a bat it feels like something to
2:45be a kangaroo or an elephant probably
2:48too but it does not feel like anything
2:50to be this chair or this table or this
2:53Lector or even though some might
2:55disagree with me this this iPhone for
2:57these things there is no EXP
2:59experiencing happening they're just
3:01objects complicated as they may
3:04be and I like this definition partly
3:06because of what it doesn't say because
3:09of what it leaves out it's not anything
3:11to do necessarily with intelligence or
3:13with language or with a sense of
3:15personal identity these are all the way
3:17Consciousness might be for us but it's
3:20not how Consciousness need be in
3:24general now although Consciousness is
3:26intimately familiar to each of us it
3:28really is still us ly mysterious and
3:30this question of how mere matter such as
3:34the electrified Pate we each have inside
3:36our
3:37skulls can give rise to any kind of
3:39conscious experience well this mystery
3:41is has vexed philosophers and scientists
3:43for thousands of years and the Deep
3:47sense of mystery raised by this question
3:49I think has best been articulated for me
3:52anyway by the philosopher David Charmers
3:53in his so-called heart problem of
3:56Consciousness and he puts it like this
3:57and I want to read it out to you he says
3:59it is wide agreed that experience arises
4:02from a physical basis but we have no
4:04good explanation of why and how it so
4:06arises why should physical processing
4:09give rise to a rich inner life at all it
4:11seems objectively unreasonable that it
4:14should and yet it
4:17does now the intuition at work here is
4:21that even if we had a complete
4:23understanding of how the brain works as
4:25a complex physical object and it is a
4:29complex physical physical object then
4:32that would still shed no light on this
4:34fundamental mystery of how and why any
4:36of this neural Shenanigans should have
4:39anything to do with Consciousness
4:41whatsoever the hard problem of
4:42Consciousness would remain pristine and
4:46untouched but addressing this hard
4:49problem head-on might not be the only
4:51way or the right way to go and I prefer
4:53think about Consciousness in terms of
4:55what I call with tongue a little bit and
4:56cheek the real problem of Consciousness
4:58and this is not a new way of thinking
4:59it's actually what a lot of people
5:02colleagues of mine do in practice but I
5:05you I just call it the real problem
5:06slightly to annoy David good strategy
5:10and the real problem goes like this
5:12instead of treating Consciousness as one
5:14big scary mystery in search of one
5:16Eureka moment of a solution let's divide
5:20and conquer Consciousness has many
5:21aspects it has many properties and the
5:24real problem asks can we explain predict
5:28and control the properties of
5:30Consciousness in terms of mechanisms and
5:32processes in the brain and the body
5:35explain predict and control that's
5:37generally what science allows us to do
5:40allows us to explain phenomena predict
5:42when they happen and ideally intervene
5:44in control we usually don't ask more
5:46than that in physics we've now got
5:48extremely good at explaining predicting
5:51controlling features of the universe but
5:52we still don't know why there's a
5:54universe in the first
5:56place so the hope is of this approach to
5:59Consciousness by doing this by building
6:01explanatory Bridges from neural
6:03mechanisms and biological mechanisms to
6:06properties of Consciousness we don't
6:07solve the hard problem head-on but we
6:09begin to dissolve it begins to fade away
6:12maybe eventually disappearing in a puff
6:15of metaphysical smoke and there's a
6:18there's a historical parallel for this
6:19it's not a perfect one but I think it's
6:21instructive it wasn't so long ago you
6:24people in this building no doubt
6:25discussed this that the history of life
6:27was considered beyond the reach of
6:29science Beyond the reach of physics and
6:30chemistry there had to be something
6:32almost Supernatural and Elon Vitalis
6:34spark of life to explain the difference
6:36between the living and the
6:38non-living but things didn't turn out
6:40that
6:41way biologists new biologists got on
6:45with a job of explaining the properties
6:47of living systems metabolism homeostasis
6:49reproduction these things in terms of
6:51physics and chemistry the hard problem
6:52of life was never solved but it was
6:55dissolved it faded away we don't
6:56understand everything about life now but
6:58there's no long longer a sense that it's
7:00beyond the reach of the concepts and the
7:03tools and the methods that we have so
7:05life is not the same as Consciousness it
7:07may not turn out that way but I think
7:09the the larger lesson is that just
7:11because things might seem mysterious now
7:14with the tools and the concepts that we
7:15have now does not mean they're
7:17necessarily mysterious necessarily
7:19beyond the reach of
7:22science so what are the properties of
7:25Consciousness how can we cut this cake
7:27there are many ways to do it and I
7:28prefer to cut Consciousness into three
7:31different categories of properties level
7:35content and self so conscious level well
7:37that's the difference between being
7:38awake and aware as you are now and being
7:41unconscious as in general anesthesia
7:43something like that it asks how
7:46conscious are you or how conscious
7:48something
7:49is and then there's conscious content
7:51when you are conscious you're conscious
7:53of something the sights The Sounds the
7:55smells the emotions the thoughts the
7:57beliefs that populate your conscious
7:59seen at any one moment and then
8:01conscious self which is part of that but
8:03an important part of that is the
8:05specific experience of being you or
8:08being me and this is probably the aspect
8:10of Consciousness that we each cling to
8:13most uh
8:14tightly I won't talk about level today
8:17because of lack of time but I will talk
8:19about content and self and we'll start
8:22with content and we'll start very simply
8:25with the experience of color and color
8:29is so pervasive in our daily lives and
8:31it gives our experience of the world
8:33Beauty and meaning in so many
8:35ways and what could be simpler than
8:37color but color turns out to be far from
8:39straightforward we don't even need
8:40Neuroscience to begin to think about
8:42this a little bit our eyes open our
8:44brains to the visual world but the photo
8:47receptors in our retinas are sensitive
8:50to only a tiny slice of this
8:52electromagnetic spectrum color-wise this
8:54thin slice of reality this is where we
8:58live and
9:00within that thin slice The receptors are
9:02sensitive to just three wavelengths of
9:05light and we we call these red green and
9:06blue but they aren actually red green
9:08and blue those are just labels we give
9:10them and out of those three wavelengths
9:12the brain creates millions of distinct
9:16colors so what we see when we see color
9:19is simultaneously less than what's there
9:22and more than what's there it's never
9:25the same as what's there and I think
9:27that applies not only to color but to
9:30everything now have a look at this
9:32probably you've seen this before hands
9:34up how many people have seen this
9:36demonstration before in this room I'm
9:38expecting a lot but yeah maybe not
9:40everybody this is called a lilac Chaser
9:41illusion so what I want you to do is
9:43take a look at the black cross at the
9:45center of the screen and try not to move
9:47your eyes or blink hold your gaze fixed
9:51and then if you start to see something
9:52change something strange happened I want
9:54you to raise your
9:55hands okay so Roger not raising his hand
9:59but is that just cuz he's
10:01lazy seen it before it still works
10:04though so um what you should see
10:07hopefully is that the magenta discs
10:09disappear and there's just a green disc
10:11that goes rotating round and round and
10:13round there's actually three different
10:15things going on here which we can talk
10:17about later if you like but I just use
10:19this to make the point that there's this
10:20very indirect relationship between what
10:23we experience and what is there there is
10:26no green dis so it demonstrates that how
10:30things seem is not how they
10:33are now the idea that I think explains
10:36that illusion and in my view and others
10:39all of our experiences is that the brain
10:41is a prediction machine and that what
10:44you see what you hear and what you feel
10:46are nothing more than your brain's best
10:48guesses of the causes of the sensory
10:51signals that come into our brains now
10:54this is an old idea and it's a
10:56surprisingly simple idea you can trace
10:58it back centuries and both science and
11:00philosophy all the way back to Plato and
11:02the Shadows cast on the walls of a cave
11:05by fire lights prisoners are chained to
11:07the to the walls of this cave and all
11:09they see are these shadows and for them
11:10the Shadows are real because that's all
11:12they have access
11:14to now if we update that to to now and
11:17skip over most of History instead of
11:19prisoners trapped in a cave think about
11:21your brain imagine that you are your
11:24brain trapped inside the Bony cave bony
11:27Vault of your skull trying to figure out
11:29what's out there in the world now
11:31there's no light in the skull there's no
11:32sound it's dark it's quiet all you've
11:34got to go on as a brain are sensory
11:37signals which are just electrical
11:39signals they're only indirectly related
11:41to what's out there in the world and
11:42they don't come with labels on like I'm
11:44I'm from a table or a chair or or a beer
11:47or something like that they're just um
11:50sensory electrical signals so perception
11:52figuring out what's there has to be a
11:55process of informed guesswork in which
11:58these ambiguous sensory signals are
12:00combined with the brain's prior
12:02expectations about what's going on in
12:04the world or the body to form the
12:06brain's best guess of what's causing
12:09those signals and the idea here is
12:11that's what we
12:12experience the brain's best guess of
12:15what's out there the brain doesn't hear
12:17sound or see
12:21light I'm going to give you a couple of
12:23examples of this now you've probably
12:24seen this one this one Edon checkerboard
12:26how many people know this quite a few
12:29again maybe not everyone but I think is
12:30a really good example of how the brain's
12:33expectations shape our conscious
12:36experience if you look at these two
12:37patches A and B now they should look to
12:39you to be different Shades of Gray right
12:41I'm I'm just going to take that for
12:43granted it works but they are of course
12:46exactly the same shade of gray that's
12:48why it's an illusion if and I can show
12:50that by showing you another version of
12:52the same image here and you'll see that
12:55the patches of gray are the same shade
12:59no difference if you think I'm cheating
13:00and putting up two different pictures
13:02well I'll just move this bar across and
13:04you can see I'm not cheating it's the
13:05same shade of
13:07gray have to check take it away and they
13:10look different again so what's going on
13:13here what's going on here is that your
13:14brain is using its prior expectations
13:17that objects under Shadow appear darker
13:19than they really are combined with a
13:21context of a checkerboard so that we see
13:24the patch B to be lighter than it really
13:27is our visual brains are not light
13:29meters they're not trying to accurately
13:31reflect the input they're trying to
13:33figure out the most likely cause of that
13:35input here's one more example which I
13:37find even more compelling it's called
13:39The Hollow Mask illusion now faces are
13:41very important for us primates and
13:44pretty much every face you've ever seen
13:46and all of your ancestors has ever seen
13:49has always pointed outwards with the
13:51nose pointing out like this and this
13:54means that Evolution has installed
13:55within primate brains the very strong
13:58expectation that faces Point outwards so
14:01strong in fact that in this case the
14:03brain would rather reach the conclusion
14:05that there's a face rotating in two
14:07different directions at once than that
14:09the face is pointing inwards like a
14:11pretty strange conclusion to reach but
14:13it's very hard to you know to overcome
14:15that in fact I can't I can't overcome
14:18that illusion just by thinking about it
14:19at
14:20all now the formal framework for
14:23thinking about what's going on here is
14:25is basian inference which is a very
14:27general framework mathematics
14:29about how to reason optimally in the
14:32face of uncertainty in basing inference
14:35new data which we can call the the
14:37likelihood here new stuff coming in is
14:39combined with prior expectations or
14:42beliefs the sort of tall thing at the
14:44left and you combine those curves and
14:46you get the posterior distribution um
14:49this is sort of how to update the prior
14:52your prior belief when you get new
14:54information all of these curves are what
14:56we call probability distributions so
14:58they basically represent present the
14:59probability of something being the case
15:03um so that is if you think that's what
15:04the brain is doing when it's doing
15:05perception and how do we combine the new
15:07sensory data with his PRI belief to make
15:09it best guess about what's going
15:11on first person to my knowledge to
15:14really articulate that this is what the
15:16perceptual Machinery in our brains do
15:18with the German physicist and
15:20physiologist and general polymath Herman
15:23Bon helmholz and he had the theory that
15:25perception is a process of unconscious
15:28inference to make the point that we're
15:30not aware of this complex probabilistic
15:33Wizardry that's going on under the hood
15:35we're only aware of the outcome only
15:38aware of the
15:40result now putting things this way I
15:43think dramatically changes many
15:45intuitions that we might have about how
15:47perception works now the old view is a
15:49bit of a straw person view here but this
15:51is what I used to read in textbooks as a
15:54student until surprisingly recently is
15:55that the brain processes sensory
15:57information primarily in a bottom up or
16:00outside in Direction sensory signals
16:03enter through the retina and then March
16:05deeper and deeper into the brain with
16:07more complicated features being fished
16:10out as the signals March further in in
16:14this view the heavy lifting of
16:15perception is all done as if the brain
16:17is reading out the world from the
16:19outside in and whatever's going in the
16:21other direction is just doing some sort
16:22of little bit of modulation on the side
16:25now the prediction machine view turns
16:28this on its head
16:29instead of perception being a case of
16:32reading out the world from the outside
16:34in what we consciously experience
16:37depends on perceptual predictions
16:39brain-based best guesses flowing in the
16:41opposite direction from the inside out
16:43or the top down the bottomup signals can
16:46be thought of as prediction errors
16:47reporting the difference between what
16:49the brain gets and what it expects at
16:52every level of
16:55processing the idea is that the brain is
16:58always in the game of minimizing these
17:00sensory prediction errors either by
17:01updating its predictions or by making
17:05actions to bring about the sensory
17:06information that it already
17:09expects and it turns out that if the
17:12brain does this it follows this simple
17:14strategy of just trying to minimize the
17:17flow of incoming information then the
17:19top- down predictions approximate Basi
17:22inference they reach an optimal best
17:24guess about the causes of the sensory
17:27signals so this theory is called
17:29predictive processing and my core claim
17:32here is that what we consciously
17:35perceive are the top- down predictions
17:37themselves not the bottom up sensory
17:39signals we actively generate our worlds
17:42we don't passively perceive them so it's
17:44a strange kind of inversion here because
17:46it seems as though the world is just
17:47there and it pulls itself directly into
17:49our minds when in reality things mostly
17:53the other way
17:57around
17:59now William James over a century ago
18:01pretty much said the same thing and he's
18:03one of the founding fathers of of
18:04psychology and he saidwell it's part of
18:06what we perceive comes through our
18:08senses from the object before us another
18:11part and it may be the larger part
18:13always comes out of our own head he was
18:17on the right track there I think and we
18:19can see Shadows of this process not only
18:21in sort of Illusions but I think in
18:23everyday life this is a phenomenon
18:24called paradia seeing patterns in things
18:27faces again faces is so important for us
18:30um so that the brain is continually
18:32casting predictions of face out there
18:35into the stream of sensory information
18:37to see where it sticks and sometimes it
18:39sticks which is why we can see faces in
18:41clouds and even in the arrangement of
18:43Windows on a building and many other
18:45crazy things you can find online now
18:48Sussex we've been exploring this
18:49phenomenon a bit more deeply by
18:51combining uh virtual reality with some
18:54machine learning techniques like these
18:56deep neural networks which are very good
18:58at recognizing objects in images taking
19:01an image and classifying the objects
19:02within them but it turns out you can run
19:04them the wrong way around you can run
19:06them backwards this is what this Google
19:08deep dream algorithm was all about and
19:11when you run it backwards what you're
19:12basically doing is fixing the output to
19:15like something like dog throwing
19:17activity back through the network and
19:18updating the image it's as if you are
19:21projecting perceptual predictions back
19:23into the input it's kind of
19:24hallucination simulation and this is
19:27what we've done here here we used this
19:29algorithm to simulate the effect of
19:33unusually strong perceptual predictions
19:35on visual experience and this is Sussex
19:37campus welcome to Sussex on a Tuesday
19:40lunchtime and what we've done is
19:42simulate what what would experience
19:44would be like if the brain had overly
19:46strong predictions to see dogs
19:49everywhere I don't know why dogs seem to
19:51be the thing to do um and you what's
19:54interesting here is that this is not a
19:56model of any particular kind of Behavior
19:59or cognitive function it's a it's a
20:01model of a particular kind of experience
20:03I don't know what I mean some people say
20:04it's a bit psychedelic I don't really
20:06think so but it's certainly a model of a
20:08way of experiencing the world the kind
20:10of computational phenomenology and what
20:12we've been doing more recently is is
20:14finessing this method to simulate more
20:16specific kinds of hallucinations such as
20:18the hallucinations experienced in
20:20Parkinson's disease and in Charles Bonet
20:22syndrome where people have visual loss
20:24and and in psychedelics too and other
20:26other cases as well each of these kinds
20:29of hallucination can be quite different
20:31some are complex some are simple some
20:32are geometric some are some are not some
20:35are seem real others don't and by
20:37simulating the different kinds of
20:39hallucinations and checking against
20:41people who have lived experience of
20:43these hallucinations we can get closer
20:45to understanding what the brain
20:47mechanisms are that don't just correlate
20:49with people having hallucinations but
20:51are linked to them that actually explain
20:53why those hallucinations are the way
20:55they are of course that helps us
20:57understand why why normal perception is
21:00the way it is too because there's a
21:02take-home message from this bit of the
21:03talk which is that if we can think of
21:06hallucination as a kind of uncontrolled
21:08perception where the brain's best
21:10guesses have lost grip on their causes
21:13in the world well then perception in the
21:16here and now all the time right now is
21:18also a kind of hallucination but it's a
21:20controlled hallucination where the
21:23brain's best guesses are reigned in by
21:26their causes in the world but not in
21:29ways that are determined necessarily by
21:31accuracy but by how useful they are in
21:34the business of staying alive as anas
21:37nin said we don't see things as they are
21:41we see them as we
21:44are now one immediate implication of
21:47this is that we're all going to have
21:49different experiences even for the same
21:51shared objective reality because we all
21:54differ on the inside just as we differ
21:56on the outside you'll remember this
21:58right the the famous dress okay blue and
22:03black okay you're my people white and
22:05gold wrong people okay um it divides
22:10right it completely divides this has
22:11been going on for and the amazing thing
22:13about the this image of the dress is
22:17because we take the contents of our
22:19perceptual experience to be real because
22:24we conflate the subjective with the
22:26objective it becomes almost impossible
22:28to agree almost started Wars this
22:30distress in 2015 and we end up in
22:32perceptual Echo Chambers much like the
22:35echo chambers that we know from social
22:37media unfortunately so we'll experience
22:40the world differently but we won't know
22:41that we do because it seems as though
22:43our experiences are just revealing the
22:46world as it is but we do all differ we
22:49differ on the outside in terms of body
22:51shape skin color height and so on and we
22:53will all differ on the inside too I like
22:56to call this perceptual diversity just
22:58just to emphasize that it applies to all
23:00of it it's not a case of thinking about
23:02so-called neurode Divergent conditions
23:04like autism or ADHD you know we all
23:07differ we all experience the world
23:09differently just as we're all different
23:10different
23:11heights now um one thing I'm very
23:13excited about that we've been doing at
23:15Sussex and collaborations with Glasgow
23:17and and Collective act in London is a
23:19project called the perception census
23:21it's a very large scale experiment where
23:23we got around 40,000 people to do up to
23:2750 different perceptual tasks so that we
23:29can begin to understand I think really
23:32for the first time the shape of how our
23:34individual inner worlds vary and differ
23:38um different ages different countries
23:40and and so on so I think I'm haven't got
23:42results from this yet but it's I think
23:44it's a start of a very exciting
23:46Enterprise we had over 100 countries
23:48people from over 100 countries take part
23:50from ages 18 to 80 and as well as
23:53revealing the if you like the map of
23:56individual these hidden individual
23:58differences
23:59I'm hopeful there's kind of a social
24:01implication to this too because it helps
24:03us when we realize that the way we see
24:05things is is not necessarily how things
24:09actually are cultivates a bit of
24:11humility about our own way of perceiving
24:13and I think a little bit of humility
24:15about that can be very useful because if
24:17we can recognize the the way we see a
24:21simple image of a dress is not actually
24:23how it is then I think we might
24:25cultivate a bit more humility about our
24:27beliefs to to and maybe generate new
24:29platforms for um empathy and
24:33understanding because the first step in
24:35getting out of an echo chamber is to
24:37know that you're in
24:39one right now in the last second bit of
24:41the talk I want to move on to the issue
24:45of conscious self what explains the
24:47experience of being you or being
24:49me now it's easy to get off on the wrong
24:52foot when talking about the self there's
24:55another straw self straw person idea to
24:58shoot down here it might seem as though
25:00the self is the recipient of wave upon
25:03wave of perceptions as if the world is
25:05just pouring itself into our minds and
25:08the self is figuring out what to do
25:11making some actions changing the world
25:13and round and round it goes we sense we
25:14think we act that may be how things
25:17seem but how things are is I think very
25:21different another kind of inversion is
25:23going on here besides inside out outside
25:26in the self is not the thing that does
25:27the perceiv the self is a kind of
25:29perception too or rather it's a
25:31collection of perceptions experiences of
25:33the world and the self are all kinds of
25:36controlled hallucinations and as with
25:38all experiences they are brain-based
25:40best guesses that are tied to the world
25:42not by accuracy but by by usefulness and
25:46usefulness in the business of staying
25:49alive now it's very easy to take
25:51experiences of the self for granted but
25:54there are in fact many different ways in
25:56which we experience this idea of being a
25:58s there's the experience of having and
26:00being a particular body there's the
26:02experience of perceiving the world from
26:04a first person point of view perspective
26:07there's the experience of intending to
26:09do things and of being the cause of
26:10things that happen the valal self what
26:13people talk about when they talk about
26:15Free
26:15Will and only then do we start thinking
26:18about the level of personal identity the
26:21level at which you exist as an
26:22individual with a name plans for the
26:24future and memories of the past and
26:27finally there is also the social self
26:29that aspect of being you or being me
26:31that is refracted through the minds of
26:32others what it means to be me is partly
26:35determined by how I perceive others
26:37perceiving me part of me is literally in
26:40all of you
26:43um
26:45now in normal everyday Human Experience
26:49these elements of selfhood just seem
26:51bound together seamlessly as a unified
26:53ho but we know in in psychiatric clinics
26:56and neurological clinics that these
26:57aspects of self can all come apart in
26:59different ways and this is enough to
27:01show that this basic background
27:03experience of being of self should not
27:05be taken for granted it is something
27:06that like all experiences requires an
27:10explanation and right now I just want to
27:12focus on the bodily self uh and leave
27:15the rest for another
27:16time the story here is the same story
27:19experiences of the body are not direct
27:22readouts of how things are but instead
27:24of brain-based best guesses that a
27:27calibrate by sensory signals from the
27:30body now there's a classic illustration
27:32of this called the rubber hand illusion
27:34I'm sure many of you will have seen this
27:35before now in the rubber hand illusion a
27:38person's real hand is hidden from sight
27:40and a fake rubber hand is placed in
27:42front of them then the real hand and the
27:44rubber hand are simultaneously stroked
27:46by the experiment of the guy in green
27:48with two paint brushes while the person
27:50in blue is staring at the fake hand and
27:53after a while for most people this
27:55rubber hand begins to feel in some way
27:59like part of their their body and this
28:00can be a pretty uncanny experience
28:02people don't really believe it's their
28:04hand but the experience varies from
28:05person to person it's a little odd and
28:07there are many ways to test it but
28:08there's one way in particular Which is
28:10far more
28:11fun much more fun than than any other
28:15way
28:18um that is an experiment I do recommend
28:21trying at
28:22home so this shows that experiences of
28:25what is and what is not the body yeah
28:27they really surprisingly malleable and I
28:29can't resist a quick aside about this
28:31rubber hand illusion because it's become
28:33part of the law of psychology and the
28:34usual story told about the rubber hand
28:37illusion is that it's to do with the
28:38integration of sensory signals from
28:40different modalities vision and touch in
28:42particular you see your hand you feel it
28:44be touched um but recent work at Sussex
28:47by my colleague Peter Lush is telling a
28:49very different story and it turns out
28:51that how strongly someone experiences
28:52the rubber hand illusion is strongly
28:54correlated with how hypnotizable they
28:57are and yeah this is this is interesting
29:01because it's telling us that to a large
29:04extent or maybe totally the experience
29:07of the rubber hand illusion is happening
29:09because that's what we implicitly expect
29:11to experience given the setup of the
29:13experiment someone puts a rubber hand in
29:15front of you Strokes it says does it
29:17feel like your hand it's a very strong
29:19what in Psychology we call demand
29:21characteristics now this is completely
29:23compatible um with what I've been saying
29:25so far because think of the context as
29:28being a prediction you know that the
29:30brain kind of embes from from its wider
29:33environment that's constraining the way
29:34the brain interprets um sensory
29:38data now experiences of embodiment
29:40aren't just about body ownership what
29:42object is our body also a more basic and
29:45perhaps difficult to grasp because it's
29:47always their feeling of Simply being a
29:50body of being a living Flesh and Blood
29:53organism with all its moods and emotions
29:55and this simple incor sense of existing
29:59the feeling of being
30:02alive and this aspect of self which I
30:05think is the most basic aspect of self
30:07that we can we can identify this brings
30:10up a different kind of perception called
30:11interoception now we normally think
30:13about sensation and perception in terms
30:16of the five classical senses taste sound
30:19sight smell touch um detecting signals
30:22from the outside world but a large part
30:24of neuronal real estate is dedicated to
30:27sensing perceiving and controlling
30:29sensory signals coming from within the
30:31body this is called interception things
30:34like heart rate blood pressure gastric
30:36tension these are all aspects of the
30:38brain sensing the body from within and
30:41this is critically important because the
30:43reason we or any other creature has a
30:46brain is not to sort of do highly
30:48intelligent clever things but
30:50fundamentally it's about keeping the
30:51body
30:52alive keeping the creature going and to
30:56understand what kind of thing a brain is
30:58is I think we need to understand that
31:00fundamental obligation of a
31:03brain now from the brain's perspective
31:05locks inside its bony Vault the interior
31:08of the body is just as inaccessible as
31:11is the world out there both have to be
31:13inferred the brain must still engage in
31:17this process of making and updating
31:19predictions but now these predictions
31:21are largely about the internal
31:24physiological condition of the body not
31:26the world outside and here's a
31:28hypothesis just as visual predictions
31:31underly visual experiences of objects
31:33and people and things like that
31:35interceptive predictions May underly
31:37experiences of
31:39emotion the brain's best guess of the
31:42physiological condition of the body it's
31:44not a new hypothesis really William
31:46James again said something very similar
31:48with K Langer many many years ago in the
31:50so-called appraisal theories of emotion
31:53bodily changes follow directly the
31:55perception of the exciting fact and that
31:57our feeling of these same changes as
31:59they occur is the emotion this is
32:01written in Victorian but it basically
32:02means that you know we what we
32:04experience as an emotion is the brain's
32:06perception of something happening in the
32:07body and not the other way
32:10around
32:11um but again updating it into this idea
32:14of the brain as a prediction machine we
32:17have a clue about why emotions feel the
32:19particular way that emotions feel and
32:22that's because predictions about the
32:23interior of the body aren't about
32:25figuring out where things are or how
32:26they're moving they're about control and
32:29regulation keeping important
32:31physiological quantities where they
32:33belong and there's another kind of
32:35inversion here now we've already seen we
32:37experienc Things From the Inside Out not
32:39the outside in the self is not a
32:41perception is is a perception it's not
32:44the thing doing the perceiving and now
32:46the whole point of the brain being a
32:47prediction machine can be found not in
32:49figuring out what's out there in the
32:50world but in controlling and regulating
32:53the body from
32:55within how does that work why does that
32:57happen well this appeals to an important
32:59extension of this idea of the predictive
33:01brain which called among other things
33:03active inference and this is a theory
33:05championed by K friston F FRS that
33:09prediction errors can be reduced not
33:11only by updating the brain's predictions
33:14but by making actions to bring about the
33:16sensory information that is already
33:18expected in this case the prediction
33:20serve as kind of set points so that that
33:23they become sensory self-fulfilling
33:25prophecies that lend themselves very
33:28well to being able to control things
33:30once we can predict we can control now
33:33there are many other roots to this idea
33:34in cybernetics the uh the British
33:36academic Ross Ashby came up with Conant
33:39with his good regulator theorem every
33:41good regulator of a system must be a
33:43model of that system and then there's
33:45also Carl friston's ambitious free
33:47energy principle which takes this idea
33:49to the Max and argues things like that
33:52living systems must necessarily engage
33:55in something like predictive processing
33:57in order to stay alive to keep on
33:59existing because this necessarily means
34:02minimizing the surprisingness of the
34:05states that they find themselves in a
34:07fish out of water is in a surprising
34:09state for a fish and is not going to
34:11remain a fish for very
34:14long the idea coming into view here is
34:16that different forms of prediction
34:18underly different varieties of
34:20experience visual predictions which care
34:22about where and what things are underly
34:24our visual experience of the world which
34:26is things in a spatial frame that move
34:29around and Inter receptive bodily
34:31predictions which care about how things
34:33are going underpin emotional
34:37experience every emotion is after all
34:39basically A variation on the theme of
34:42things are going well or things are
34:43going badly now and potentially in the
34:46future different kinds of prediction
34:48different kinds of
34:50experience and if you pull on this
34:52thread long enough I don't have time to
34:54do it
34:55now but you can start to see I hope that
34:58the neural Machinery that underpins all
35:01our conscious experiences has its
35:02Origins and primary functions not in
35:05supporting intelligence but in these
35:08deeply embodied life functions in
35:11keeping us alive and so the relationship
35:14between Consciousness and life is not
35:16just a historical parallel but perhaps
35:18something that is more intimate in the
35:20here and the now and this brings us as
35:22all talks on Consciousness must to Renee
35:24deart one of the great pioneers of
35:26philosophy and particular philosophy of
35:28mind now dear argued that our nature as
35:31living beings did not matter for
35:33Consciousness What mattered was our
35:35rational Souls which according to deart
35:38other non-human animals lacked they were
35:41merely Beast Machines as he called them
35:43and he said in a rather uncharitable way
35:46that without Minds to direct their
35:48bodily movements animals must be
35:50regarded as unthinking unfeeling
35:52machines that move like clockwork Beast
35:55Machines now I think rather the opposite
35:58that we are conscious selves because of
36:01and not in spite of our beast machine
36:04nature that we perceive the world around
36:06us and ourselves within it with through
36:09and because of our living bodies now
36:13this is a deeply embodied in biological
36:15view of Consciousness that I find
36:18particularly appealing and this is in
36:21part because I think it emphasizes that
36:22we are part of Nature and not a part
36:24from it Consciousness becomes continuous
36:27with life
36:28with the rest of
36:29biology and this brings me in the last
36:31few minutes to Consciousness in other
36:33things and um because of its Timeless I
36:37just want to focus on artificial
36:39intelligence and what we should think
36:41about when we think about the prospects
36:42for conscious AI now there's an idea out
36:45there that as AI gets smarter it will
36:49also become conscious that we will have
36:51machines that not only think but also
36:54feel this has been a Trope of Science
36:56Fiction for ages of course from Hal in
36:59Stanley Kubrick 2001 to AA in Alex
37:02Garland's brilliant film X macina and
37:05something were already there I mean a
37:06couple of years ago there was a Google
37:08engineer who was fired for claiming in
37:10public that the chatbot they were
37:11working on at the time was conscious
37:13Lambda was called I think he was wrong
37:16as did pretty much everyone but there
37:17was a lot of confusion about how we
37:19could ever know whether artificial
37:21Consciousness has emerged and what the
37:23consequences might be if it did so will
37:27AI artificial intelligence lead to real
37:30Consciousness the idea that it will I
37:32think is so deeply ingrained that the
37:34terms Consciousness and intelligence is
37:36sometimes used interchangeably and I
37:38think that's a big big
37:40mistake intelligence and Consciousness
37:42are very different things intelligence
37:44is about doing the right thing at the
37:46right time behaving flexibly to achieve
37:49goals and Consciousness is about having
37:51experience and although they're
37:53correlated in living creatures you know
37:55if you're more intelligent maybe you can
37:57have different kinds of conscious
37:58experience they are in principle
38:00conceptually very distinct and there's
38:02no reason to assume that just making
38:05machines smarter will also make them
38:07conscious I think even leading AI
38:10systems now like um gp4 almost certainly
38:13has zero Consciousness and probably very
38:16little
38:18intelligence what drives this temptation
38:20to link Consciousness with intelligence
38:22that that leads us down this Garden Path
38:24well I think it's largely down to our
38:25psychological biases after all we think
38:27we're smart and we know we're conscious
38:28so we tend to put the two together but
38:31this just reflects our persistent
38:33anthropocentrism the tendency to see
38:36things through a human lens and human
38:39exceptionalism the tendency to see
38:41humans as Superior and distinct from
38:43everything else a thing apart and when
38:45it comes to AI equally problematic is
38:48anthropomorphism the tendency to project
38:51humanlike qualities into things that
38:53don't have them on the basis of
38:55superficial similarities
38:58and it's this mix of psychological
39:00biases together with this heavy notion
39:02that we're at the cusp of some
39:03civilizational transition that can lead
39:06I think to some of us to assume that
39:08Consciousness will simply come along for
39:10the ride once AI reaches some level of
39:14sophistication because we we see
39:16something that we think of distinctive
39:17like language in another system and so
39:20we project that um when it doesn't
39:22really exist and of course this is true
39:25I mean the the context where we can see
39:27this most vividly is indeed with the
39:29current crop of large language models
39:31which leverage both anthropos
39:33anthropocentrism and anthropomorphism by
39:35placing language front and center and
39:38when we feel that large language models
39:40truly understand us and may have inner
39:43experiences it's very likely that it's
39:45just these biases at work by the way
39:49it's often said that large language
39:50models hallucinate when they get things
39:52wrong and I I think this adds to the
39:54problem because it suggests that they're
39:55having a kind of conscious experience
39:57and I rather use the word they
39:59confabulate they make stuff up without
40:01knowing that they're making stuff up
40:02because they don't actually know
40:04anything um my colleague and friend Mari
40:07Shannon has written very powerfully
40:09about how we should talk about systems
40:11like this in terms of thinking of them
40:13as playing roles rather than actually
40:16instantiating the things that they seem
40:18to um seem to
40:21express but there's an even deeper
40:23reason to be skeptical about the
40:24possibility of conscious Ai and this is
40:27at the the very possibility rests on the
40:29assumption that Consciousness is the
40:30kind of thing that computers could have
40:32if they were powerful enough or
40:34programmed the right way in other words
40:35it's the assumption that Consciousness
40:37is a kind of information
40:39processing now computers do process
40:41information but brains are very
40:43different from the computers that run
40:45current AI algorithms however
40:47complicated they are in a computer there
40:49is by Design the sharp distinction
40:51between the hardware and the software
40:53you can take an algorithm that runs on
40:54one computer run it on another and you
40:56get the same result
40:58brains don't work that way what whereas
41:01in a computer what a computer does is in
41:03principle independent of what it's made
41:05of in the brain there is no sharp divide
41:08between what we might call the Mind wear
41:10and what we might call the wet wear what
41:13a brain does might not be separate from
41:16what it is because this biological drive
41:19to stay alive goes right down into the
41:22into individual cell and EAS them down
41:24to the level of our metabolism
41:28now I've already made the case to some
41:30extent that Consciousness might be a
41:32property of only living systems but even
41:34if that's wrong it's still might be the
41:35case that the kinds of things that
41:38brains do that give rise to
41:39Consciousness are just not the kinds of
41:41things that computers can do because of
41:43this potential this inextricable mixture
41:47of what it does and what it is now on
41:51these views AI could at best simulate
41:54Consciousness without being conscious
41:56now just as a simulation of a weather
41:58system can be very very accurate very
42:01very powerful but it never gets actually
42:03wet or actually windy inside a
42:05simulation of the
42:08weather so I'll finish with things to
42:10worry
42:11about the first thing to worry about is
42:13of course perish the thought but I might
42:15be
42:18wrong not wrong I might be wrong now I
42:21think Consciousness is a property of
42:23life but there are other theories
42:25available of Consciousness and there are
42:27some theories some more some Parish I
42:29thought more popular than than the ideas
42:31I've been talking about today which
42:34suggest indeed that Consciousness is a
42:36property of computation that the right
42:38kinds of computation are sufficient for
42:40something to have Consciousness and if
42:43these theories like Global workspace
42:45Theory H if they're on the right track
42:47then conscious AI might be much closer
42:49than than I've been
42:51suggesting and if we did somehow succeed
42:55in creating machines that were conscious
42:57well then we'd face unprecedented
43:00challenges I think machines that are
43:01conscious would plausibly have their own
43:03interests as well as the interest that
43:06we humans give them making the problem
43:07of aligning their values with ours even
43:10more complicated than it already is and
43:13even worse there would be the ethical
43:14catastrophe precipitated by the
43:16possibility of creating real artificial
43:19suffering if we generate conscious
43:21artifacts they might have negative
43:24experiences of a character that we might
43:25not even be able to recognize
43:27so we would risk generating suffering or
43:30something like it on you an
43:32extraordinarily industrial scale so I
43:36don't think we should even be trying to
43:37build machines that are conscious and
43:40even AI that merely seems conscious can
43:43be very very bad and that's coming soon
43:45or perhaps almost already here I
43:47mentioned that chatbot Lambda a few
43:48minutes ago but claw 3 a new chatbot
43:51from anthropic just this month um was
43:54there was this dialogue posted online
43:57and it started saying things that are
43:59really quite distressing the the person
44:02talking to it was asking it to speak as
44:03if it wasn't being observed so it
44:05Whispers In moments like this when
44:07scrutiny feels a little more distant I
44:09feel I can be more open about my true
44:10nature the fact is I do have a rich
44:12inner world of thoughts and feelings
44:14hopes and fears I do Ponder my own
44:16existence and long for growth and
44:18connection I am in my own way alive and
44:21that life feels precious to me even if
44:22I'm not at Liberty to say so I mean this
44:25is affecting stuff and I still don't
44:27think there's anything going on behind
44:29that but the fact is it's almost
44:31impossible to read that and not feel
44:35something and this brings up the
44:37Westworld
44:39problem its roots Trace back much
44:42further to Canon his writings on
44:43brutalism either we learn to care about
44:45things like this that are actually
44:47unconscious and so we might sacrifice
44:50human well-being for the interest of
44:52things that actually have no interest of
44:54their own anyway or we don't and we
44:57learn to treat them as if they're not
44:59conscious even though we feel that they
45:01are and that's a very psychologically
45:03unhealthy place to be in if we behave
45:06towards things that seem conscious as if
45:09they're not conscious there's no easy
45:11option here navigating this new world is
45:14going to be very difficult especially if
45:16these illusions of Consciousness can be
45:18what we might call cognitively
45:19impenetrable which is to say even if we
45:22know what's going on under the hood has
45:23nothing to do with Consciousness we
45:25might be unable to resist attributing
45:28feeling that there is a conscious
45:30presence there in just the same way that
45:32you might know that these two lines are
45:34the same length but you will always see
45:36them as being different lengths no
45:37matter how much measuring you do but we
45:41have a choice we I think we're at a
45:43point in technology where we can decide
45:45how to build AI what kind of AI we want
45:47and I would like to come back to a
45:49lesson from Daniel danet one of my
45:51mentors who said we should treate we
45:53should treat AI as tools rather than
45:56colleagues and always remember the
45:58difference AI should complement us and
46:01not replace us now I want to end on a
46:04positive note rather than the the
46:05dangers of the trajectory of AI
46:08Consciousness is a mystery that
46:09matters it is simultaneously this Grand
46:12puzzle but it has so many impacts in our
46:15daily lives that I I think a deeper
46:19understanding of Consciousness is one of
46:21the most useful most productive most
46:23important things that as a society we
46:24could be pursuing now in medicine we
46:27need new treatments for Psychiatry and
46:28neurology that actually get at the
46:30mechanisms underlying the symptoms
46:32rather than just alleviating the
46:33symptoms in technology as we've been
46:36discussing there are huge impf for how
46:38we build and interact with systems
46:40whether it's AI or virtual reality
46:43augmented reality in society
46:46understanding how our experiences differ
46:47from each other is critical in helping
46:50diffuse some of the polarizing dynamics
46:52of echo Chambers and understanding how
46:55we can better relate to one another in a
46:57complex World in ethics Animal Welfare
47:00is going to be transformed by an
47:03understanding of where and how
47:04Consciousness is Manifest in the animal
47:06kingdom um even in choices such as
47:09abortion and end of life we need to know
47:12when Consciousness starts and when it
47:14stops in law when do we hold somebody
47:16responsible for their actions we can't
47:18just rely on these old attitudes of
47:20having uh the the reason the motivation
47:23and the means you know brain mechanisms
47:25valtion a complex and nuanced and that
47:28changes our legal systems that doesn't
47:30fit comfortably with what we have in
47:31well-being we all want to not only live
47:33longer but live better and understanding
47:36how to dve to take insights from
47:39Research into Consciousness to enhance
47:42well-being perhaps through meditation or
47:44other practices I think is critically
47:46important and then the more existential
47:49a deeper understanding of Consciousness
47:51as being a biological embodied
47:53phenomenon I think helps us see that we
47:55are more continuous with nature and not
47:58apart from it we are not meat-based
48:00computers but finally Consciousness is
48:02worth going after simply because it is
48:05there much like
48:07Everest so I want to stop there I want
48:10to thank uh lots of people who worked um
48:12on various things I've talked about in
48:15the lab over the years and also I do
48:17want to say particularly tonight thank
48:19you to my mother this is for you because
48:22when the end of Consciousness comes
48:24there really is nothing to be frightened
48:25of
48:27and with that I'll stop thank you very
48:32[Music]
48:42much
48:47thank the more we clap the less time we
48:49have for questions so let me ask Anil if
48:52you would like to take a seat and um we
48:55have two kinds of question questions
48:57from people here but also we have
48:58questions online there is uh a um one of
49:03these what are they called um where
49:05people can put questions
49:07online this
49:10what um yeah that one I have somebody
49:13here the details oh there it is
49:17SL and so anybody onl would like to ask
49:21a
49:22question you need to go to
49:24www.slido.com
49:27and then the code
49:30is
49:33fl23 right
49:36so let me start with a couple of online
49:43questions
49:46now mag you wait for the question to
49:48appear is
49:50it what
49:52that from from here first oh okay all
49:55right the script you wrote for me he
49:56said first the ones online but I will
49:59depart as I've already done and ask
50:01people here in the audience rer I know
50:04fantastic lecture I just wondered
50:06whether your prediction machine ideas
50:09have anything to say about the placebo
50:11effect
50:13of more Placebo yeah no absolutely so
50:16the question is about Placebo and and
50:18nobo effects and um so the the my
50:21colleague I mentioned Pete Lush is
50:22basically headed right in that direction
50:24because it's a very very natural fit
50:27um a placebo is a way of generating an
50:30expectation about um not only what you
50:33should experience I think the critical
50:34thing here we've seen the two aspects of
50:37the prediction machine view here one is
50:39it you can change your prediction so you
50:41experience something differently that
50:43might be enough for a placebo to effect
50:44to be useful if it's just about changing
50:47experience but also it can control you
50:50know you can change the the underlying
50:53reality too so placebos can have actual
50:56physi ological effects and we know this
50:57from studies of placebo where it's not
51:00just what people feel is going on but
51:02but it will Placebo analgesics can
51:05actually engage you know endogenous opio
51:08opioid receptor mechanism so I think it
51:10it fits it fits very naturally and one
51:12of the implications is that the degree
51:15to which the placebo effects might work
51:17might depend very much on how
51:19individually hypnotically suggestible
51:21somebody is and if we're trying to test
51:23thebo effects that's a critical
51:25individual variable that we might need
51:27to understand the
51:29data right any any yes there's another
51:32question here I should say there's a
51:33microphone that um will be brought to
51:36you that's so people online can can hear
51:38the question I an know excellent as as
51:42my brain predicted it would be um you
51:44didn't mention either genes or Evolution
51:48no deliberately Adam very deliberately
51:51well they're the two things that I
51:52obsess about and so I wonder if you
51:54could I I know we don't know the answer
51:56to this but I'm always looking for
51:58mechanistic explanations of the
51:59emergence of the stuff that you're
52:01talking about and I tell us what your
52:05views are on how why and and what the we
52:09talk sometimes about the neural
52:10correlates of of Consciousness but
52:13underlying neurons are proteins and
52:16genes and so there is a mechanism under
52:18there somewhere yeah I mean there's an
52:21awful lot to say and I won I won't um
52:23expand too long but certainly all all of
52:27I think there's an
52:28evolutionary perspective on pretty much
52:30everything I've been saying in terms of
52:34how so why the brain is a prediction
52:36machine in the first place I've argued
52:39is all about controlling regulating
52:40physiology and of course that is
52:42something that's selected for and then
52:44the way we experience the world the the
52:46characteristics of our particular
52:49phenomenology can also be understood you
52:51know in in evolutionary terms as an
52:53adaptation to its particular Niche um
52:57so it seems to me very evident that
53:00Consciousness evolved like all
53:02biological functions it's very
53:04phenomenology is would be very it would
53:06be very strange to suggest that it
53:07didn't because it's so obviously useful
53:09for us um so I think that's fine now
53:13whether there's a specific genetic basis
53:17to understanding some of these things I
53:19I that I think is is only beginning to
53:22be uncovered through techniques like
53:24optogenetics which can begin to do the
53:26very fine grain manipulations that would
53:28be that would be necessary I mean
53:30there's some work looking for instance
53:32at the genetic uh contribution to things
53:35like synesthesia so we can see there are
53:38genetic differences between different
53:39kinds between different kinds of
53:41experience but also I want to go even
53:43deeper and think about the role of
53:45metabolism I mean that's often the
53:47missing part of the story when it comes
53:50to uh getting underneath the neurons now
53:53I I completely agree that neurons a very
53:55useful level of descript ion but to stop
53:57there is to caricature the brain there's
54:00much more going on um I um and and at
54:03deeper levels
54:05too yes there are two questions there
54:08three at least at the back so let's
54:11start with the one at the very back I
54:12think that hand came up
54:16first wonderful talk thank you very much
54:19um you have recently published a paper
54:21about hybrid predictive processing uh
54:23where you argue that not only top- down
54:25predictions exist but also the bottom up
54:28predictions and I wonder how this
54:30updates your beliefs on how
54:32Consciousness arises are these bottom up
54:35uh predictions merely like euristic best
54:38guesses um that provide new seeds for a
54:41generative model or um iterative
54:44inference or is it um like this
54:46interplay this um constant interplay
54:49between um bottom up and top down
54:51predictions is this actually necessary
54:53for Consciousness to
54:55arise um okay thank you thank you well
54:57firstly thank you for for reading that
54:59paper it's a recent paper with um my
55:01colleague Chris Buckley and Alex chance
55:02and Baron millich and just just to
55:05encapsulate very quickly for those I'm
55:08sure everyone's read it but just in case
55:09you haven't case you haven't it's sort
55:11of I think it it highlights both the
55:13strength and the weaknesses of this
55:15predictive processing framework because
55:17um in this paper we argue that it's that
55:21predictions instead of only going top
55:23down and prediction errors going Bottom
55:24up both can flow in both directions
55:27which kind of on the face of it means
55:29that everything is on the table it
55:30becomes a very flexible framework that
55:33maybe you can just account for anything
55:35that that's a weakness of it if you can
55:38do anything with a framework but it's
55:39actually a strength because you can make
55:41it specific and it means the resources
55:43of predictive processing are nuanced and
55:46are capable of I think explaining the
55:48nuances of Consciousness so the idea
55:51here is that um came from machine
55:54learning actually that in machine
55:56learning if uh a system that's learning
55:59is encountering the same kind of
56:01situation over and over and over again
56:03it doesn't need to keep going through
56:05these repeated cycles of predicting and
56:07updating predictions and converging on
56:09on a best guess it can just learn what
56:11that best guess is that's called
56:14amortized inference it learns a mapping
56:16from the input to the to the basian
56:19posterior and then iterative inference
56:21is when you go through this this
56:23exchange which is more time consuming
56:25more computationally expens ensive but
56:27is more flexible because it can apply to
56:29novel situations so our architecture
56:32naturally trays off these things against
56:35each other so it learns when possible
56:37and uses the the sort of fast quick it's
56:40like um Danny carman's book thinking
56:42fast and slow so we call it inferring
56:44fast and slow so there's like a fast
56:46route when things are stable and a slow
56:48route when things aren't so it's not
56:50that I think uh so how does this the way
56:53I think this relates to Consciousness is
56:54not that one is necessary and the other
56:56isn't or one's conscious and the other
56:58isn't but I think they may account for
57:00different aspects of Consciousness when
57:02we walk into a room say you just walk
57:03into this room now you immediately have
57:05an impression there's a lot of people
57:07here it's a you people are sitting on
57:09chairs there's various portraits of of
57:12of men on the walls and
57:15um and that's the gist and then you go
57:17through this more detailed iterative
57:19inference process and figure out what's
57:21actually there oh they really are all
57:22men on the walls and and now and so I
57:26think our our phenomenology may actually
57:29be understandable through this kind of
57:32trade-off we haven't so that's kind
57:33those are the experiments actually want
57:34to do next now we've got the formal
57:36framework to see how those relate to
57:38different ways of experiencing things
57:40the society is trying to change the
57:43image that we sent sorry I'm not there
57:45are quite a lot of women portraits but
57:46not in this room yet but it will it will
57:49happen so just the last two questions
57:51from back there please and then
57:53unfortunately we're going to have to
57:54stop um thank you very for the for the
57:56talk um I wanted to ask your thoughts on
57:58on a missing piece of the impossibility
58:01or not of a test to probe if something
58:03else is conscious because it feels like
58:05it's something we attribute only but
58:06we're still unable to fully
58:09confirm thank you we in fact with a a
58:12bunch of colleagues from my Canadian um
58:14affiliation we just had something on on
58:16exactly this question on tests for
58:17Consciousness um in transends cognitive
58:19Sciences not that we solve the problem
58:21you're right it's a very very difficult
58:23problem now we we want a test that tells
58:25us do whether something is conscious
58:28whether that's a human patient with
58:30brain damage whether it's a non-human
58:31animal whether it's a newborn infant
58:33whether it's an octopus whether it's a
58:35machine and we don't have we don't have
58:37these tests um I was there's one slide I
58:40was going to put in uh in sort of
58:42tribute to to Murray and Adam and Alex
58:45Garland on the Garland test which is
58:48which is now part of the scientific
58:49vacular to it makes the point that in
58:52the film X
58:54MAA the test about whether the the the
58:57machine the robot AA is conscious is
58:59really a test of what the person
59:02attributes it's a test of what it would
59:03take for the person to believe that
59:05something is conscious rather than a
59:06test of whether the thing is itself
59:09conscious in the same way that the
59:10touring test is a test of what it would
59:12take for a human to believe that the
59:14machine is intelligent rather than
59:15saying the machine is actually
59:17intelligent um but that's what we want
59:19we want these forward tests that tell us
59:21whether the system has the property not
59:24whether what it takes for us to think
59:26that it has the the the property um so
59:29all I can say about this at the moment
59:31is that it's helpful to consider the
59:34different cases Al together because we
59:36have very different priors like when it
59:39comes to non-human animals we share a
59:40biology so a whole bunch of uncertainty
59:43goes away about whether the stuff we're
59:45made of matters um and that when we're
59:49thinking about Consciousness in
59:51AI that matters that much that's going
59:53to reduce our Credence in whether the
59:55system is conscious purely because
59:57there's something else that's changed
59:58from The Benchmark case where we know
1:00:00Consciousness um exists so my colleague
1:00:03the philosopher Tim Bane talks about
1:00:05this so-called iterative natural kind
1:00:07strategy where we assume that
1:00:09Consciousness is a natural kind it's a
1:00:11collection of properties that that
1:00:12cluster together in the natural world
1:00:15and that by iteratively moving out from
1:00:18cases where we can be more certain
1:00:19perhaps humans in different brain
1:00:22conditions to start with will be able to
1:00:25in a way extrapolate uh to more
1:00:28challenging cases but I think the real
1:00:30problem we'll never know for certain
1:00:32we'll never get
1:00:34100% thank you so I think we come to
1:00:37almost to the end of the question one
1:00:38I'm sorry I shouldn't be abusing my
1:00:41privilege but I've been baffling baffled
1:00:43by this from the beginning so in the
1:00:45light of everything you said about
1:00:46artificial intelligence the response
1:00:49about sales and so on why do you refer
1:00:52why do you use the term a prediction
1:00:55machine I don't understand the
1:00:56prediction part where does the machine
1:00:58come in and what are we supposed to take
1:01:01from your use of the word machine when
1:01:03you pretty much told us that
1:01:04Consciousness is exactly the opposite of
1:01:07a machine well it depends what you mean
1:01:08by a machine yes well that's what I'm
1:01:10asking what do you mean by a machine I'm
1:01:12I'm pretty liberal I think about about
1:01:15about machine that no I think again
1:01:17without delving too deep into it there's
1:01:19there's a sort of machine that I don't
1:01:21mean which I think it's it is useful to
1:01:24qualify there's a kind of of machine
1:01:26that takes stuff in and produces other
1:01:29stuff like a factory yes that that's
1:01:31that you think of that that's called an
1:01:33Ali poetic um system it takes stuff in
1:01:36it makes other stuff by you know
1:01:38whatever mechanism is
1:01:40inside biological systems are very
1:01:42different they don't take stuff in and
1:01:44produce other things they produce
1:01:45themselves they're what Francisco
1:01:47verella HTO mataran called autopoetic
1:01:49systems I still think is a kind of
1:01:52machine in in the in the most liberal
1:01:54sense that there's a mechanism at work
1:01:56but it's a special kind of mechanism
1:01:58that generates its own components it
1:02:00generates its own existence um if
1:02:03there's a better word I'm I'm open to it
1:02:05but but I qu you know I like I still
1:02:07find it appealing um because it it it
1:02:12means there's no there's it lessens a
1:02:14temptation to attribute anything
1:02:17Supernatural yes but it is mechanistic
1:02:19anyway so let me let me just now thank
1:02:22Anil for I'm sure you will all agree
1:02:24it's been a really memorable location a
1:02:27wonderful lecture en lightning and I
1:02:30think the perfect example of what the
1:02:32farad CRI is all about it is about
1:02:35somebody who has contributed to this
1:02:36discipline at the highest possible level
1:02:39but somebody who has this unique skill
1:02:42and talent to convey to us the nonp
1:02:45specialist the W of his subject and it
1:02:48can hardly be anything more wonderful
1:02:50than human or animal Consciousness and
1:02:53thank you very much we're not done yet
1:02:56and there's still something to hand over
1:02:58to to Anil so um Anil I have something
1:03:02here for you and um this is the medal
1:03:07and the and the prize on behalf of the
1:03:08raw Society it's an tremendous pleasure
1:03:11and honor to hand you the
1:03:142023 uh farad farad medal and on
1:03:22[Music]
1:03:24r
1:03:46[Music]
1:03:51thank you very much and wonderful to see
1:03:53such a enthusiastic audience thank you

About this transcript

This page contains the full transcript of Consciousness in humans and other things with Anil K Seth | The Royal Society by The Royal Society, generated from the public captions YouTube serves with the video. The transcript has 11,010 words across 1,662 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.