How the brain shapes reality - with Andy Clark

[Applause] would you know if a 15 cm nail penetrated your foot in 1995 a construction worker fell off some Scaffolding in New York City found that he'd Fallen onto a 15 cm nail which had pierced the work boot entirely the construction worker was in great pain was given pentanol and madzelan for the pain was taken by the paramedics to hospital where it turned out that the nail had passed cleanly between the toes the boot the boot itself was severely injured but the toes were entirely intact and yet the construction worker had been in in genuine great pain needed the fenil in order to have the nail gently removed what's going on so that's a dramatic case it's a little sort of anecdotal case in the British medical journal 1995 but what I want to do today is just to convince you that regardless of dramatic cases like that all of our experience Moment by moment is constructed in something like that way it's constructed by some weird mixture of what we expect to be experiencing like great pain because I've seen this great big nail go right through my workbook and what the sensory information the information from the NOS receptors in this case is actually delivering and I think everything that we experience is built in that way and in some of these sort of false perspective pictures you can see photographers playing around with your um with your brain's predictions about what's what so the picture I want to put on the table is about brains as prediction machines I think that's a fundamental thing that a brain is it's a prediction machine that is busy predicting all kinds of different stuff not just um what you see what you hear but what's going on in your body as well um and also of course what you might be doing in the near future but a lot of it about predicting the present Al that kind of sounds and we don't really realize the extent to which this is going on mostly because the predictions are normally unconscious loads you know you make a few conscious predictions we'll come back to those later but most of what really matters here is going on unconsciously and you have very little control over that um so here's a a demo um those of you that are sitting there can probably see this demo okay it's a so-called Hollow Mask illusion so this is an ordinary kind of um sort of like a joke shot mask I guess it's Einstein in this case and um this is the hollow side of the Mask now those of you sitting back there that sign's probably looking fairly convex to you probably looking a bit like a normal face but if I rotate carefully here we go yeah yeah yeah that's the um normal phase side and then we come around I'll try and do it for people on the side in a minute and then we come around and that's the con okay so now let me I'm just going to do this now let try to do this since we brought all this Hightech apparatus with um here we go for people over there that's the um that's the concave side there right and oh you're going to be difficult aren't you um how's that going to work hold on there you're getting there yeah right cool okay excellent if all that down now that was the difficult bit gone off without much of a hitch that's good um so what is going on there your brain is getting perfectly good when you're looking at the the concave side your brain is getting perfectly good information saying that that is a concave side um no lack of information being transmitted to the brain there on the other hand a bit of your brain is kind of saying that faces aren't like that I know a lot about faces faces are convex that's just what a face is particularly a famous face I've seen it lots of times um and that prediction of convexity is kind of trumping all that real sensory information that is coming in from Vision in this case um so in a way this is a good thing your brain is kind of trying to sort the signal out from the noise using what it knows about the world but sometimes it gets things wrong as a result as when you're looking at the concave side of the Mask or if you happen to be falling onto a 15 cm nail and there's a is that was in case it didn't work so you know I just's go through this bit now that was another version right so I think we will have all experienced something in this ballpark if you've ever experienced Phantom phone vibrations you probably experienced that feeling of the phone going off in your pocket even when it's actually not in your pocket or it's turned off or nobody's calling you um they're very common um I actually get Phantom wrist vibrations now thanks to the Apple watch my Phantom wrist vibrations are very clever now because if I start walking quickly I expect the watch to give me a double little ping saying hey you're exercising and I feel that even if I'm walking quickly I'm not wearing the watch it's really weird so um I think we do all um experience our predictive brains in action what we don't realize is how much of everything that we experience is like that that's mostly what I'm going to be talking about there's a recipe for Phantom phone vibrations um they're extremely common in people that are strongly expecting a call no big surprise there um medical interns for example have a very high incidence of phantom phone vibrations uh they're exacerbated by stress and caffeine and both of these mess around with dopaminergic systems that were returned to later um because dopamine is a big player in the predictive brain and and other neurotransmitters well come to that but there's a little recipe for getting Phantom phone vibrations going do one more demo if it works this one is um sine wave speech so what's going on in S Wave speech is there's an ordinary speech recording and then it is stripped of a lot of the usual um acoustic information so you get a kind of skeletal version um it sounds a little bit like the clangers anybody that remembers the clangers it's British thing if you're not British you know don't worry about this but um back in whenever it was there were these things on the moon and they kind of went kind of like that and that's what sway speech sounds like um when you hear it the first time what I what you'll experience here is very rapid learning on the part of the brain so I'll play you the sine wave sentence I'll play you the real sentence the non-strip down one then you'll hear the sine wave one again and all I want you to do is just appreciate the difference in your experience the huge difference to your phenomenology as we philosophers sometimes say that um a good prediction can make so here we go it was a sunny day and the children were going to the park real sentence coming up it was a sunny day and the children were going to the park same sign wave again it was a funny day the children were going to the park should sound very very different with it want two more the C was kept to the cage at the zoo the camel was kept in a cage at the zoo the camel was kept in a cage to the zoo come he was sitting at his desk in his office he was sitting at his desk in his office [Music] so if you listen to that to those kinds of recordings enough you become a native speaker of sine wave speech you can play me a brand new recording I'll hear it fine you know it's just oh yeah I know what that is um so when we hear speech sounds in our ordinary language we're performing the same kind of trick you hear spaces between the words well sometimes I'm not talking this quickly you hear spaces between the words that I'm saying but if you look at the actual acoustic signal it's pretty much continuous those spaces are inserted by the predictive brain because it's correctly predicting that there are word boundaries there so everything that we experience is built this way this was in case that demo didn't work it's like it's a visual version of the same thing see the what's going on now see that one again um so what this suggests to use a phrase that comes up in this literature is that perception is a kind of controlled hallucination the brain is continuously trying to guess what's most likely to be coming at it from both interception and extracephalic forecast made a little difference to the actual weather in that region that's a weird kind of world and let's not even think about what goes on in the Border Lines between regions and things but you know a confident forecast of rain made rain itself just a little bit more likely a confident forecast of sun made Sun itself just a little bit more likely we don't live unfortunately in that kind of weather world but I think that this is the experiential world we live in our confident forecasts tweak our own experience in the direction of the forecast now you know people have said things like this for a long time um there's stuff like this in Emanuel K kind of constructivist picture of perception and experience it's really super clear in Herman Von helmholtz uh the idea that perception is the product of an unconscious inference and in the 20th century um Rick niser Richard Gregory and others the analysis by synthesis movement um had the same sort of picture and here's Richard Gregory um delivering the Christmas lectures I think here in 1979 so I feel very um as if I should have more bangs and whistles that was all I had yeah um so what's new now well since about 2005 what's new I think is a comprehensive neuroscientific theory of all these effects and it's actually a theory that deals not just with perception but with action too tonight's not long enough for me to give you the account of action but it it will be important in discussion I imagine um so maybe you know one of the pioneers of this is Carl friston at UCL here in London 2005 published theory of cortical responses which was beginning the hard work of of turning these sort of um intuitions into a story that was both computationally clear and also so neurophysiologically plausible that's sort of where we are now with it I think um so much so that lots of people are writing books about it me included I'm selling one The Experience machine um so there are lots of good books about this out there um my colleague annel Seth has one an early one by yakob houie the predictive mind wonderful book Lisa Felman Barrett's book on how emotions are made a big stress there on inceptive prediction predicting the states body and the importance of that my own earlier book and the current sort of Bible the active inference Bible because predict I'll call it predictive processing lots of other people call it active inference which I find a bit clumsy but it's the same thing so what I'm going to do is introduce as it says a sketch of a part of the theory and then look at some applications in computational Psychiatry a particular interest of mine what might these accounts say about um both neurotypical and atypical forms of experience and then look at some puzzles and challenges towards the end so oh what did I do there the more traditional view had perception of Sonic that works from the outside in we get this in daycard all those beautiful drawings of energies hitting the eyes impressing things like little tubes almost like little wax Impressions going deeper and deeper into the brain eventually something cool happens with the p gland we won't worry about that um but it was all outside in perception was the world coming in from the outside and making a richer richer impression on the brain and this picture came through in 20th century Neuroscience where most textbooks leading textbook when it's talking about Vision has all of the arrows moving um inwards from the sensory peripheries deeper into the brain in artificial intelligence David Mah had a basically feed forward processing picture of um Vision super influential wonderful work uh but predictive processing turns a lot of it upside down most of the brain's work is being done from the inside out that huge metabolic budget that we'll come back to soon is mostly in the business of keeping your model alive and using that model to make predictions just like you know a weather forecast needs a weather model to make predictions your brain needs a model of the world to make predictions so yeah brains like that uh predict in the present the consequence of that that is a very efficient and interesting thing is that in order to get to grips with the world all the brain needs to do is deal with the errors in its own predictions so if I'm busy predicting a certain kind of visual scene and I'm getting information from the world that is consistent with my prediction I don't need to do anything this is all as expected but if I get prediction errors stuff that that I've been unable to predict then something ought to happen ideally so prediction error signals are are a sort of anti-hero almost of these accounts they carry the news they are the sensory information that is currently unexplained so they are a form of sensory information they're just that particular bit of it that is currently unexplained by the brain's best ongoing predictions um they carry the news and what do you use them for you use them to select better predictions so if I'm going to see something I didn't expect to be there then I better get a flurry of prediction error um use that prediction error to recruit a better bit of the model a bit of the model that has a plastic yellow duck or something in it so that I can get to grips with the Unexplained bits of sensory information uh oh and that's sort of how these models actually sort of look if you're um schematically drawing them thing to take away from that is just that these are multi-level models the brain's not just kind of sitting there making one prediction and I don't know um uh what shall I have uh computer laptop on the table instead high levels are making predictions about I know something's going on low levels are making predictions about the precise details of the arrangement of stuff on the screen and each intermediate layer is trying to predict the layer below it so you know just like you might use information stuff you know about sentences to predict stuff about words and stuff you know about words to predict stuff about letters and stuff you know about letters to predict the distribution of Ink on a page so it be a bit like that so it's super efficient you use the active World model to make an educated guess and if the guess doesn't accommodate trusted and we'll come back to that in a minute bits of sensory information prediction eror is your friend and it ought to improve your guessing it's super efficient it's used in commercial um motion compressed video for example that goes back to 1959 where people realized that you don't need to send the information in the next frame of a video if it's more or less the same as a frame before it all you need to do is send the difference so if someone's running against a static background just send the information about the difference in where their legs are going let the background stay the same suppose you also knew their gate just how they were running then you don't even need to um update where their legs are going because you predicted that too so you can get as clever as you want about this the clever you are the more efficient you are and I think this helps us make sense of what's a really really puzzling fact about the brain the neuronal connections in the brain that run as they sometimes say top down from deep in the brain to towards the sensory peripheries they greatly outnumber the ones carrying information inwards that's a bit of a surprise on the perception is a big thing we're doing it all the time and it's all from the outside in it's not it's mostly from the inside and out and prediction errors just kind of tidy up around the edges the picture here so for example at least 90% of the input to lateral genicular nucleus which is just after the retina comes from deeper in the brain um on average throughout cortex the these outflowing Pathways outnumber the inflowing pathways by at least two to one and in some areas by four to one so it's a the brain is Wired from the inside out and these kinds of accounts make a lot of sense of that wiring so the anomal is resolved the hard work's being done from the inside out and hey no wonder the brain is spending an awful lot of energy all the time even when it's at rest because it's maintaining the world model it's maintaining a model that can be used to make these very detailed predictions it's as if you're constantly running a simulation of reality in your brain well you are running a simulation of reality in fact you're experiencing one right now and if I do something unexpected I'll just give it a few prediction errors I know we go go so prediction error anchors the brain's imaginin if you like the brain's constructions to the world but this raises a puzzle question if the brain can do this if it really can use FL as a prediction error to rapidly correct its own guessing why do we some get things wrong you know what happened with the worker that fell on the 15 cm nail you know there was no no receptive pain information coming from the toe where they were feeling the pain um to the brain why didn't the brain then immediately get prediction errors and correct its guessing indeed so this brings us to what I think is arguably the most important bit of the Contemporary picture it's something that wasn't there in the earlier versions of this story um what it is is the idea that the the balance between prediction and sensory information is never fixed it's something that the brain itself has to estimate Moment by moment so your brain is estimating how much to um take account of certain predictions how much to take account of certain bits of apparently sensory counter evidence you know the case of the Hollow Mask your brain's ignoring a lot of sensory counter evidence because it's really confident in its predictions which of course again um is interesting from the perspective of you because you know it's a hollow M but you're still seeing it the other way around because those intermediate levels of processing are just absolutely convinced of the convexity that go on so it's a kind of a kind of weight in game the brain is estimating how much confidence it has in certain predictions or certain bits of sensory information this is known as Precision weighting um and the Precision waiting game is kind of a zero sum game so if I up the Precision on uh my predictions I am lowering the Precision on the bits of sensory evidence that are meeting those predictions and vice versa and this is happening at every level of processing and if these accounts are right in every area of the brain they're probably not all processing prct and prediction errors in quite the same way the amigdala has got its own special thing to do um but that's kind of what they're all doing if these accounts are on track we'll come back to that question um yeah so on a foggy day for example you might not want to rely too much on what your eyes are telling you particularly if you know the area you've got a great model of the world it's like I'm going to rely on my model and just um you know I'll take some account of the visual information but not too much so in the brain the idea is is that this Precision weight in economy centers on the complex neurotransmitter system so uh all of the things there dopamine noradrenaline atic acetal Coline um Garba glutamate they're all there somewhere serotonin um so that's a big bit of the Precision weight in economy is being done by complex neurotransmitter systems it's also being done probably on faster definitely on faster time scales by the neuronal phase by the temporal phase of neuronal oscillations low versus Fast frequencies of neuronal oscillation will come back to those towards the end if there's time so Precision increases the influence of particular predictions or particular bits of sensory information it's kind of what attention is if these accounts are right then what attention really is is uh variable Precision weight in going on in the brain which I think is kind of useful we all use this word attention no one really knows exactly what it is there a proposal here that's what attention is it's exactly that um this is a Mooney image it's a bit like the uh woman with the horse earlier I if you can see this one but here's the actual image and here's the Mooney again right now I think that what we can take away from these things is that predictive brains have a very very delicate job to do they have to trade a very fine line if you're going to spot the hidden frog you've got to up the Precision on an awful lot of bits of information about froggy Contours in particular and you downgrade other bits of information that's kind of what's going on when you allow the when you the Frog can I was going to say jump out at you yeah exactly um and I think it's because we can do this delicate Precision weighting that predictive brains are good Detectors of faint patterns in world full of noise so it's important to get these weightings right and if they go wrong very bad things will happen that's what that's trying to say you know if your brain starts to misestimate the value of certain predictions or to overestimate the value of certain bits of sensory information then bad things will happen and we'll look at a few of those bad things in the next um half hour um at this point though you might also be asking well can't it just get the balance right you know um that's actually a very tough question we may come back to it in discussion because I would say there's no right balance there is no balance such that that balance will always reveal how things are because it all depends on the kind of world you happen to be living in at that time how volatile it is what's going on there so I don't think there's any right answer to how this Balancing Act can be done there are some ways it can um change so computational Psychiatry as it appears through the lens of this account is mostly a matter of trying to see how differences in that Precision weighted balance in act might generate different profiles of experience that's that's that's pretty well what's going on in an awful lot of work um I think it's a good way of bringing together a systematic understanding of a lot of um atypical experiences the wide range of typical experience and also what happens under the effects of psychedelic drugs for example um which we won't get to so for example suppose you assign a very high weight into the sensory evidence very very high weight into sensory evidence what's that going to do to you as a prediction machine one thing it's going to do is it's going to make it hard for you to detect faint patterns in a noisy environment to detect faint patterns in a noisy world you want your model of how things are to to be sort of I don't know taking a firm hand over the sensory evidence your model is what's going to give you the best grip on things probably so there's a lot of interesting work now suggesting that autism spectrum condition involves just this kind of overweighting of the sensory evidence that um in fact it's sort of it's very much enhanced sensory evidence is the the kind of profile is emerging from quite a lot of detailed studies here and there's a very interesting side debate that I won't go into but it's a side debate about whether what's going on is weakened influence of the models or heightened influence of the sensory information from a basian perspective um these are going to look very very similar but they won't look quite the same and the evidence is now strongly suggesting that autism spectrum condition is not about weakened model use it's actually about enhanced sensory information um so what would happen if you're enhancing sensory information like that well a lot of environments will seem difficult to negotiate you may become wary of noisier hardto predict environments social worlds where lots of very small cues really matter might be especially challenging um so again I don't one thing that comes out of these accounts is it's if there is a sort of social understanding um difficulty in autism spectrum Condition it's a side effect of the real thing which is enhanced sensory information it's not a fundamental thing I think that's quite interesting um it also makes sense of the uh cases where people with Autism Spectrum condition just do better so stuff like uh the embedded figures task where you're showing something and ask to find the geometric figure in the thing there the rocking horsey thing whatever it is um people with Autism Spectrum condition do better at that than neurotypicals probably because they're better at allowing the sensory information to speak for itself you know they're not getting um wayed too much by that sort of um strong top down it's a whatever it is I don't know what that is actually rocking HSE thing so now imagine something different imagine that instead of assigning extra weight to the sensory information you assign extra weight to some of your own predictions what's going to happen then it's pretty obvious what's going to happen then at that point you really do start to hallucinate things you know you've taken some of the control out of controlled hallucination and You' just got hallucination um and you know I think we all do a bit of this this is why we see faces in in the clouds and things like that um because uh we can have strong predictions about faces and if we just allow a certain sort of relaxation to happen you can do this spot in so there are accounts of um delusions and hallucinations in schizophrenia and psychosis that are emerging at about this point where the idea is that there's a kind of overweighting of unconscious predictions and that that in turn might arise if your having false prediction errors if for some reason the brain is generating false prediction errors then you're going to have to come up with some sort of model to try to deal with this stuff that is kind of saying to you this is important you're not getting this this is important stuff revise your model so the model gets revised it becomes very firm and then it can alter experience pulling it in its own Direction just like with the Hollow Mask illusion yeah so they become highly weighted uh the these can be predictive models of complex conspiracies impersonations nowadays it's the Internet it's the internet is a source of a lot of these models um at that point a a strong prediction is altering your experience in a way that then seems to confirm the model that is making the predictions and that's something that I think is really important to get to grips with the the the tendency of predictive brains to get locked into spur cycles of self-confirmation where your own predictions are changing the way you sense the world and then you think oh I'm getting evidence for the model that I'm using to make these predictions but that evidence is furious I think we have effects like that in in daily life actually our own unconscious expectations alter the way that we see things and experience things and we think we're getting evidence for our own models so that's a kind kind of a killes heel I think of the predictive brain spuriously self-confirming predictions but it will also turn out to be something that we might be able to turn to our advantage I'll come back to that later checking the time yeah so as just one experimental demo of these sorts of self spuriously self-confirming Cycles little bit of work where um subjects were they had expectations created in them by teaching them to associate a particular geometric Q with higher or lower levels of heat as indicated by the little thermometer thing so in the induction phase as they called it expectations were put in place so that if you saw this geometric Q you're associating it with low heat if you see the other one you're associating it with high heat in phase two of the experiment real heat stimuli are applied to people's bodies following the geometric cues so now they're actually getting some heat but unbeknown to them the applied heat levels were all exactly the same so they get the geometric Q they already associate it with more or less heat um but they're actually getting exactly the same amount of heat now as you might expect subjects reported the expectation induced effects more or less pain in line with the induced expectations geometric cues and they supported this they didn't just ask them this question it was supported by something called the neurologic pain signature a slightly contentious um fmri profile for um something like real pain is going on rather than I'm just reporting what I think you want to hear me report so they they did that as well as a report so that explains I think why they didn't update this is what I think is interesting here you you you should be asking yourself but why don't you update you're given the same heat can't you just feel it and update your model but you can't because your model is making you feel more or less heat and so you think you're getting more evidence for your model you certainly don't think you're getting counter evidence for your model um so I think it's a nice example just a simple sort of controlled example of the way experience when it's shaped by prediction can get locked into to these sorts of spirous these self-confirming Cycles as they say expectations modify pain perception and this modified perception drives subsequent expectations so you maintain the misguided ones even when you've got good sensory evidence to the contrary so what I want to do now is to see how Cycles like this could be at work in stuff that's maybe a little bit closer to home chronic pain uh is the example that I'll start with here so um somewhere between a third and half of the population of the UK at least experienced chronic pain so it's a huge um it's a huge Health burden so there's a puzzle in chronic pain I mean chronic pain as you all know is kind of long defined as a sort of longer term version of pain it's not the acute pain that you get when you um fracture an ankle or something this is something that's been going on for a while so the match between the experience of chronic pain and identified standard causes is only high for acute and localized things like you know the the fractured ankle and so on if you move to longer term conditions the match is very bad and there's a lot of examples of this reported breathlessness in pulmonary disease discomfort with atrial fibrillation asthma cancer related fatigue lower back pain uh very very striking example and these are all cases where the amount of pain experienced just doesn't sit very well with the peripheral etiology if you like what seems to be um going on structurally in the relevant bits of the body and this isn't just across different patients this is across the same patient at different times as well so it's a a huge amount of variation for what's apparently pretty much the same level of sort of basic structural um bodily damage and it affects the same person at different times too I think this is that me no you got the same ringtone as me you're just panicking me here that's that's right so yes wide variation in experienced pain I think this falls into place really really well from a predictive process in perspective I mean what goes on over those longer time spans one thing that goes on is you You' got an awful lot of time to form expectations about how your experiences of pain are going to vary across different context um different life histories will also um do this differently so I think that we can expect different life histories to differently balance that sort of precision weighted Balancing Act and sort of random things that happen contextually like I don't know you experience an asthma attack in a particular kind of context that could make it more more likely to experience it in another context given the role of prediction in those events um so they can become highly sensitive to context and I think that can happen even when the context turns out not to actually be clinically relevant um the longer you live with a condition the longer you've got for things like that to start to happen it's not surprising um so this is just kind of of like this is kind of the same thing again about predictions changing experience and then you think that you're getting evidence for the model that you're using to make the predictions even when you're not I do think this is part of the puzzle of chronic pain and it's something that we can begin to address um let's get to that so no one there is saying that the chronic pain isn't as it were really experienced as pain of of course it is you know just like the um just like the person that fell on the nail was experiencing excruciating pain so so far it it will seem like bad news it just seems like you know predictive brains are good they're rather efficient but they're getting us into all sorts of nasty little Corners um what can we do about that so there's some good news with apologies to John cinski um I think there's a kind of opportunity here to push back against the bad predictions and install different predictions that might be be helpful they might even be spuriously helpful in a certain sort of way you know maybe it's a a trick in just the same sort of way so I'm going to end by looking at some of those interventions emerging chances to sort of hack the predictive brain in a good direction so there's an obvious example the careful use of placebos a very interesting area I think um it's known that the efficacy of placebos varies a lot it varies basically according into how confident the patient can become of relief from the placebo um you can get this by differently describing the potency of the thing that you're being given um you can also get it by doing Placebo surgery for example instead of just a placebo pill Placebo surgery is very effective um for certain things um mostly I think because we think surger is pretty serious you know I think that's a pretty serious intervention I'm expecting I'm expecting something from this um at least that's I think what the brain's thinking um so patients receiving Placebo surgery for osteoarthritis of the knee reports similar relief to those undergoing normal surgery and they get more relief Than People given placebos um pills or coaching so you know pretty good studies here athletes show improved performance when hooked up to what they think are pure oxygen Delivery Systems when in fact they're just air um they also show improved performance if they're given a drug that they're told will increase red blood red blood cell um amounts even though it doesn't yes a 1.5% increase which is big if you're a elite athlete um so there's obviously tempting ways to get around the doping laws here interesting to me that honest placebos work too um an honest Placebo is one where you're told this is a placebo you know there's no no clinically active standard clinically active ingredient here um but nonetheless a lot of people report significant amounts of relief I think that's just another um excuse me gone too far I think that's just another manifestation of the fact that most of the predictions are unconscious that you know if you're given something that comes out of a nice little sort of um Bubble wrapped Thing by someone authoritative in a white coat then even if you know it's a placebo mid levels of your neuro processing don't it's like oh yeah this is this is going on um so another sort of systematic kind of push back here is something called pain reprocessing Theory it's very interesting and promising I think so this to get into this sort of bit of the the the account we need to think that the basic experience of pain is kind of it has a sort of role to play and that role is telling us that our bodies are in imminent danger that if we basically if we keep on doing what we're doing this is not going to be good um and that's true if you've just fractured your ankle you don't want to you know keep walking on that right now it's very Cy borgy ankle there know what's going on there um but in a lot of these cases of chronic long-term Pain and Disability it seems like the prediction of imminent bodily damage has become the problem if you like what's got locked in is an expectation that if you continue doing stuff it's really going to be bad for you that's a big component of what pain is on these Accounts at least is that very expectation so as people do say in this literature your experience of pain there's a little bit like a malfunctioning warning light in a car it's a bit like you know that light's going on it's saying you better pull into the side of the road right now don't keep doing this um looks like you've got to stop what's going on but suppose it's a warning light that's SM functioning um and of course what's interesting here is what if you think of a warning like malfunction you probably think oh there's a bit of wiring that's gone wrong but in the predictive processing accounts of warning light malfunctions there's not even a wiring problem it's just a kind of it's just a sort of balance in act problem there's nothing structurally wrong with the the wiring of the brain um so I think it's quite it's an interesting sort of software problem if you like so chronic back pain case I mentioned earlier in 85% of cases there's no standard peripheral cause um that's not say that there's no cause it's just not a cause that matches up to the to the pain um what pain reprocess in theory is trying to do is to reframe the pain in a way that pushes back against those predictions basically involves um doing a talk a bit like this I suppose um so saying that Chron IC pain you know might be that kind of false alarm um involving locked in predictions but then trying to install different predictions mostly by getting people to do a little bit more despite the pain and oddly enough doing more makes experiential pain become less so it's a little bit as if just by driving the car a bit more you could make the warning light get a little bit dimmer which is I think that's the trick that we want there's a a very nice um document called this might hurt on that that I think you can probably still get on on Netflix or prime so this will be a helpful cycle this I think is the flip side of all those unhelpful Cycles where expectations tweak experience in in kind of bad and self- entrenching ways this is a way of tweak an experience that might be helpful and become entrenched in a a good way so the best systematic study here is Asher Al 2022 and 2023 study of chronic back pain report and neuroimaging as mentioned earlier and basically they just found that it works um large and sustained reductions in Pain and Disability following PRT as compared with both Placebo and usual care in about 73% of cases so it's promising for you know for chronic back pain a particular kind of pain okay how am I doing for time oh this is interesting I may just have time to do I wanted to do imagine let's see um so another example is functional neurological disorders sometimes called psychosomatic or psychogenic those terms not so much in use now um functional neurological disorders neurological symptoms without standard structural causes so you know chronic pain actually is a is an fnd by uh by this definition at least 16% of new referrals to neurology clinics are Di diagnosed in the end with some form of functional neurological disorder meaning basically um that you can't find anything um so examples included unexplained cases of blindness deafness pain fatigue weakness abnormal gate trema seizure um functional problems are normally diagnosed by variabilities of impairments so typically typically stuff like um um distracting a ention from a particular bodily part can reveal intact abilities in that in that part this I think Hoover sign works that way I I won't have time to go into the these examples but one example we can do this is an example shared with me to With Me by John Stone ex colleague up in Edinburgh this is m a who had a slow but eventually total loss of vision she'd had severe migraines before but she'd had no other history of stress trauma abuse head injury an anxiety none of uh you know none of nothing really just severe migraines fmri was normal um what stone noticed though was that when she was talking to him about her total blindness she was copying his bodily movements and would sometimes follow his gaze all the while absolutely not experiencing anything visually not consciously um but she could do Gaye following a movement copy and that's a rather strong cue clue that some there's something more intact there than you might imagine her eyes responded normally to an optokinetic drum so that's just a thing that um it kind of extracts an a stigmus response from the eyes and so you they kind of twitch a bit in um in response to the change in pattern so she exhibited the normal pattern of response with the optokinetic drum ston and colleagues conjectured that the severe migraines had basically made her brain start to to constantly predict Darkness you know she spent a lot of time retreating from the migraines into darken places she began to they thought um just expect Darkness her experience then had to fit in with that expectation be a little bit like your experience of the Hollow Mask or Phantom phone vibrations you know all of these things that we all experience just a rather dramatic version so the treatment there was to try and push back a bit just like pain reprocessing really against those hidden predictions um one thing that stone and colleagues did was just shared the science of predictive brains and how predictions locked in predictions can affect experience um also um when I say shared videos shared videos with people she trusted because she couldn't see the videos you know she's still still um functionally blind um also induced experiences of flashes of light using trans magnetic cranial stimulation so you know know the big the big thing that looks like a a wind up mechanical toy thing that applies a big uh a big pulse of um of magnetism to the to the brain that can induce phosphines and so that's kind of proven to her brain that's the idea it's proving to the brain that seeing is possible so she made a full recovery of course you don't know whether those things were responsible for the recovery you never know might have just recovered anyway but she did recover um and the main message today is that these are just extreme examples of the way that we're all constructing our sensory experience all the time functional neurological disorders are just a just a dramatic kind of um manifestation of that thing that we're all doing all the time and that's true even when there's a standard cause when you know you've got illness Burns other kinds of um injury so I think the more we appreciate this the less resistance there ought to be to the diagnosis of functional neurological disorder so one thing that happens right now is patients don't want to hear that they they really do not want to be told that they've got a functional neurological disorder and I think that's because we've got the wrong metaphysical model of um what it's like to be a normal human being to be a normal human being is to build experience in just this way so functional neurological disorder suggest a an extreme Exemplar hopefully getting these sorts of accounts on the table might help us see past old dualisms of Mind and Body might even erode the distinction between psychiatric and neurological disorders altoe you know personally I don't think it's a very helpful distinction so more generally lots of other promising ways to hack and nudge our predictive brains um immersive virtual reality therapy for pain this is very effective it's used sometimes as an anesthetic during Dentistry there are good good results when um changing the dressings of burns patients so this will be roughly your in a sort of immersive VR seeing an undulating jellyfish or something kind of nice and soothing like that um and I think what these accounts give you is this not just distraction you might think oh it's just distraction and it is in a way but now we understand what distraction is doing just like verbal reframing another example it's a way of altering those precision weightings and Seed in different predictions and all of this is going to change your experience verbal reframing also very powerful so you know I get little tingles in my hands before I having plenty of them behind that door um and you know I've tried to reframe that as chemical Readiness to deliver a good performance rather than hey it's all going to crash and burn it does seem does seem to help um so all of this stuff I think falls into place self-affirmation verbal reframing the therap apeutic use of Music touch and ritual um I think that we've got a systematic way of understanding that these are genuine and potent interventions um they're all nudging predictive brains meditation is another one I'm very interested in um my ex postdoc Mark Miller has done a lot of work on this what he thinks meditation is doing is giving you greater control over the Precision weighting mechanism um and you can kind to see how that might work you know I can kind of enhance the sensory experience so that I sort of um reduce the grip of models I can uh I can allow things to to drift or to become very precisely fixed if we could get control over our own Precision weighting mechanisms we would have what my colleague Al and Deans calls phenomenological control which would be an amazing thing to have you would you know you would be able to tweak your own experience in the ways that are best for you and your life right 8 minutes to say what's missing that should that should be enough shouldn't it um so it's a very incomplete picture altering the brain's best predictions isn't going to heal a fracture it's not going to cure a cancer it's not going to kill a virus so um I don't think we know what the scope and limits of these effects are yet still to be determined clearly good for stuff like pain anxiety breathlessness fatigue Sports Performance cancer related fatigue so you know lots of clear applications and unknown boundaries there are big individual differences in how effective some of these therapies are um therapies like Placebo hypnotism um uh pain reprocessing Theory and there's some very very interesting uh clues that there might be a genetic difference underlying that so there's an enzyme called CT and and CT controls the metabolism of dopamine in the brain so the more CT you've got the more um you're kind of eating up dopamin uh you can kind of see what might go on here if you don't eat up so much dopamine then if dopamin is a big player in the Precision waiting game you may have more phenomenological control and that is in line with the results here which is that um which is that people with high levels of genetically determine CT are less susceptible to um these things like placebos hypnotism PRT so I think that's very interesting sort of bringing this stuff right down to the molecular level and then all the way back up again to the experiential level um a big challenge is working out what the contents of the neuronal predictions are in different brain areas you know that's really not that well known um there's nice work by the L mle lab up in Glasgow and the title of his paper there says what it found scene representations conveyed by cortical feedback to early Vision can be described by line drawings so the idea there is that it's kind of line drawing level information being fed right back down but he's got beautiful experiments um suggesting that nice work on the amydala recently we need to uncover the detailed neuronal circuitry of prediction and error exchange there's a lot of very cool some wet Neuroscience work finding explicit prediction error responses in layers two and three of mouse cortex and that's got a nice sort of recent twist I'm not going to mention the time um there's nice work on the role of Alpha Beta versus gamma frequencies in communicating predictions and errors with the idea that the fast gamma frequencies are carrying error through the system and the slower Alpha Beta frequencies are carrying predictions down the system this has even LED to a suggestion for a particular implementation of these accounts which is different to the standard one and I won't go through that difference because time will trip me up and I won't get to the last bit of metaphysics if I do um but the idea there is just that um is just that you have the same neuronal circuits doing prediction error but differently prepared by Alpha Beta rhythms which allow gamma through where they're not prepared that didn't make sense to come back to that in discussion okay um I think at any rate this is a good moment in the history of this account because we've got some competing stories about what the implementation is in the brain and it's only when you settle on an implementation that you can really begin to test these stories properly so you can play pin the um pin the Tale on the cian cycle here where are these accounts in the cycle of pre- science normal science model drift model crisis um I'm not going to give away where I think we are in that cycle but certainly uh having competing implementation level proposals is a really important moment um assessment of the evidencebased don't worry about that um conceptual challenges is prediction really the most fundamental thing that brains are doing um some people think that if it was we'd all just crawl into a dark corner and stay there making very um successful predictions there's a good way out of that dark corner but I'd have to give you the account of action but there'll be time in discussion maybe that's sometimes called the dark and room puzzle what is the relationship between all those mid-level unconscious predictions and the top level conscious predictions no nobody knows um how does language fit into the predictive process in picture nobody knows um what about culture and the human built environment we we build great worlds to think in we build worlds that allow us to minimize different kinds of error to get our jobs done like painting lines on roads or building Large Hadron colliders um and we have a nice big project on this at the moment uh that takes me out to Spain sometimes to look at dolmans in galsia so why not archaeologists Vision scientists computational modelers and cognitive philosophers like me working together to explore material culture and how the worlds we build change the jobs that predictive brains need to do and how do these things really work together we're going to have to worry a lot about this in the near future because productive brains are now sharing their worlds with AI based forms of predictive intelligence we need to understand these new um ecosystems communities of mutually predicting agents um for some speculations on that see work by the versus AI group Carl friston the person that did a lot of the work on active inflence is the chief scientific ific officer of that group so keep an eye on them um oh yeah the hard problem why and how is conscious experience possible at all you might have noticed the book's called The Experience machine but I didn't mention the hard problem um I think there are big Clues I go through them in the book big Clues on that involve the account of interception the body predicting its own bodily States and the account of action action is a way of resolving prediction errors by changing the world in to make some predictions come true so once you get all those bits together I think we nibble at the edges of the hard problem so I think there's real progress I tried to tell the sort of upbeat story in the experience machine I think we're closing in on a principled multi level science of at the very least the structure and variety of human experience we got a small set of factors predictions precisions prediction errors and they combine to yield these striking differences in experience so I think that we're heading for a periodic table of experiential variation one that captures neurotypical atypical and Altered States but at the very least I think that this brings us a few steps closer to appreciating both the diversity and the um continuity of Human Experience thank [Applause] you

Channel: The Royal Institution

Views: 158,708

Keywords: Ri, Royal Institution, royal institute, andy clark, andy clark predictive processing

Id: A1Ghrd7NBtk

Length: 59min 50sec (3590 seconds)

Published: Tue Mar 05 2024