Become a Creator today!Start creating today - Share your story with the world!
Start for free
00:00:00
00:00:01
Active Inference, the Free Energy Principle, and Therapy
176 Plays5 months ago
In this episode, Grant and Fara are joined by esteemed neurophysicist and authority on brain imaging, Dr. Karl Friston. His revolutionary impact on studies of the brain derives from his inventive use of probability theory to analyze neural imaging data. They discuss how we might think about psychotherapy from the point of view of Dr. Friston’s work and briefly touch on artificial general intelligence, AGI, through Karl’s work on VERSES AI.
We hope you enjoy!
In This Episode
[0:00] Introduction
[1:35] Episode Overview
[02:30] Journey into Brain Imaging Research
[6:30] Mathematics and Brain Processing
[12:10] Active Inference and its Relevance to Psychotherapy
[24:10] Therapeutic Alliance as Computational System
[27:50] Revising Entrenched Beliefs Through Therapy
[32:20] Trauma, Condition, and Altered Rationale
[55:00] Mindfulness and Managing Prediction Errors
[1:00:00] Consciousness and Selfhood
[1:04:35] Verses AI and the Future of Psychiatry
Resources and Links
Doorknob Comments
https://www.doorknobcomments.com/
Dr. Karl Friston
https://profiles.ucl.ac.uk/2747-karl-friston
Dr. Fara White
Dr. Grant Brenner
Recommended
Transcript
Introduction to Dr. Carl Friston and Discussion Focus
00:00:00
Speaker
We're very excited to offer this conversation with Dr. Carl Friston, whom Farah introduces in more detail in just a bit. For me personally, it was a huge treat and honor to interview him on his work with Active Inference, Mathematical Psychology and Computational Psychiatry. Carl is truly a world-class intellect. Most of his interviews range across computer science, physics, and mathematics. We've chosen to focus on the process of psychotherapy from the point of his work, with basic mechanics a kind of fuzzy logic which more closely models the way we actually think about the world even if we don't know we're doing it. Listeners are encouraged to take time with this episode some of it may be beyond reach while other areas offer deep insights into the nature of brain mind and behavior we hope you appreciate it as much as we do.
Hosts Introduce Themselves and Discuss Doorknob Comments
00:00:49
Speaker
Hello i'm dr fairway.
00:00:51
Speaker
And I'm Dr. Grant Brenner. We're psychiatrists and therapists in private practice in New York. We started this podcast in 2019 to draw attention to a phenomenon called the door knob comment. Door knob comments are important things we all say from time to time, just as we're leaving the office, sometimes literally hand on the door knob. Doorknob comments happen not only during therapy, but also in everyday life. The point is that sometimes we aren't sure how to express the deeply meaningful things we're feeling, thinking, and experiencing. Maybe we're afraid to bring certain things out into the open or are on the fence about wanting to discuss them. Sometimes we know we've got something we're unsure about sharing and are keeping it to ourselves. And sometimes we surprise ourselves by what comes out.
00:01:36
Speaker
if so Hi, thanks for tuning into doorknob comments today. It is my pleasure to introduce Dr.
Dr. Friston's Background in Neuroscience and Psychology
00:01:44
Speaker
Carl Friston. He's a theoretical neuroscientist and an authority on brain imaging based at the Wellcome Center for Human Neuroimaging. He is considered by many to be the world's foremost neuroscientist. Although Dr. Fristen trained in psychiatry, his revolutionary impact on studies of the brain derives from his inventive use of probability theory to analyze neural imaging data. These contributions were motivated by schizophrenia research and theoretical studies of value learning formulated as the disconnection hypothesis of schizophrenia.
00:02:17
Speaker
He invented many techniques used universally for brain imaging and he received a Golden Brain Award from the Minerva Foundation in 2003 and the Weldon Memorial Prize in 2013. I'm delighted to have you with us today on the Doorknob comments podcast where we plan to discuss how we might think about psychotherapy from the point of view of your work and get a preview about artificial general intelligence AGI i to which we may look forward with your work on versus AI genius. That's never to be here. Thank you very much for inviting me. It's a pleasure, Carl. um Could you tell us a little bit about how you got into the field and how you shifted into neuroscience and now artificial intelligence? I as a teenager wanted to be what was then known as a mathematical psychologist, so wanted to study your
00:03:11
Speaker
the most interesting thing in the universe, which of course was me and my parents and my family and my friends, but had a potential and an interest in formalizing things mathematically. So I chose a dual career in physics and in um psychology initially, but then trained as a clinician in psychiatry, and then entered research through fortuitously brain imaging at a time when that field was being established.
00:03:43
Speaker
requiring a lot of problem solving and interesting the kinds of problems that needed solving to do that research involved making sense of data, brain imaging data, and it became very clear that of course this is exactly the problem that we face as people foraging as little scientists making sense of our world, making sense of our of our sensory data, So it's a fairly easy step to repurpose all the analytic tools and ideas um and apply them not to the analysis of medical time series but to the analysis of the kind of time series that our brains process.
00:04:25
Speaker
Oh, that's fascinating. It sounds like, you know, potentially you were in the right place at the right time, but also I wonder in terms of, and and you talked a little bit about your family of origin and finding some inspiration there.
Influences and Transition into AI and Neuroscience
00:04:41
Speaker
Do you remember, you know, any specific moments, like what was the first moment that you really got excited about the potential for either new imaging tests or things that you'd sort of come across? Well, the first time I felt, I'm not very excited, but sort of compelled to commit a life to this kind of inquiry was after reading all the books that my mother used to give me. shit So she was a nurse and she yeah she loved popular psychology. So I'd have to read your Edward de Bono thinking laterally and all of those wonderful sort of 1960s, 70s treatments of popular psychology and to a certain extent psychopathology.
00:05:24
Speaker
At the same time, my father had a passion for maths, so he used to make me read things like Sir Arthur Eddington's Space, Time and Gravitation. So I think it will be useful, if not exciting, to put these two perspectives on the way the world works together and hence the mathematical psychology. in latter years um you know I told a story there where we take the same principles that apply or are applied by scientists to scientific data medical time series for example and ask whether the brain is doing the same kind of thing of course this was not a new insight yeah this kind of insight has been.
00:06:03
Speaker
you could argue around since the days of Plato and realising that and just understanding the legacy and the deep history of thinking about sentient behaviour, the conscious brain, me as a sentient creature, you as a s sentient creature, just realising that that literature existed and discovering that um over the past 20 to 30 years, everything you see, you say, Oh, yes, that makes entire sense. And it all starts to fit together in a relatively simple way. So every time that happens, every time you have experience when there's aha moments, there's a little a little bit of excitement that you seek. as a good scientist, and I imagine a good therapist, you know those insights.
00:06:48
Speaker
You know, it's interesting, i I can relate though, I didn't have as much influence in mathematics. I always loved it and read books like Chaos by James Gleich or Goital Escher Bach by Douglas Hofstadter at a fairly young age because of I think my life circumstances, things didn't go down the hard science path. But one thing that I'm curious about is is how you see the world and how you think people who really understand the deep, mathematics and structure of how things work with the kind of formalism that mathematics offers that is not accessible to, you know, sort of the everyday person necessarily. Like how does that change your subjective view of the world? Do you
Simplifying Complex Concepts with Mathematical Models
00:07:29
Speaker
think? And also your, your experience of, you know, being in the world? Yeah, not so very, very good question.
00:07:36
Speaker
Which i want to answer in two ways, the first is implicit do you need to be um a physicist to understand the active influence all the formulation of behavior and cognition that the free energy principle brings to the table. No, you don't. yeah Some of the most skillful and useful applications of this mechanics are done by medics and psychologists, so you don't need to have you a degree in physics to understand this. On the other hand, the simplicity afforded by being able to naturalize the processes that go on inside my head,
00:08:16
Speaker
in terms of physical processes that can be written down in terms of equations does I think simplify your view of sense making and just being or making sense of your world ah to the extent that you are now able to put that into a computer so you can now start to simulate the belief updating of all the attendant processes you think is going on inside your subject's head or your indeed your your own head. And being able to do that um paradoxically makes things a lot simpler. um And so some of the insights were really um abandoning certain notions that we reify um simply because you could you can demonstrate the same kind of emergent behaviour in silico by just applying the principles of simulating, effectively having little
00:09:15
Speaker
digital twins of a very very simplified you in a very simplified um formalized say, experimental paradigms here, psychophysics paradigm or a a game theoretic paradigm. So than ah that being able to naturalize in terms of physics, um where the semantics are the the semantic ah very precise, very, very crisp, I think allows you to see um the simplicity of the way that we work and crucially be able to to um reinterpret and de-reify a lot of semantics. you know So one interesting example and
00:10:03
Speaker
implicitly celebrated with my colleague and friend Andy Clark, is things like qualia, qualitative experience. ah when you ah when you When you drill down on it, it transpires that anything you can articulate has to have some kind of physicalized representation in your head, including things like qualia. So the very notion that Qualia exists and there's something that we can talk about tells me immediately that my world models my internal models my generative models have to have the concept of qualia and you can take this.
00:10:42
Speaker
to the extreme. Just me as a concept is just a hypothesis that I bring to the table that simplifies and and makes the but the most sense in terms of providing an accurate but very simple, minimally complex explanation for all the myriad of sensory streams that I have to assimilate and make sense of. So even things like selfhood now emerge as something which is not relative, not in essence at least mathematically distinct from more elemental percepts um you know such as you know visual percepts or auditory percepts or indeed gut feelings tummy rumblings and the like.
00:11:24
Speaker
Yeah, so qualia for people who may not be familiar with the term, it's a philosophical term and it's related to, I think, what is often called the hard problem of consciousness, which is kind of like, how do you know that when I see a red ball, it looks the same inside your head as when when you see a red ball? We can agree that there's a red ball, um but we don't really have any way of knowing whether it actually is experienced the same way. And and so I also hear something like, And this is true for for me, if you can get through the complexity, whether it's through kind of formal maths or through metaphor, then it simplifies things and you can function you know better in the world.
Active Inference and Building Psychotherapy
00:12:10
Speaker
Now, you're using the concept that you that you that you created called active inference, and we can talk a little bit about what that is. it's a kind of um
00:12:19
Speaker
a trial and error process of becoming more and more efficient for moving through the world um by comparing your models with the outcome and looking at how your perceptions and beliefs and your actions um can become more efficient over time based on what you call the free energy principle. So maybe before we ask you about the The first question that I think we wanted to ask is, if you were building a form of psychotherapy from the ground up based on active inference, um what would that look like? you know What would it be? um Maybe it'd be helpful to try to put it into layman's terms what your theory is. Right. Well, it's it's useful you close with your um
00:13:06
Speaker
if you had to build a therapy from the ground up, how would you apply this kind of, and what sometimes it is referred to as a basic mechanics that comes from the physics of sense making, the physics of sentience? And and I think in you know in that question, you know there is um an important premise that if I want to understand the processes that would relieve some suffering or some perceived suffering or would remediate in some way or mitigate some what you might or I might call a psychopathology. um Then i and I need to understand the mechanics of what's gone wrong and I have to identify the nature of what it is that has gone wrong and just asking these very simple questions leaves you now or at least motivates having
00:13:58
Speaker
a formalism and mechanics of belief updating. um I use belief updating because this is a common phrase and in Bayesian statistics that I have some um prior beliefs about the causes of my sensorium. And I now have some new data. It could be sort of you know what I hear you say or what I see. um And I now update my beliefs on the basis of seeing those data. And technically, this is updating from my beliefs prior to seeing the data, prior beliefs to after seeing the data, a posterior or posterior beliefs. So this is a really fundamental process, which means that me as a process
00:14:43
Speaker
can be adequately described as a process of Bayesian belief updating. And I think that's important because, of course, most of the phenomenology and most of the the expression of distress or indeed even different aspects of well-being is all about or can be framed in terms of belief updating, technically, mathematically, a kind of inference. So if you want to understand and how to make things better, then you have to understand what is wrong. And what is wrong is a failure of inference. So to me, that is one example of how simple things can be can be made through this very um dispassionate mathematical approach.
00:15:34
Speaker
that you and I mean by false inference quite literally, um the kind of false inference you would have done as a student in terms of type one and type two statistical errors, you're inferring something is there when it's not a false positive or a type one error, which of course is um a useful or one way of describing hallucinations and delusions. And also you can make false negatives type two errors inferring something is not there when it is and now we can start to think about neglect syndromes, dissociative syndromes in terms of a failure, a particular failure of false inference. So just putting though the the phenomena
00:16:15
Speaker
at hand within that frame means now you have to ask, well, how does inference work and how could it go wrong? And then you get to, if you like, the motivation for understanding the brain as an inference machine, as a statistical organ, a constructive organ that's always trying to Make the best inferences or inferences to the best explanation or prediction and I use prediction explicitly because I think that's one simple way to understand the um the Physics of this kind of plea for updating and implicit your Again in your question that we can summarize
00:16:59
Speaker
me as a process, me as a process of inference, me as complying with this Bayesian mechanics, simply as trying to minimize the surprise that I experience when I receive some new sensory evidence. And you can read this surprise in many, many different ways, and and they all make equal sense, which again speaks to the you ah but simplicity of or all of this.
The Brain as a Prediction Machine
00:17:25
Speaker
So you can read surprises literally the degree to which my predictions do not explain.
00:17:33
Speaker
my sensations in the moment. And that can be scored by a prediction error. Literally, I am generating predictions of what I would see, hear, feel, if my understanding of my belief about the world was correct. And I can compare the predicted sensorium with what I'm actually sampling. And then I just subtract, I can just measure the mismatch. And if the mismatch is zero, job done, I've got a good enough explanation for my world. But if it's not, then I can use that prediction error then to update my beliefs, to drive my Bayesian belief updating until the prediction error is resolved. It doesn't mean to say I know what's out there, but at least I've got a good enough explanation for what's going on. So that would be surprised as a prediction error. If you're a physicist, then you would call this something called self-information. And the average of self-information is entropy.
00:18:34
Speaker
And entropy basically scores the uncertainty or the dispersion of your sensory states. And your sensory states include not just what you hear and see, but what you feel, your your interception. So what we're talking about here is a sort of generalization of homeostasis. It's minimizing surprise in the moment means that I'm, on average, minimizing entropy or the dispersion of all my states, including my physiological sensations. So I'm just a homeostatic creature, and did I keep all my sensed or measured states within within viable bounds. And this applies and not just to physiology, to everything. The final interpretation, which is my favourite, is that the self-information or the surprise is the complement of something called the evidence
00:19:31
Speaker
technically the marginal likelihood for your model of the world, which means that by minimising surprise, it looks as if I'm trying to maximise, indeed search for evidence for my model of the world. um People in philosophy sometimes call this self-evidencing. It's a lovely notion, you know, that we spend our entire lives trying to avoid surprises Simply, or that would be one way of describing a self-evident creature, just trying to gather the right kind of information in order in order to um provide the sample evidence for my model of my lived world, or at least sensed one.
00:20:10
Speaker
Yeah, this this is really deep. And it's interesting because it's like an evolutionary model. Like people talk about entropy is always increasing. But locally, we build order, right? We create things and we have a model which may be off and then we ought to get feedback and update our model and then do better.
Therapy as Hypothesis Testing
00:20:30
Speaker
um If you think of an example, let's think of an example from therapy that's very familiar, right? Um, with interpersonal relationships. So someone comes into a therapist's office and Farah chime in here too. And they say, I want to send an email about this problem, but I'm scared. I'm afraid the person is going to get mad at me. I want to, ah I want to ask my boss for a raise, but I'm afraid they'll fire me. Or, you know, I walked past my coworker.
00:20:58
Speaker
the other day in the hallway they're usually friendly but they didn't even acknowledge me and then you have a model um does the person not like me did i do something wrong or maybe they're just having a bad day so with a lot of pathology. People make misattributions in cognitive behavioral therapies called personalizing i must have done something wrong they must not like me um and then a therapist may say something like well why don't you check. um ask them, right? Or they may say, well, why don't you try asking your boss for a raise and see how it goes. And the the stereotypical story is, you know, they try it and it works out and then they go, oh, I was off about that.
00:21:41
Speaker
Or there's data that antidepressants can change the perception of another person's face, where you know you may think someone's mad at you, but with antidepressants, you interpret it as, hey, maybe they're unhappy and it's not me. So Farah, how do you think about that with therapy? Because there's this, okay, Freud had the idea of the repetition compulsion, which is just doing the same thing over and over again and it doesn't work and then there's that phrase that people like to. um The definition is quote the definition of being crazy is doing the same thing over and over again and expecting a different result so.
00:22:16
Speaker
um I really sort of like the concept, you know, Carl, when you were sort of talking about people minimizing surprise, I think about from analytic theory, all of these defenses, you know, some very sophisticated and some very primitive that um maybe my patients are sort of employing to ah maintain a homeostasis because of the change can be very scary. But Grant, the examples that you gave when you talked about, well, let's say I'd like to ask my boss for a raise, um but I'm scared to, you know, why would somebody be scared to advocate for themselves? What sort of belief system do they have right now if they tried to advocate for themselves in childhood or a primary school, were they repudiated by an authority figure and did that then become their belief system? And now it's my job to help them update it.
00:23:14
Speaker
um So I really like the idea of thinking about things in this way. And I think sometimes our job as therapists is to really help our patients make sense at the outside world, to sort of break it down and metabolize it for them. And I think we all pull from lots of different theories and um kind of use whatever is most helpful and most available in the moment. And so I can really see how this would be very helpful in psychotherapy. So you would adopt a kind of a rational stance, but a rational stance based on Bayesian epistemology, Bayesian knowledge. And when I first encountered Bayesian um theories, they called it fuzzy logic in computer science. And in in a sense, it's more accurate because there's some there's a way to deal with uncertainty, whereas kind of classical logic is very black and white, Bayesian logic is is indeterminate.
00:24:14
Speaker
So with with that in mind, I'm curious how you think about the therapy process, ah given that you have two people together who are kind of forming, ah you know, a kind of um a computational system, if you will, which is meant to be more effective for the for the patient in navigating reality, and presumably, if successful, will allow them not only to update their models, but to update the way they update models. Yeah, well, I think you've you've provided the answer in in that exchange and those little vignettes and perfectly. So if you allow me, Alice, pick out some really key points you've you've implicitly made and and indeed explicitly made. First of all, if we are self-evidencing things, then we are compelled to provide explanations to ourselves for everything that we witness. So, you know, the example, that person didn't smile at me, she normally does.
00:25:08
Speaker
I have to find an explanation for that. Perhaps I've done something wrong. So it's interesting that we have to find these explanations just by being being here as a physicist, but as a philosopher, by being self-evident, trying to find and those explanations that provide a good account of my of my sensory experiences. In so doing, you are effectively becoming a good little scientist in the sense that you are now testing hypotheses all the time that provide the best explanation. And I use that analogy simply because of the the last statement, ah yeah that the the appeal to Bayesian Y probability theory, why Bayesian mechanics, and you're absolutely right, it's all about the uncertainty. So the average surprise is
00:26:03
Speaker
entropy, it is uncertainty. This is about not only decision-making under uncertainty but also quantifying the uncertainty, the confidence, the precision that you ascribe to one belief versus another belief. So to and to recognise that is very important because when you start to simulate these explanations and what they entail for what you are going to do next and indeed the consequences of acting, you know, asking your boss for a pay rise, you rehearse this. um So when simulating this kind of basic in mechanics and self-evidencing, notice that what we're talking about is largely making decisions about what to do. It's not just
00:26:47
Speaker
perception. And this is the active part of the active inference. It's basically inferring what is something like me most likely to do in, you know, in this particular context. And the way that we have to do this is to have a world model or an internal generative model of the consequences of behaving like this, or behaving like this, or behaving like that. And then you choose a one that is most likely, that is, if you like, delivers the least expected surprise mathematically minimises the uncertainty. So it would subsume both avoiding surprises such as getting fired, and which I would find very surprising given my self model is I'm very employable, but also I'm minimising uncertainty that I know more. I now know whether my boss likes me or not. So asking the boss
00:27:40
Speaker
does he value me um asking for a pay rise it also has an epidemic affordance is also gives me ah information that minimizes my uncertainty it minimizes my expected surprise it all comes back to self-evident maximizing evidence of minimizing surprise by acting upon the world in exactly the same way that a scientist does by performing an experiment so i use that analogy because of course people also think of perception. as hypothesis testing. And of course, if we're now in charge of the data that we are using to generate hypotheses to explain, then we also have to we can also now think of our entire lives as a continuous rolling experiment designed to together gather the right data to resolve
00:28:34
Speaker
are surprised or maximise the evidence yeah for for ah for our models. so they the Coming back to that nice example, you know should should I or should I not um ask my my boss for a rise? Well, I'm going to evaluate the consequences of that in my world model. and Now, if I have prior beliefs that being needy or asking for nourishment or support or nurture is um I'm the kind of person that will be rejected and I will come away more surprised and um more damaged than um than I would if I avoided even asking for this kind of valuation or this kind of validation.
00:29:20
Speaker
And then if I rehearse that, then I'm highly unlikely to commit to that particular action because it it ends up in a surprising and it will provide evidence that is counter to my self model. So if I've been brought up in a way I've had some traumatic experiences from carers or bosses in the past that I have found demeaning that are egotistonic. I am not going to engage now in that kind of epistemic foraging. because I think that's going to lead to a surprising outcome, so I'm going to avoid. But notice the pernicious problem and in committing to that kind of completely base optimal and egosynthetic behavior. The problem is, you'll never know if you were wrong. You'll never know if your prior beliefs were actually apt for the particular familial or work situation. Because because you've chosen not to go and perform the experiment, there is no evidence at hand
00:30:17
Speaker
to challenge your prior beliefs. So and what can think of many examples, you know, depression or compulsive disorder, social withdrawal, um catatonia, you could even argue is is one way of precluding the opportunity to get sensory evidence that will allow you to update your prior beliefs. So And I frame it like that just to answer your final question. What is the therapeutic dyad for or in the group therapy? I think it's a safe laboratory in which you can explore, you can perform these experiments and slowly render your prior convictions less precise and um explore other ways of transacting so that you can now
00:31:04
Speaker
revisit some entrenched, literally being stuck in a rut, having very, very precise probabilities. So you can then go out and experiment in the in the real world, ah you know, as a good self evidence and creature or scientist ah should actually do. yeah i I love that idea of looking at the therapeutic diet as a safe laboratory because we sort of go with this assumption, and and I think, Grant, you would agree that everything that's happening within the therapy office
00:31:35
Speaker
is also happening outside of it. So bids for connection, um things like secretly harboring resentment because you know they had to rush to their appointment and they really wish that I could see them later, but somebody else has the later time, right? And they're faced with that as as soon as they leave. So sort of expressions of anger or um and neediness or questions about lovability. If those things can get um revised or resolved within the therapy office, then it can really change how people function outside of it.
00:32:15
Speaker
And I think that that is really the hallmark of a successful therapy. I just hadn't thought of it in these terms. yeah Well, I think, you know, some you know psychoeducation is helpful, right? So you kind of will have patients who don't approach life as a natural scientist. And I think there's an idea that children are probably born as natural scientists. I remember like my earliest memory is looking at the sun through the trees in the winter and feeling a sense of curiosity and wonder. um So a lot of people aren't ready to take those risks and they may not even have that mental model that it's good to explore because they've been burned. Right. It hasn't worked in the past. And so they have a model. It's often related to developmental trauma, um though there there may be depression that restricts
00:33:02
Speaker
the exploratory urges or the the play circuit that Panksepp has studied, maybe impaired in some ways. And so in therapy, people will um need to kind of learn that they can take those tentative steps. And that's often quite challenging. it's it's It's kind of the first big move. And yeah, of course you can think of therapy as a developmental process. um What I was curious about is I think a lot of people don't approach reality in a methodical
00:33:35
Speaker
and Bayesian rational way.
Curiosity, Belief Updating, and Distrust in Science
00:33:38
Speaker
um A lot of people seek sort of magical explanations. And we run into this problem writ large, I think, in politics with disinformation um and the way people can get pulled into sort of um delusional belief systems that are supported by a large group, right? um So I'm curious what your thoughts are about On one hand, I'm thinking, well, by definition, the brain is functioning in this Bayesian self-evidenting way. It has to be by definition. But if you actually look at the behavior of people, they often appear like they're not following the free energy principle. And I'm wondering if that is true. Can people fail to follow the this way of most efficiently tracking free through things. And would that energy principle? be a source of pathology? Like a failure of the basic approach to reality that they're not, as you like to say, palpating the world. They're not seeking information the right way.
00:34:42
Speaker
Yeah, no, I think that's absolutely right. um Again, you've sort of answered your own question in in the way that you've set it up. But just to pick out, you you've talked about curiosity. i I think that will be a really nice way of summarizing what we are and what we do. um whether we're curious about how to get better, whether we're curious how how to be a better therapist, whether we're a curious scientist or we're just trail walking in an unexplored forest. All of this is is just evincing curious behavior which is in the service of resolving uncertainty or minimizing expected surprise. It can be no other way. So I just wanted to celebrate your your use of the word curiosity there.
00:35:24
Speaker
um And then you your your specific question, is it possible to exist in violation of the free energy principle? um I'm afraid not, no, ah mathematically speaking. I know what you mean. What you mean is that you your models may not be apt for the the world at hand. um However, there's something called the complete class theorem, which means that that um For any pair of loss functions and behavior, there are some prior beliefs that render that behavior completely base optimal or base rational. So what that what the complete class theorem says, which comes from statistics and the statistics of experimental design and the like, um um what that basically means is you can reduce the failure
00:36:16
Speaker
or the pathology of inference to people's prior beliefs and of course everybody has their own prior beliefs we in yeah and because by definition they are based upon prior experience and that is unique and private to everybody. So everybody's different everybody can brings different prior beliefs to the table It is certainly possible for your prior beliefs to not be um as apt for this so for this environment, usually when the circumstances have changed.
00:36:48
Speaker
so you know clear that our worlds are changing all the time, but you know under the free energy principle we are basically trying to minimise our uncertainty, maximise the predictability of the world, we choose the right kind of news channels that elude ah fake news, we talk to people who are like-minded to speak our language, we affirm our beliefs by talking to yeah to people like us, creatures like us. which um in itself maintains predictability and minimizes surprise. I don't get surprising answers. And we all do that uniquely because in so doing, we're doing it in an ecosystem where there are lots of other things like me around. And and of course, when um when we talk about the environment or the niche or the world,
00:37:36
Speaker
that has to be modeled correctly by updating your prize into posters to catch up with the changing world. That world is constituted by other things like me. So now we're in the other tricky game of attributing causality to the sensory consequences of behavior that I could have caused, or you could have caused, or she could have caused, or he could have caused. And so now we have we've got quite a difficult, which not every thing or every creature has to deal with, but we do, we have to and not only and work out how some sensory consequence, some news was generated, but we also have to um infer who
00:38:18
Speaker
generated that content and why did they generate that content and are they like me? Can I trust them? So we now have not only this epistemic affordance of go out there listening to other views but also we have to be very careful in selecting those people whose um but the sources in which we place epistemic trust. And of course, we can only infer, we'll never know, we can only infer that this person is a source. So that's a really tricky game. And of course, you know, we talked earlier about sort the therapeutic dyad, the very fact that the that we're in a joint belief updating context really adds a lot of pressure on getting your world models or your generative models right, by which I mean simply finding the right priors.
00:39:08
Speaker
So I think that you can say, um you certainly can say that the prize that this person brings into a therapeutic setting are not apt for her actual lived world at this point in this city, at this point in her life. And it may be not so much a question of rehearsing neurodevelopment, but slightly unpicking those and of inappropriate priors that have been installed in a different context, in a different stage in this person's life, um and that usually, certainly in my world, in a simulation world, in a purely theoretical world,
00:39:50
Speaker
basically means um relaxing these prior convictions and allowing thats the yeah ah person to explore other hypotheses and other um ah ah other options, as it were, simply by removing the prior precision from the way that they think that they act and the the consequences of their particular um that particular um behavior. So we come back exactly to where we were at the end of the last but i answer the last question. It's all all about providing the right opportunity
00:40:26
Speaker
to relax those ah pathological priors that have been installed through what kind of upbringing, perhaps in a traumatic situation or prima um ah or you traumatic um or inappropriate interpersonal relationships that do not cannot be migrated, the rules, and the contingencies that a priori now you have learned. cannot now be gracefully migrated to a new set of relationships, a new group or a new ah a new mean work family.
00:41:03
Speaker
um I could carry on about not because I'm talking too much, but I just wanted to also mention that yeah by bringing in the notion of fake news, we're we're now talking about collective belief updating, doing it together, and minimising joint surprise, and what does that and usually entail? Unfortunately, or fortunately, depending upon how what you mean by magical thinking, um magical thinking is an extremely good way to minimise surprise. It provides a really simple explanation. and so and I should qualify this. I know that we've spoken about this you off-camera as it were. but yeah So, evidence.
00:41:45
Speaker
can always be expressed as accuracy minus complexity. That's just a mathematical truism. um One way of thinking about this is that to find the best explanation for some data, some sensory consequence of something, to find the best cause is to provide an accurate account of why she didn't smile today that is as simple as possible. So minimizing the complexity is a crucial part of maximizing evidence. So there's no true explanation, because you never know what's actually beyond your skull. oh you there There is a best explanation. And the best one is the one that provides the most accurate account that is as simple as possible. And very often, the simple one is actually a magical explanation. And that is the best explanation. And if you're doing it together,
00:42:39
Speaker
then if you all are light upon the same magical explanation, job done. you know That's base optimal. Yes, it's like a shared delusion. So there's kind of neurosis, which causes this excessive complexity and unhealthy self-doubt. And then there's maybe this kind of healthy self-doubt, which allows you to relax some of those priors, those tightly held beliefs. And many times, right, people with anxiety, rely on inaccurate and excessive sense of certainty, which can kind of lock them in. And then in the group setting, they get locked in by these group dynamics, which are really problematic. And and especially nowadays, because there is a lot of distrust of science, which I think is pernicious um and insidious.
00:43:30
Speaker
And there's a proliferation of pseudoscience. And so if you're trying to sort of pitch people on being good scientists and there's this prior belief installed in the outgroup that science is to be mistrusted, it's not clear to me, and this is a little bit of a digression, but it's not clear to me how you get out of that Kind of valley that probability pet that you're in and how you achieve a different type of dialogue of course this comes up with war with conflict resolution and i'm curious if you have any thoughts about. About what happens when you get into that kind of um trap i do. um
00:44:13
Speaker
I'm just wondering which ones I'll tell you the thought that was out of my mind, which may not be the one that you you. Yeah, I'm thinking mainly of the mistrust of science, right? It's kind of like a being yes yeah like a perfect. It's a belief that kind of locks locks the chest up and then throws it into the ocean. um Once you get can once you convince people not to believe in being good scientists, it's a really problematic place to be, I think. Yes, but then I suspect one wouldn't use the word science scientist. I mean, um what I'm trying to say that we're all actually little ideal scientists and you know scientists and um as a popular meme. and It's not quite what I was inferring that you know we are
00:45:01
Speaker
just by our very existence curious about each other. We have to be. It couldn't it couldn't be any other way. and The scientist, the beam scientist, is just an extreme expression of that, just using a particular set of measuring devices that go beyond um the you know the the normal sensory organs using telescopes and microscopes or but even mathematics as a kind of ah measurement tool, um at least a language. ah but But the principles I think are right. um So are you suggesting that a distrust of science then will preclude people
00:45:40
Speaker
committing to exploring the their world, ah their political world, their their their social world, their interpersonal world.
Ruminations and Trauma in Current Contexts
00:45:49
Speaker
um It would be a great shame. A failure of curiosity. um In a sense, you could view it as an existential threat, but I don't necessarily want it to go in that direction. Farrah, I know you had some interesting thoughts about um ah about how to think about how how to use active inference in psychotherapy as well. That might be interesting to get back to the question about kind of noise or difficulty processing information.
00:46:17
Speaker
Well, one thing, Carl, that ah you brought up was ah this idea, um and I don't remember exactly how you phrased it, but I'll say that when people are functioning um or they have the behaviors such as ruminations, um ah dissociative tendencies, you you know those are defenses and habits that may have been from a difficult and traumatic time in their lives that really no longer serve them. And I think it is important to understand how certain things came about um to to sort of honor that, but then to also say, hey, this doesn't make sense for your world. And Grant, I don't know if that was
00:47:08
Speaker
you know answering your question, but it's... i was thinking I was thinking, so we're talking about developmental kind of influences. I was thinking about like ADHD or what people call ADHD. So let's say your brain intrinsically has some difficulty processing information for whatever reason. I don't have a concrete question, but I'm i'm curious, Carl, how you would how you would think about that um as it be approached in therapy. Right. ah Well, I think that there are probably two lines you could you could take. One ah is to actually understand the mechanics. yeah um
00:47:44
Speaker
and work with that as a therapist. The other trick is to actually explain to the client how you're thinking and why it's working like this. So almost like cognitive behavior therapy for self-evidencing. So just have even if ah if basic mechanics is a load of rubbish, it's still a really useful, magical idea that has and that provides an accurate account that is as simple as possible of the way that we do transact and ah navigate our world. So whatever you say or I say along these lines, I think could be very easily shared um with with with a client. But specifically, you're just taking something like potential disorders or hyperactivity disorders,
00:48:29
Speaker
come To understand the mechanics I think is quite useful here and by the particular mechanics I have in mind is something that you'll find in the philosophy the the philosophy and computational psychiatry literature. that describe the space in mechanics in terms of something called predictive processing or predictive coding. And the emphasis here is on something called precision weighted prediction errors. And I am patting that, you know, this particular notion of formulation of surprise very carefully because it's the precision, which is usually the bit that is broken. So um if you um imagine that
00:49:12
Speaker
all our surprise signals, all our prediction errors, which I repeated just the difference between what I predicted generating in an inside out way to the sensory periphery predictions of what I should be seeing in the moment and comparing them with the actual sensations that my sensory organs, my sensory epithelia are providing, then the prediction errors is the difference. And I said before that we can use that now to effectively vitiate or resolve those prediction errors despite updating to provide better predictions and explain away the prediction error. It's not quite as simple as that. It has to be precision weighted and the precision is the confidence you ascribe to this prediction error. So if you're a psychologist, this would be, do I attend to this particular prediction error?
00:50:00
Speaker
or not. So to use the analogy of fake news, the prediction error is the news. It is the newsworthy bit of all your sensorion, because it's a bit you didn't predict. So when I watch the 10 o'clock news on the BBC, for example, I'm watching it because I didn't know that it was unpredicted. So it's a prediction error is the newsworthy stuff, because you have to select the right channel, because it could be a lot of fake news. So you have to assign a certain precision to the prediction errors before you will let it drive your belief updating. And I repeat, the psychological analogue of that is attenuation. If you're decreasing the precision, you're ignoring something, which we do all the time when moving, for example.
00:50:46
Speaker
attenuating pain reception by rubbing our tummy by acting for example we do sensory tenuation or decrease in the position or we attend to it if we have high epistemic trust that basically can be read as predicting a high precision for this source of information. So I think that in nearly every case I can imagine, in certainly in computational psychiatry, one could argue ah and several neurological syndromes that rest upon a synaptopathy as opposed to a lesion or a stroke.
00:51:19
Speaker
and That's where the pathology lies. It's not in the predictions and it's not in forming the prediction error. It's in assigning it the right weight, in assigning the right precision, paying the right amount of attention to it in relation to your prior beliefs. and so What might you then do to try and redress the balance between the sensory precision and the prior precision? Well, you could give drugs. What kind of drugs would you give? You'd give the kind of drugs that um um either attenuate, um well, would ah redress the balance in terms of the excitability of new neuronal populations reporting prediction error. We're talking about classical so um and um psychiatric drugs. We're talking about psychedelics.
00:52:10
Speaker
in the visual system or the auditory system for example. So these neuromodulatory or drugs that affect the neuromodulatory transmitter systems are exactly the kind of effects that you would anticipate or read as setting the right position or the right attention. So one way to get out of the rut, so the rut literally defines something that's very very precise, a very precise prior belief. No amount of sensory evidence is going to move me around my Wellington landscape, my fitness landscape, or technically a variational free energy landscape, ah simply because I'm in a rut. So this is a very, very precise prior belief. And I'm going to need some very, very precise evidence to move my prior belief. But I also have to predict it's very precise. And if I got very precise beliefs, a priori, I'm not going to do that. Again, another aspect of this the vicious self-maintaining kind of behavior.
00:53:08
Speaker
So in conditions I think that are clearly um biological in nature, you know, that I suspect would be, so I'm thinking about severe autism, for example, yeah the severe, very severe autism. i I would imagine that you'd be looking for, you know, as a first line therapy, you'd be looking for the pharmacological treatment in conjunction with providing the right kind of sensory predictable sensorion. bearing in mind that these people have lost the ability to attend away from anything. They have to make sense of everything, because they can't implement the sensory attenuation. They can't switch off the channel. They're list they're not they're not filtering. Yeah, exactly. ah Exactly. yeah and so Yeah, so you could actually read this or as isomorphic with a broadband
00:54:01
Speaker
attentional filter model, absolutely. So they're listening to everything that could be fake news, it could be good news, they don't know, they have to explain everything. And of course, like the in so doing, they they will have to necessarily have ah incoherent, low level explanations that overfit all the sensory data because they can't ignore it, they can't they can't decrease the precision. So it's only you know people like you and me who have the joy of being able to ignore all the rubbish and find these simple hypotheses that make sense yeah having central coherent long-term narratives yeah that that allow us to form a good theory of mind and all those other very selfhood.
00:54:44
Speaker
you know, just a minute of selfhood, you know, the first step developmentally. But that does depend upon ignoring all the fake news, which doesn't depend upon having the right kind of attentional mechanisms, which does depend upon having an intact synaptic transmission of a modulatory sort. If you haven't, how could you remedy it? Well, anything that allows you to gain volitional control over your attention is going to really help. So we're talking mindfulness, meditation, some people, um some of my friends and colleagues in psychotherapy, you'll commit to this line of thinking, talk about sort of mentalizing homeostasis, mentalizing certainly those people who think that much of feeling, you talked about Panskeft before, and I'm sure you you
00:55:31
Speaker
and Mark Somes would wax lyrical about but about this, but for those people who think that, well, certainly Mark thinks that feeling in and of itself is just the precision. It's the thing that ignites the prediction errors and makes them do work, that enables the brief updating. Furthermore, there are lots of people but um who um I think that the interception is really very important when it comes to belief updating about self.
00:56:04
Speaker
because you've got lots of evidence from your own body, your heart rate, your blood pressure, your mesenteric nervous system, you've got lots and lots of sensory evidence about the state of you as an embodied person. any um um And of course, that brings with it the job of attending to this and that. And sometimes if you can't ignore certain bodily sensations, you can get yourself, you can find um
00:56:33
Speaker
you can find yourself in a place where your explanations for why i am I'm looking this shape, for example, or I'm getting these abnormal mesenterical gut sensations, you know, I'm obese, I've got stomach cancer. So these are very simple, base optimal, magical explanations for these sensations. The problem is you can't ignore them, so you have to have to explain them. Whereas if you can now teach somebody or skill allow them to become skillful in exerting volitional control over what they attend to, to mentalise these usually sub-personal processes,
00:57:12
Speaker
they can now actually visit other hypotheses. They thought, oh, it's probably not stomach cancer, it might actually just be a stomach ache, I might have eaten too much, or i've got my trousers aren't fitting properly. But you can only do that when you explore other options, which means that you have to carefully sort of attend to this or attend to that, attend to this, attend to the somatosensation, as opposed to the interception. um so but That's how I think you would you would you would you would start to use this kind of theory to ah appreciate
00:57:49
Speaker
why things like mindfulness and meditation and certain kinds of drugs could go hand in hand with, say, talking therapy and and the laboratory ah notion of yeah or ah a therapeutic dyad that makes sure that the content is appropriately explored. Whereas the context established by the precision of the confidence of the tenure attention or attenuation um is now brought to the surface by this notion of a precision weighted ah prediction error. Does that make sense?
Feelings as Feedback in Therapy
00:58:24
Speaker
It does. In some way, no matter what kind of therapy it is, it can attend to different types of errors or patterns, but one of them is is to become more open-minded generally. So in psychoanalysis, that might be through free association and the occasional well-crafted interpretation. I'm the therapist might say it seems to me there's a way where your thoughts keep coming back to the same thing no matter what is going on around you where it could be cognitive behavioral where they point out a cognitive error. um And it seems to me that feelings are an important source of information but it's really what we make of them that's important because they can be both misleading and they can also kind of lead the way i'm fair what are your thoughts.
00:59:07
Speaker
Oh, I um frequently will say, you know, feelings or feedback are not fact, right? So that feelings can be really valuable information um and something to integrate into our worldview or feelings can be fake news. um And so the ability to tease that apart and to make use of it in a helpful way, I think is um something that we should aim to do in any treatment situation um with sort of these interpersonal interactions. um You know, I felt really insulted when you didn't say hi and then maybe the person was distracted or maybe finding that simplest explanation or finding that, you know, the the most accurate one ah sort of relieves that burden.
00:59:58
Speaker
um and and allows us to sort of maintain functioning in the world. Yeah, so I wanted to ask a question about consciousness and then and talk about versus AI a little bit. My question about consciousness is, and I know everyone talks about this, is is whether you think it serves a function or not. um Does it make a difference to have consciousness? um In my world, everything has to serve a function. In my world, and everything that we witness has to be, if you like, and
01:00:31
Speaker
an emergent property of of of existing in some characteristic states. um and ah There are lots of stories about consciousness. I have no philosophical training, and but I've got lots of friends who who could give you an answer. We'd probably take them about out about an hour each. Mark Sones would be probably the best person in this particular conversation. and I would say, what is the purpose of having self-consciousness? If I was a virus, would I need to be have any selfhood in order to make sense of and act reflexively, given changes in the intracellular milieu in which I find myself? No, probably not. If I was an insect, would I need to um have a model of me as distinct from you? I suspect not.
01:01:26
Speaker
If I was a a pet or a person, I think I probably would need to to have a model. Part of my explanation for the world would have to include the notion, the hypothesis that I am a thing, I am me. You don't need to do that, and sometimes, obviously, with depersonalization, the precision of that particular pride belief just um you know goes away. It's quite, quite ah I imagine. and I'm told it's a horrible experience, but yeah we don't necessarily have to have selfhood, even of a minimal sort, but certainly um most of us, um I infer that most of most people do have one, so that must in some way pay its way in terms of i um
01:02:15
Speaker
must having self-consciousness must pay its way in terms of maximizing the model evidence, which means that it must afford a greater accuracy than the complexity that having a certain selfhood or self-consciousness entails in terms of more parameters and greater model complexity. It sounds like it might be quite important for relationships, the more complex social context
01:02:46
Speaker
Absolutely. um But you have to, so what sorry, the punchline of this argument, which is ah yeah is is ah is so effectively a Mark-Somes argument. um What does it buy me um in terms of how can I increase the accuracy of my explanations just by knowing I'm me? And usually the answer is I can now perform mental actions. What are those mental actions? They're exactly what we're talking about. They are attenuating and attending to various sources of information. So there's a deep connection between this uncertainty quantification, this fundamental part of inference, the precision
01:03:27
Speaker
uhm weighted prediction error story the neuromodulation story and now mental action and selfhood so in order for me to act. on the inside to attend to this and to attend to that, I have to be a me. you know I have to have at least a minimal agency or a minimal and a minimal selfhood. So and I think that would be, if you like, the purpose of consciousness. It's really providing the ability
01:04:00
Speaker
to infer me in this context and of course as you were intubated that brings along something which are probably necessary equipment when the world i am trying to navigate contains things like me and then we get back to the attribution that. aspect It makes sense to me. There's an idea in philosophy called illusionism that consciousness is kind of not real. But I think if you have the brain is burning glucose to be self reflective and aware of other people. And there's got to be a reason for that. The brain you know doesn't want to waste glucose and blood flow.
01:04:37
Speaker
um Farah, maybe maybe um we could shift over and talk a little bit about versus AI and genius.
Dr. Friston's Role at Versus AI
01:04:46
Speaker
And I also have a thought about how AI might be a useful prosthesis for people, say, with certain types of psychiatric or mental illnesses to function better ah in terms of augmenting some of these processes. Yeah, I think ah Dr. Fristen would probably like to hear about some of your ideas. I think they were kind of interesting. Well, I'd rather hear your ideas, Carl, but... Oh, well... So, okay, so I'll ah frame it. So there's, yeah I think, are you the chief sign scientific officer for versus AI? Is that right?
01:05:24
Speaker
yeah yeah um I am the chief chief scientist at Aversus AI. I've listened to some perspectives. Aversus AI is effectively a cognitive computing company. so that the um They don't make products. They make things that enable people who may may or may not make something you could buy off the shelf. So it's it's the aspirations of this company is in particular and all the
01:06:02
Speaker
all the academics and philosophers who are investing in this initiative is really to provide an enabling kind of software that could be deployed on the world wide web and specifically in fact there is a non-profit aspect to versus commitments via the special web foundation that are in conversation with the IEEE to set the next generation of standards for the way that we exchange information, the way that we pass around our fake news on the web, to make sure that that that it it now has the space and the machinery to hold
01:06:43
Speaker
the kind of sense making and belief updating of a federated distributed sort that we've been talking about. it And the emphasis here is very much on trying to identify those principles that underwrite natural intelligence as opposed to artificial intelligence. and ensure that they can be realized in the next generation of web protocols and the right with the right um protocols and the markup languages. They are very much like sort of um the the markup languages, the protocols we use when when surfing the web now, but now equipped crucially with the ability to score and quantify uncertainty
01:07:26
Speaker
which in principle now means you can now have agents, not just human agents like you and me, but also assistive helper agents, autonomous agentsifactual agents, agents that now have the same kind of the biometric curiosity. that we have. And here's the twist. and Once you've got this kind of message passing and exchange and interaction in play, you're now exactly in the situation we're talking about before, where we've got lots of people either dyadic or in a group therapy or in ah in ah in an ecosystem um in a lived world.
01:08:08
Speaker
who are all trying to jointly minimize their um the surprise, their joint surprise, their joint free energy. What does that look like? It means that they're all trying to predict to each other and they all want to learn about each other. They all become curious about each other. yeah So you know this is just a description of a sustainable ecosystem of intelligence that has attained or is attaining a um a surprise a joint surprise minima or a marginal likelihood or self- joint self-evidencing maximum, where the whole point of being part of this ecosystem
01:08:48
Speaker
um is to learn about the other participants in this ecosystem, so that you are now rendering your inputs and your outputs as predictable as possible, leading to yeah ah mathematically what would be a joint free energy minimum um from the point of view of dynamical systems, this would be known as a generalized synchronization. ah from the point of view of climate change, it would be the the thermodynamically the most sustainable way of doing this via things like, you know, the Janitsky quality. From the point of view of a user, it's basically a kind of, um you know, harmony
01:09:29
Speaker
in which everybody on this web is trying to make sense of, on this network, is trying to make sense of those people that it is connected to, ah technically the Markov brackets. So that's the ambition. you you know to To get an adoption and uptake, you have to provide proof of principle. And that will be in the form of effectively ah software that can be used at different levels. It can be used either as a professional and um application writer, you know through software development kits, right through to um
01:10:04
Speaker
sort of um user interfaces that somebody is not a software engineer could could actually use to to to to you know to build apps. Could they be used in therapy? ah there is That is certainly the aspiration. I'm not saying it's going to be this year or next year, but certainly there are people and teams of versus who are themselves mandated to look at the possibility of creating emotional artifacts. um and and that have not only the ability to recognize the emotional state, and the context, self inferred context of you, ah you know, and now we can get into a different conversation about what it means to
01:10:46
Speaker
have us have a model of me where I can be in different states and then I have to infer what kind of state I'm in. Am I embarrassed? Am I in love? Am I angry? Am I frightened? you These are all hypotheses that I have to infer actively and do all that mental action in so doing. ah But if you now want to have an artifact, you know a companion, for example, um in s Silico, Way that you could have a meaningful exchange with the artifact will also have to have a self model that means I am embarrassed I am in love that I am you know frightened whatever and also be able to recognize and infer that you are in this emotional state because the it is seeing the same consequences and
01:11:30
Speaker
that it would generate where it in that so it's true sort of mirror neuron and if you like phenomena but now at the level of representations about the very nature of self and self in the moment and what kind of self and what kind of person am I in in this particular moment So that's a long-term ambition. The short-term ambition and is really to provide a select um set of the user community with software development kits to just to see how far they can take this enabling kind of cognitive computing.
01:12:08
Speaker
I can see it's kind of the idea as a smarter world. There's this idea that the whole planet may develop consciousness. It's been called the noosphere. I sometimes call it global emergent consciousness. And I think there's a sense that we need to be smarter. People are familiar with large language models. How does this differ from a large language model? um Because I think you're talking about something that actually thinks. yeah Yes, I am, yes. and So I think the simple answer to that is, We've been talking about um agency throughout, whether it's the agency of our clients, or whether it's our agency, whether it's the agency of the community or the agency of the promulgators of fake news.
01:12:53
Speaker
um what is is it it What is it to be an agent? and that's you know very much The answer to that question very much under ah underlies my response to you in terms of what it is to be conscious. to To be conscious, you, I think, have to be an agent. What does it mean to be an agent? I have to do something. what and Furthermore, I have to choose what to do, which means I have to have a world model that entails the consequences of my actions. None of that is part of large language models. Large language models don't choose anything. They don't do anything. They're not agents. They're beautiful and constructs. I would imagine probably most people would say that you know the most beautiful invention of the 21st century, but they don't do anything.
01:13:39
Speaker
They just take past content and generate the next most likely one. So they don't make any decisions. There's no decision under uncertainty. I could give you a whole bunch of technical differences. um So the the other fundamental distinction between um the deployment of large language models and generative AI and what the the vision of a smarter world of the kind articulated by versus rests upon the fact we're talking about agents. We're talking about agents and ecosystems of agents that all have to act together and literally interact.
01:14:16
Speaker
Yeah, I can see the digital therapeutic application could be amazing. um For example, someone with ADHD, like that model could notice when it wasn't when the person wasn't noticing stuff and sort of and provide a level of mindfulness. And maybe over time, it would also have a training effect, depending I suppose on the plasticity. um I know we're running up against time. Farrah, I'm curious what's on your mind. Well, I'm just thinking about this as a potential, you know, when we get towards the end of our careers, what is sort of the next therapeutic frontier and what are things going to look like? And I know there's a lot of talk about
01:14:59
Speaker
psychedelics and different, you know, sort of, I guess they call it interventional psychiatry, TMS. And um just wondering how, you know, Carl's work is going to fit in. And I think it's just, ah you know, really inspiring. And I'm really grateful that you came and spoke with us today. If our listeners want to find out more about you or about your work, um where can they go? and Yeah, usually just Google my name and and that been fine i mean I'm trying to be a recluse because I'm quite shy, so so but I'm sure you can find bits about me on on the web somewhere. Certainly. that That would be an interesting conversation, kind of the future, um not just of the human sort of consciousness and intelligence, but also specifically mental health, because there is all this really amazing work with brain networks and the really the infancy of interventional psychiatry. And I think um
01:16:04
Speaker
the idea of hybrid intelligence, that a person and a machine intelligence would would be greater than the sum of the parts is sort of scary to some people, but also holds a lot of promise, but also in terms of targeting some of these therapeutics much more effectively. Right now, we're we're kind of... essentially blind. There's minimal data. So as someone who does TMS, like we can understand, I'm targeting the certain part of the brain to treat OCD or another part of the brain to treat depression, but it's really very crude. um Carl, any any any final words? but Well, well and just to endorse what but both of you have just said in terms of potential
01:16:46
Speaker
applications in the future. um yeah i I think this notion of um you know the hybrid intelligence, the your ecosystems of intelligence, I think it's great. ah But you could argue we've already got that with mobile phones. I mean, you know in the sense that you're a lot of but in the spirit of Andy Clark, a lot of our cognition has actually gone out there and has gone out there a long time ago, thank you, to everybody's benefit, well, largely if we ignore social media, to to most people's benefit. yeah and And and yeah like the the ah the role of things, and I can imagine sort of you setting homework for clients along with their um the in silico therapist, which could actually just be a large language model. You don't have the you don't need to have a thinking ah thinking ah therapist to yeah to to to fill in the the time between say weekly appointments or group therapists. I can see that being used very artfully, but also TMS, psychedelics,
01:17:45
Speaker
ah have you Have you had Robin Carhartt Harris on this? We've not. yeah Yeah, I think you'd enjoy speaking to him because that's one, and you know the use of psychedelics literally to and engender this open mind you were talking about. So he would interpret that, as would I, as basically rendering shallow these ruts, these very precise ruts, neurochemically. offering the opportunity now for an exploration or that you would have to therapeutically carefully navigate or hold the hand of the client ah whilst they're navigating through this opened up opened up landscape of possibilities. so I think you'd have an interesting conversation about how one could use TMS in conjunction with psychedelics. um There are people and in in Spain who are pursuing this at the at the moment, which yeah are
01:18:37
Speaker
So I would love an introduction. And the cell phone thing, I often think of that as the cyber cortex. And Andy Clark's work I read many years ago, I i read, I think maybe one of his first breakaway books, Natural Born Cyborgs. And so I agree, we're already there, um but it's it's really... We are not seeing the full potential. And and I think you can glimpse the future um probably more because of all the work you've done. So um thank you very much for being so generous with your time. My pleasure. I really enjoyed talking to you. Thank you very much. Remember, the Doorknob Comments podcast is not medical advice. If you may be in need of professional assistance, please seek consultation without delay.