How the brain shapes reality - with Andy Clark

Would you know if a 6-inch nail penetrated your foot? In 1995, a construction worker fell off a scaffold in New York City and discovered he had fallen on a 6-inch nail that had completely pierced his work boot. construction worker was suffering a lot? The paramedics brought him pentanol and madzelan for the pain, where it turned out that the nail had passed cleanly between his fingers. The boot, the boot itself, was badly injured, but the toes were completely intact, and yet the construction worker had been inside. A great genuine pain needed the phenyl to be able to gently remove the nail.

What is happening is a dramatic case. It's kind of an anecdotal case in the British medical journal from 1995, but what I want to do today is just convince you that, regardless of In dramatic cases like that, our entire moment-to-moment experience is constructed that way, it's constructed by a strange mix of what we expect to experience as great pain because I've seen this big nail go through my workbook and what sensory information, NOS receptor information in this case, is actually being delivered and I think all that we experience is constructed that way and in some of these types of false perspective images you can see the photographers playing with your um with your

brain

's predictions. about what is what, so the image I want to put on the table is about

brain

s as prediction machines.

I think it's a fundamental thing that a brain is a prediction machine that's busy predicting all kinds of different things, not just what you see, what you see. You listen, but also what's going on in your body, um, and also, of course, what you might be doing in the near future, but a lot of that is about predicting the present. Al, that kind of sounds and we don't really realize to what extent this This happens mainly because predictions are normally unconscious uploads, you know, you make some conscious predictions, we'll come back to them later, but most of what really matters here it happens unconsciously and you have very little control over it so here's a demo um those of you who are sitting there can probably see this demo okay it's the so called Hollow Mask illusion so this is kind of um ordinary, kind of like a joke 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 probably looks pretty convex to you, it probably looks a little bit like a normal face, but if I turn carefully here we go, yeah, yeah, that's the normal phase side and then we turn around. I'll try to do it for the people on the side in a minute and then we come back and that's the con, okay, so now let me, I'm going to do this now, I'll try to do this since we brought all these high-tech gadgets. with um here we go for the people over there that's the um that's the concave side right there and oh, you're going to be difficult, right? um, how is that going to work?

Wait, you're getting there, yeah, right, great, okay, great. If that's all down now, that was the hard part that went without much trouble, that's good, so what's happening there, your brain is getting perfectly good when you look at the concave side, your brain is getting perfectly good information. , saying. that that is a concave side um there is no lack of information that is transmitted to the brain there, on the other hand, a part of your brain is saying that faces are not like that I know a lot about faces faces are convex that's what The face is particularly a famous face.

I've seen it many times, and that convexity prediction is overriding all that real sensory information coming from Vision, in this case, so in a way this is a good thing for your brain. It's like trying to separate signal from noise using what you know about the world, but sometimes you get it wrong, like when you look at the concave side of the mask or if you fall on a 15cm nail and there's a shock. that it didn't work, so you know, now I'm going to go over this part which was another version, so I think we'll all have experienced something at this stage if you've ever experienced Phantom.

Phone Vibrations You probably experienced that feeling of your phone ringing in your pocket even when it's not actually in your pocket or it's turned off or no one is calling you. They are very common. In fact, I now have phantom vibrations on my wrist thanks to the Apple Watch. The vibrations on my Phantom wrist are very smart now because if I start walking quickly I expect the watch to give me a double ping saying "hey, you're exercising" and I feel like even if I walk quickly I'm not wearing the watch. Really strange, so I think we all experience our predictive brains in action, what we don't realize is how much of everything we experience is like this.

That's mainly what I'm going to talk about. There is a recipe for the Phantom phone. The vibrations are extremely common in people who strongly expect a call, it is not a big surprise that internal doctors, for example, have a very high incidence of phantom phone vibrations, they are exacerbated by stress and caffeine, and both are mix with dopaminergics. systems that came back to later, because dopamine plays an important role in the predictive brain and other neurotransmitters we'll get to that, but there is a little recipe to make ghost phone vibrations do one more demo, if it works, this is sine wave speech.

So what happens in S Wave speech is that there's an ordinary voice recording and then a lot of the usual acoustic information is stripped away, so you get a sort of skeletal version. Sounds a bit like clangers. Anyone who remembers clangers. It's a British thing, if you're not British, you know, don't worry about this, but in the past there were these things on the moon and they were like this and this is what influence speech sounds like, um when you hear it the first time, what you'll experience here it's very fast learning by the brain, so I'll play you the sine wave sentence, I'll play you the real sentence, the one that doesn't fall apart, and then you.

I'll listen to the sine wave again and all I want you to do is simply appreciate the difference in your experience, the enormous difference in your phenomenology, as we philosophers sometimes say a good prediction can make, so here we go, it was a sunny day and the kids were going to the park real phrase coming it was a sunny day and the kids were going to the park the same sign says hello again it was a fun day the kids were going to the park should sound very, very different with that I want two more the C was kept in the cage at the zoo the camel was kept in a cage at the zoo the camel was kept in a cage at the zoo see he was sitting at his desk in his office he was sitting at his desk in his office so If you listen to that type of recordings enough, you will become a native speaker of sine wave speech.

Can you play me a new recording. I will listen to her well. You know, it's just oh yeah. I know what it is. So when we listen. speech sounds in our ordinary language we are performing the same kind of trick, you hear spaces between the words, well sometimes I don't speak that fast, you hear spaces between the words I'm saying, but if you look at the actual acoustic signal it's quite continuous, those spaces are inserted by the predictive brain because it correctly predicts that there are word boundaries there, so everything we experience is constructed this way, this was in case the demo didn't work, it's like it's a visual version of the same.

Look at what's happening now, look at that one more time. So what this suggests, to use a phrase that appears in this literature, is that perception is a kind of controlled hallucination, the brain is continually trying to guess what is most likely to happen. Both the intercept and the extracephalic forecast made little difference to the actual weather in that region, that's kind of a strange world and let's not even think about what happens on the border lines between regions and things, but you know, a reliable forecast from rain it made rain. itself a little more likely a reliable forecast of the sun made the sun itself a little more likely, unfortunately we don't live in that kind of climate world, but I think this is the experiential world we live in in our reliable forecasts , we modify our own experience in the direction of the forecast now you know people have been saying things like this for a long time um there are things like this in Emanuel K a kind of constructivist picture of perception and experience is really super clear in Herman Von Helmholtz uh the idea that perception is a product of unconscious inference and in the 20th century, Rick Niser, Richard Gregory and others, analysis by synthesis movement, had the same kind of image and here is Richard Gregory, giving the lectures of Christmas, I think here in 1979, so I feel very.

Like I should have more bangs and whistles, that's all I got, yeah, um, so what's new now? Well, since about 2005, the new thing I think is a comprehensive neuroscientific theory of all these effects and it is actually a theory that deals not only with perception but also with Action also tonight is not enough for me to give you an explanation of the action, but it will be important in the discussion. I imagine you may already know that one of the pioneers of this is Carl Friston at UCL here in London. He published the theory of cortical responses in 2005.

He was beginning the hard work of turning these kinds of intuitions into a story that was computationally clear and also as neurophysiologically plausible. That's where we are now. I think so much so that many people are writing books about It's Included. I am selling a The Experience machine. um, so there are a lot of good books about this out there. um. My colleague Annel Seth has one of the first of Yakob Houie, the wonderful book of the predictive mind. Lisa Felman Barrett's book on how emotions are created. There is a big emphasis on inceptive prediction predicting the body of states and the importance of that.

My own previous book and the current type of Bible, the Active Inference Bible because predicting, I'll call it predictive processing, a lot of other people call it active inference, which I find. a little clumsy but it's the same thing, so what I'm going to do is present, as you say, an outline of some 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 we looked at some puzzles and challenges towards the end, so what did I do there?

The more traditional view had a perception of Sonic that works from the outside. We have this on the card of the day, all those beautiful drawings of energies. hitting the eyes impressing things like little tubes almost like small wax The impressions go deeper and deeper into the brain eventually something cool happens with the p gland we won't worry about it um but it was all outside in the perception it was the world coming from the outside and making a richer impression on the brain and this image emerged in 20th century Neuroscience, where most of the leading textbooks when talking about Vision have all the arrows moving inward from the sensory peripheries. deeper into the brain in artificial intelligence David Mah basically had a breakthrough processing picture from um Vision, wonderful work, super influential, uh, but predictive processing turns a lot of this on its head, most of the brain's work is It works from the inside out, that enormous metabolic budget that we will return to soon. mainly 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 is a very efficient and interesting thing is that to understand 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 type of visual scene and I am 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 that I haven't been able to predict, then something should ideally happen, so prediction error signals are kind of an antihero, almost all these accounts carry the news, they are sensory information that is currently unexplained, so they are a form of sensory information, they are just that particular part that is. currently not explained by the brain's ongoing best predictions, they carry the news and what you use it for, you use it to select better predictions, so if I'm going to see something I didn't expect to be there, I better hurry up . of prediction error um I use that prediction error to recruit a better part of themodel, a part of the model that has a yellow plastic duck or something so that I can familiarize myself with the inexplicable fragments of sensory information uh oh and that's some of what these models really look like if you draw them schematically what to take away from this is these are multi-level models, the brain is not just sitting there making a prediction and I don't know, um, uh.

What should I have? Laptop on the table. The high levels are making predictions about I know something is happening. The lower levels are making predictions about the precise details of the arrangement of things on the screen and each middle layer is trying to predict the layer below. so you know the same way you could use information you know about sentences to predict things about words and things you know about words to predict things about letters and things you know about letters to predict the distribution of ink on a page, so it's going to be a little bit. like that, so it's super efficient.

You use the active world model to make an educated guess and If the guess is unreliable and we'll come back to that in a minute, bits of sensory information, prediction error is your friend and should improve your guess. It's super efficient. It is used in commercial compressed motion videos, for example. We go back to 1959 when 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 the previous frame, all you need to do is send the difference, so If someone is racing 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 know their gate, how they were running, then you don't even need to update where their legs are going because You predicted that too, so you can be as smart as you want about this, the smarter you are, the more efficient you will be and I think this helps us understand what is a really puzzling fact about the brain: the neural connections in the brain that work as they sometimes say that from the top down, from deep in the brain to the sensory peripheries, they greatly outweigh those carrying information inward, which is a little surprising in terms of perception, it's a very important thing, which we do all the time and everything is from the outside in.

It is not primarily from in and out and prediction errors simply order the edges of the image here, for example at least 90% of the input to the lateral genicular nucleus, which is just after the retina, comes from the most deep in the brain, um. On average, across the entire cortex, these exiting paths outnumber incoming paths by at least two to one and in some areas by four to one, so the brain is connected from the inside out and this type of accounts make a lot of sense in that wiring. then the anomaly is resolved, the hard work is done from the inside out and it is not surprising that the brain expends an enormous amount of energy all the time, even when it is at rest, because it maintains the model of the world, it maintains a model that can be used to make these very detailed predictions it's like you're constantly running a simulation of

reality

in your brain, well you're running a simulation of

reality

, in fact you're experiencing one right now and if I do something unexpected I'll just hit it. a few prediction errors.

I know we come on, so prediction error anchors the brain's imagination if you like the brain's constructions for the world, but this raises a puzzling question if the brain can do this if it can actually use FL as a prediction error to correct. quickly your your own assumption why some of us were wrong, you know what happened with the worker who fell on the 15cm nail, you know there was no receptive information about the pain coming from the toe where they felt the pain, um to the brain, why not? the brain immediately gets prediction errors and corrects its guesses, so this brings us to what I think is possibly the most important part of the contemporary picture, it's something that wasn't there in the previous versions of this story, um, it The idea that 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 take certain predictions into account, how much to take into account. into account certain fragments of apparently sensory information. counterevidence, you know the Hollow Mask case, your brain is ignoring a lot of sensory counterevidence because it's very confident in its predictions, which of course, again, is interesting from your perspective because you know it's a hollow M, but you're still Looking at it the other way around, because those intermediate levels of processing are absolutely convinced of the convexity that occurs, so it's kind of a weight in the game: the brain estimates how much confidence it has in certain predictions or certain bits of sensory information. information, this is known as precision weighting, and the precision waiting game is a kind of zero-sum game, so if I raise the precision on my predictions, I'm lowering the precision on the bits of sensory evidence that meet those predictions. and vice versa.

This is happening at all levels of processing and if these accounts are correct in every area of ​​the brain, they probably aren't all processing accuracy and prediction errors in the same way that the amygdala has something special to do, but that's kind. of what everyone is doing, if these accounts are on the right track, we will return to that question. um, yeah, on a foggy day, for example, you might not want to rely too much on what your eyes are telling you, especially if you know the area where you have a big model of the world, it's like it's going to trust my model and, you know, I will take into account a little bit of visual information, but not too much, so in the brain the idea is that This precision weight in the economy focuses on the complex neurotransmitter system, so all the things there's dopamine, norepinephrine, attic acetal, choline and garba glutamate, they're all in there somewhere, serotonin, um, so that's a big part of the precision weight in economics. neurotransmitter systems is also probably being done on faster time scales definitely on faster time scales by the neuronal phase by the temporal phase of the low neuronal oscillations versus the fast neuronal oscillation frequencies will return to those towards the end if there is time , so precision increases the influence of particular predictions or particular chunks of sensory information is kind of like what attention is, if these accounts are correct, then what attention really is is uh variable Accuracy weight on what it happens in the brain, which I think is a useful thing, we all use this word attention, no one really knows exactly what it is there's a proposition here that's what attention is is exactly that um this is an image of Mooney it's a bit like the woman with the horse from before.

Yes you can see this one, but here's the real picture and here's Mooney again. Right now I think what we can learn from these things is that predictive brains have a very, very delicate job to do, they have to walk a very fine line, if you want to detect the hidden frog, you have to turn up the Precision by a huge amount. amount of information about particular Frog Outlines and degrade other pieces of information, that's what happens when you allow the Frog to. I was going to say jump at you, yeah, exactly, um and I. I think it's because we can do this delicate precision weighting that predictive brains are good detectors of weak patterns in a world full of noise, so it's important to get these weights right and if they go wrong, very bad things will happen, that's what you're trying to say, you know?

If your brain starts to underestimate the value of certain predictions or overestimate the value of certain bits of sensory information, then bad things will happen and we'll see some of those bad things in the next half hour, but at this point. You might also be asking: Can't the right balance be achieved? You know, it's actually a very difficult question. We can return to it in the discussion because I would say that there is no right balance. There is no balance such that that balance always exists. reveal how things are because it all depends on what kind of world you live in at the time, how volatile what's happening there is, so I don't think there's a right answer for how you can do this balancing act. some ways that can change, so computational psychiatry, as it appears through the lens of this account, is primarily a matter of trying to see how differences in that weighted balance of precision in action might generate different profiles of experience.

That's pretty much what's happening in A Lot of Work, I think it's a good way of putting together a systematic understanding of many atypical experiences, the wide range of typical experiences and also what happens under the effects of psychedelic drugs, e.g. something we won't understand. So, for example, suppose you assign a very high weight to sensory evidence. A very, very high weight on sensory evidence. What will that do to you as a prediction machine? One thing it will do is make it harder for you to detect. weak patterns in a noisy environment to detect weak patterns in a noisy world you want your model of how things should be I don't know, take a firm hand at the sensory evidence your model is what will give you the best There is probably a lot of interesting work that suggest that the autism spectrum condition involves precisely this kind of overweighting of sensory evidence, which is, in fact, a much enhanced kind of sensory evidence, is the kind of profile that is emerging from quite a lot of detailed studies here and there, there are a very interesting parallel debate that I won't get into, but it is a parallel debate about whether what is happening is a weakened influence of models or an increased influence of sensory information from a basic perspective.

They will look very, very similar, but they will not look exactly the same and the evidence now strongly suggests that the autism spectrum condition is not about weakened use of the model, it is actually about enhanced sensory information. So what if you were enhancing sensory input? So, well, many environments will seem difficult to negotiate, you may become wary of noisier and harder to predict environments, social worlds where many very small signals really matter, can be especially challenging, so again, don't I know nothing of what emerges from these accounts. is if there is a kind of social understanding um difficulty in the autism spectrum Condition is a side effect of the real thing that is enhanced sensory information is not something fundamental I think it's quite interesting um it also makes sense in cases where people with the Autism Spectrum condition just makes it better, so things like the embedded

shapes

task where you're showing something and you ask to find the geometric figure in what's there, the rocking horse, whatever it is, um, the People with the Autism Spectrum condition do better than neurotypicals probably because they are better at allowing sensory information to speak for itself, you know, they don't get carried away too much by that kind of strong top-down um, it's whatever, I don't know what that really is.

The HSE thing, so now imagine something different, imagine that instead of assigning extra weight to 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 start to freak out. things that you know you've taken some of the control away from a controlled hallucination and you just have a hallucination um and you know I think we all do a little bit of this that's why we see faces in clouds and stuff like that um because uh we can have predictions. solids about faces and if we allow a certain type of relaxation to occur, you can make this place so that there are accounts of delusions and hallucinations in schizophrenia and psychoses that are emerging at this point where the idea is that there is a kind of overweighting of unconscious predictions and that, in turn, could arise if you have false prediction errors, if for some reason the brain generates false prediction errors, then you will have to come up with some kind of model to test. to deal with these things that tell you this is important, you're not understanding this, this is something important, revise your model so that the model revises itself, becomes very firm and then can alter the experience, pulling it in its own direction. like with the Hollow Mask illusion, yes, so they become very weighted.

These may be predictive models of personifications of complex conspiracies today. It's the Internet, the Internet is a source of many of these models. At that point, a strong prediction is altering your experience. in a way that later seems to confirm the modelthat's making the predictions and that's something that I think is really important to understanding the tendency of predictive brains to get caught in self-confirming stimulus loops where your own predictions are changing the way you perceive the world and then you think, oh , I'm getting evidence for the model I'm using to make these predictions, but that evidence is furious.

I think we have effects like that in daily life, in fact, in our own unconscious. Expectations alter the way we see and experience things and we think we are getting evidence for our own models, so it's kind of a kill. I think of the predictive brain falsely confirming its predictions, but also turning out to be something we could take advantage of. They created expectations in them by teaching them to associate a particular geometric Q with higher or lower levels of heat, as indicated by the little thermometer, so that in the induction phase, as they called it, expectations were set 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 geometric cues, so now they are actually receiving some heat, but unbeknownst to them, the heat levels applied were all exactly the same, so they get the Q geometric. They already associate it with more or less heat, but actually now they're getting exactly the same amount of heat that you would expect the subjects to report. expectation induced effects more or less pain in line with expectations induced geometric signals and they supported this not only did they ask them this question, it was supported by something called neurological pain signature a slightly controversial fmri profile for um something like real pain go on instead of just reporting what I think you want to hear me report so they did and also a report so that explains.

I think why they didn't update. This is what I think is interesting here. You should be asking yourself, but why don't you update? They give you the same warmth. Can't you just feel it and update your model? But you can't because your model makes you feel more or less hot. I think you're getting more evidence for your model, you certainly don't think you're getting contrary evidence for your model, so I think that's a good example, just a simple kind of controlled example of the way that experience, when shaped by prediction, you can get caught in these types of self-confirming spiromatic cycles, since they say that expectations modify the perception of pain and this modified perception drives subsequent expectations, so you keep the wrong ones even when you have good sensory evidence to the contrary , so what I want to do now is see how cycles like this might work on things that are maybe a little bit closer to chronic pain.

This is the example I'll start with here, so somewhere between a third and half of the UK population. at least I experienced chronic pain, so it's a huge burden on your health, so there's a conundrum in chronic pain. I mean, chronic pain, as everyone knows, has been defined for a long time as kind of a long-term version of pain, it's not the acute pain that What happens when you break an ankle or something like that is something that's been around. happening over a time, so the match between the chronic pain experience and the standard identified causes is only high for acute and localized things like ankle fracture etc.

If you move on to longer term conditions, the overlap is very bad and there are many examples of this shortness of breath in lung diseases, discomfort with atrial fibrillation, asthma, cancer-related fatigue, low back pain, uh, a very, very example surprising, and these are all cases where the amount of pain experienced just doesn't fit very well with the peripheral etiology if you like what seems to be happening structurally in the relevant parts of the body and this doesn't just happen in different patients, but in the same patient at different times. Additionally, there is a huge amount of variation for what is apparently more or less the same level of basic structural damage type and also affects the same person at different times.

I think this is that I, no, you have the same ringtone as me. You're just making me panic, that's right, so yeah, a wide variation in pain experienced. I think this fits very well from a predictive process perspective. I mean what happens during those longer periods of time. One thing that happens is you. I have plenty of time to form expectations about how their pain experiences will vary in different contexts. Different life stories will also do this differently, so I think we can expect different life stories to balance that kind of weighted precision differently. Balancing act and random things happening contextually like I don't know if you experience an asthma attack in a particular type of context that might make you more likely to experience it in another context given the role of prediction in those events, um then they can become very sensitive to context and I think that can happen even when the context turns out not to be clinically relevant.

The longer you live with a condition, the more time you have for things like that to start happening, not surprising. um, so this is kind of the same thing again about predictions changing experience and then you think you're getting evidence for the model you're using to make the predictions even when you're not. I think this is part of the chronic pain puzzle and something we can begin to address. Let's get to that so that no one says that chronic pain is not, so to speak, really experienced as pain, of course it is. You know, just like the person who fell on the nail was experiencing excruciating pain, so far it will seem like bad news, it seems like you know that predictive brains are good, they are quite efficient, but they are understanding us. in all sorts of nasty little corners, what can we do about them?

So there's good news, with apologies to John Cinski. I think there's kind of an opportunity here to reject the bad predictions and install different predictions that might be useful. it might even be falsely useful in a certain way, you know, maybe it's a hack in the same way, so I'll end by looking at some of those interventions, emerging opportunities to hack the predictive brain in a good way. direction, so there is an obvious example: the careful use of placebos, a very interesting area. I think it's known that the effectiveness of placebos varies greatly, it varies basically depending on how confident the patient can be in the relief from the placebo.

Describe differently the power of what they are giving you. You can also get it by doing placebo surgery, for example, instead of just a placebo pill. Placebo surgery is very effective for certain things, mainly I think because we think surgery is pretty. Seriously, you know, I think it's a pretty serious intervention. I hope, I hope for something from this. At least that's what I think the brain thinks. So, patients who receive placebo surgery for osteoarthritis of the knee report relief similar to those who undergo regular surgery. and they get more relief than people who get placebos or pills or training, so you know, there are some really good studies here.

Athletes show better performance when connected to what they believe are pure oxygen delivery systems, when in reality they are just air. They also show better performance. If they are given a medication that they are told will increase red blood cell counts, even though it doesn't, yes, a 1.5% increase, which is big if you're an elite athlete, so obviously there are tempting ways to to get it. Around the doping laws here, it's interesting to me that honest placebos work too. An honest placebo is one where they tell you it is a placebo. You know there are no standard clinically active ingredients here.

But still, many people report significant amounts. of relief, I think that's just another um, sorry, I've gone too far, I think that's just another manifestation of the fact that most predictions are unconscious that you know if they give you something that comes out of a kind of small and nice bubble wrap. Something from someone authoritarian in a white coat, even if you know it's a placebo at the mid levels of your neurological processing, right? It's like, oh yeah, this is happening, so another kind of systematic pushback here is something called pain. Reprocessing theory is very interesting and promising, I think to get into this kind of part of the account, we have to think that the basic experience of pain has some kind of role to play and that role tells us that our bodies are in imminent danger. that if we basically keep doing what we're doing this is not going to be good and that's true if you just broke your ankle you don't want to keep walking In that right now it's very Cy Borgy ankle, you know what's going on there, but in many Of these cases of long-term chronic pain and disability, it seems that predicting imminent bodily harm has become the problem if you like what's blocked.

There's an expectation that if you keep doing things it's going to be really bad for you, that's a big component of what pain is in these Accounts, at least that's the expectation, so as people say in this literature, your pain experience is a bit. a bit like a malfunctioning warning light on a car it's a bit like you know that light is coming on it's telling you that you better pull over to the side of the road right now don't keep doing this it looks like you have to stop what's going on but let's say it's a warning light that's working SM and of course the interesting thing here is if you think of a warning as a malfunction you probably think oh there's some wiring that went wrong but In the predictive processing accounts of warning light malfunctions there isn't even a wiring problem, it's just some kind of balance problem in act, there's nothing structurally wrong with the wiring of the brain, so I think it's pretty interesting, it's kind of a software problem if you like the chronic back pain case.

I mentioned earlier that in 85% of cases there is no standard peripheral cause. That doesn't mean there isn't a cause, it just isn't a cause that matches the pain. What pain reprocessing is theoretically trying to do is reframe pain in a way that counters those predictions basically involves doing a talk a little bit like this. I guess saying that Chron IC's pain, you know, could be that kind of false alarm, involving blocked predictions but then trying to install different predictions primarily by getting people to do a little more despite the pain and, interestingly, doing more does that the experiential pain is less, so it's a little bit like just driving the car a little bit more can make the warning light dim a little bit, which is what I think is the trick we want. .

There is a very good document called "This Could Hurt." I think you can probably still stay on Netflix or Prime, so this will be a useful cycle. I think it's the other side of all those useless people. Cycles in which expectations modify experience in bad, self-consolidating ways. This is a way to modify an experience that could be useful and entrenched in a good way, which is why the best systematic study here is Asher Al's 2022 and 2023 study on chronic back. report pain and neuroimaging as mentioned above and basically found that it works with large and sustained reductions in pain and disability after PRT compared to placebo and usual care in about 73% of cases, so shows promise for chronic back pain. a particular type of pain is fine, how am I doing for the time?

Oh, this is interesting, maybe I'll have time to do it. I wanted to do it. Imagine, let's see. So another example is functional neurological disorders, sometimes called psychosomatic or psychogenic, those terms aren't used as much. now, um, functional neurological disorders, neurological symptoms without standard structural causes, so you know that chronic pain is actually a finding, by this definition, at least 16% of new referrals to neurology clinics are ultimately diagnosed with some type of functional neurological disorder. which basically means, um, that you can't find anything, um, so examples include unexplained cases of blindness, deafness, pain, fatigue, weakness, abnormal gait, trema, seizures, um, functional problems are typically diagnosed by variabilities of disabilities, so usually things like um, um, distract attention from a particular body. part may reveal intact abilities in the sense that in that part I think the Hoover sign works that way.

NoI'll have time to get into these examples, but one example we can do is an example shared with me on With Me by John Stone ex. colleague in Edinburgh, this is a woman who had a slow but eventually complete loss of vision, she had had severe migraines before, but she had no other history of stress, trauma, abuse, head injury, anxiety, any of, you know , nothing at all, really. severe migraines fmri was normal um what Stone noticed was that when she talked to him about her total blindness she was copying his body movements and sometimes followed his gaze the entire time without experiencing anything visually or consciously um but she could do Gaye following a copy of movement and that's a pretty strong clue that there's something more intact there than you imagine, his eyes responded normally to an optokinetic drum, so that's something that draws a stigma response from the eyes and so on.

They kind of twitch a little bit in response to the change in pattern, so she exhibited the normal pattern of response with the optokinetic drum ston and her colleagues surmised that the severe migraines had basically caused her brain to start constantly predicting Darkness, You know? She spent a lot of time retreating from migraines to dark places. She began to think that they were just waiting for the Darkness. Her experience then had to fit that expectation, be a bit like your experience with the Hollow Mask or the Ghost Phone vibrations that you know. these things that we all experience are just a pretty dramatic version, so the treatment was to try to push back a little bit, like pain reprocessing, actually against those hidden predictions.

One thing that Stone and his colleagues did was just share the science of predictive brains and how predictions locked in predictions can affect the experience um also um when I say videos shared videos shared with people I trusted because I couldn't see the videos you know she's still functionally blind um also induced experiences of flashes of light using transmagnetic cranial stimulation so you know know how big, how big a mechanical wind-up toy looks like that applies a large pulse of magnetism to the brain that can induce phosphines and that's proven in your brain, that's the idea you're demonstrating. to the brain that seeing is possible so she recovered completely, of course you don't know if those things were responsible for the recovery, you never know, maybe she recovered anyway, but she recovered, and the main message today It's just that these are just extreme examples of the way we are all constructing our sensory experience all the time.

Functional neurological disorders are just kind of a dramatic 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 have a disease Burns other types of injuries, so I think the more we appreciate this , there should be less resistance to the diagnosis of a functional neurological disorder, so one thing that happens now is that patients don't want to hear that, they really don't want to be told that they have a functional neurological disorder and I think that's because we have the wrong metaphysical model of what it is to be a normal human being.

Human beings must construct experience in this way, so functional neurological disorder suggests an extreme example. Hopefully, putting these kinds of narratives on the table might help us see beyond old mind-body dualisms—it might even completely erode the distinction between psychiatric and neurological disorders. You know personally, I don't think it's a very useful distinction, so in general, there are many other promising ways to hack and push our predictive brains. An immersive virtual reality therapy for pain. This is very effective. It is sometimes used as an anesthetic during dentistry. There are good ones. Good results when changing bandages on burn patients, so this will be more or less in some kind of immersive virtual reality watching an undulating jellyfish or something nice and relaxing like that, and I think what these beads give you is not just a distraction . you might think it's just a distraction and it is in a way, but now we understand what distraction does, just like verbal reframing, another example, is a way of altering those accuracy weightings and the Seed into different predictions and everything.

This will change your verbal experience. The reframing is also very powerful, so you know, I get little tingles in my hands before I have a lot of them behind that door. um, and you know, I've tried to reframe that as chemical preparation to deliver a good performance instead of, well, everything's going to crash. and burning it, it seems to seem to help, so I think this all falls into place, the verbal self-affirmation, the reframing of the therapeutic therapeutic use of music, touch and ritual, I think we have a systematic way of understanding that these are genuine and powerful interventions um they're all pushing predictive brains meditation is another one that I'm really interested in um my former postdoc Mark Miller has done a lot of work on this, what he believes meditation is doing is giving you greater control over precision. weighting mechanism um and you can see how that works, you know, I can enhance the sensory experience so that it sort of reduces the grip on the models.

I can allow things to drift or be set very precisely. If we could control our own precision-weighting mechanisms, we would have what my colleague Al and Deans call phenomenological control, which would be amazing if you knew you could modify your own experience in the best way. for you and your life right 8 minutes to say what's missing that should be enough shouldn't it? um so it's a very incomplete picture to alter the brain's best predictions it's not going to heal a fracture it's not going to cure a cancer it's not going to kill a virus so I don't think we know what the extent and limits of these effects are yet. determined.

It's clearly good for things like pain, anxiety, shortness of breath, fatigue, sports performance, cancer-related fatigue, so you know, there's a lot of clear applications. and unknown limits, there are large individual differences in the effectiveness of some of these therapies, therapies like placebo hypnotism, pain reprocessing theory, and there are some very, very interesting clues that there might be an underlying genetic difference, so that there's an enzyme called CT and and CT controls the metabolism of dopamine in the brain, so the more CT you have, the more um you're consuming dopamine. You can see what could happen here if you didn't ingest so much dopamine.

So if dopamine plays an important role in the precision waiting game, it may have more phenomenological control and that is in line with the results here, which is that people with genetically determined high levels of CT are less susceptible to these things like placebos hypnotism PRT so I think it's very interesting to take these things directly to the molecular level and then go back up to the experiential level. A great challenge is to determine what the contents of neural predictions are in different brains. areas that you know are really not that well known. There is some good work from the L mle lab in Glasgow and the title of their paper tells what they found.

Scene representations transmitted by cortical feedback to Early Vision can be described by line drawings, so the idea. there's a kind of line drawing level information that feeds back, but you have beautiful experiments that suggest good work on the amydala recently we need to discover the detailed neural circuitry of error prediction and sharing. There are many interesting things, some wet. Neuroscience work finds explicit prediction error responses in layers two and three of the mouse cortex, and that's a nice recent twist. I'm not going to mention the moment there is good work on the role of Alpha Beta frequencies versus gamma frequencies in communicating predictions and errors with the idea that fast gamma frequencies are carrying errors through the system and the slower Alpha Beta frequencies are bringing predictions to the system, this has even led to a suggestion for a particular implementation of these accounts that is different than the standard one and I won't go through that difference because time will trip me up and I won't get to the latest part of the metaphysics yes I do, but the idea is that you have the same neural circuits that make prediction errors, but prepared differently by Alpha Beta rhythms that allow gamma to pass where they are not prepared, there was no point in going back to that in the discussion, okay, I think, in any case, this is a good moment in the history of this account because we have some competing stories about what the implementation is in the brain and only when you decide on an implementation can you really start to test these stories properly so you can play pin the story in the cyan cycle here, where are they? accounts in the prescience cycle normal science drift model crisis model um I'm not going to reveal where I think we are in that cycle but certainly uh having competitive implementation level proposals is a really important moment um evidence based gift evaluation Don't worry about That, the conceptual challenges, is prediction, really the most fundamental thing that brains do, some people think that if it were like that, we would just crawl into a dark corner and stay there making very successful predictions, there is a good way out. from that dark corner, but I would have to give you the account of the action, but there will be time to discuss it, perhaps that is sometimes called the puzzle of the room and the darkness, what is the relationship between all those mid-level unconscious predictions and the higher conscious level?

No predictions, no one knows, how does language fit into the predictive process of the image? Nobody knows, what happens to culture and the human-built environment? We build big worlds to think in, we build worlds that allow us to minimize different types of errors to get our jobs. done like painting lines on roads or building large hadron colliders, um, and we have a big project on this right now, that takes me to Spain sometimes to look at dolmans in Galsia, so why not archaeologists, scientists from the vision, computational modelers and cognitive philosophers like? Me working together to explore material culture and how the worlds we build change the jobs predictive brains need to do and how these things actually work together.

We will need 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 communities of agent ecosystems that predict each other. For some speculation on that, see the group's work versus AI Carl Friston, the person who did much of the active work. Inflence is the scientific director of that group, so keep an eye on them. Oh yes, the hard problem is why and how conscious experience is possible. You may have noticed that the book is called The Experience Machine, but I didn't mention the hard problem. um, I think there are great clues.

I discuss them in the book Big Clues involving the explanation of interception, the body predicting its own bodily states, and the explanation of action. Action is a way to resolve prediction errors by changing the world to make Some predictions come true, so once you put all those parts together, I think we brushed the edges of the hard problem, so I think there's real progress. I tried to tell the kind of optimistic story in the experience machine. I think we're getting closer. a multi-level, principled science that looks at at least the structure and variety of human experience, we have a small set of factors, predictions, accuracies, prediction errors, and they combine to produce these striking differences in experience, so I think We turn to a periodic table. of experiential variation, one that captures neurotypical, atypical and altered states, but at least I think this brings us a few steps closer to appreciating both the diversity and continuity of human experience, thank you.