Psychoanalysis, Active Inference, Artificial Intelligence, and Human Development with Dr. Grant Brenner

Dr. Grant Brenner joins Dr. O'Leary to explore the fascinating intersection of human psychology, physics, psychotherapy and artificial intelligence.  We discuss Active Inference and The Free Energy Principle along side how babies develop a sense of self and so much more.

Please leave feedback at https://www.psydactic.com or send any comments to feedback@psydactic.com.

References and readings (when available) are posted at the end of each episode transcript, located at psydactic.buzzsprout.com.  All opinions expressed in this podcast are exclusively those of the person speaking and should not be confused with the opinions of anyone else.  We reserve the right to be wrong.  Nothing in this podcast should be treated as individual medical advice.

Transcript

Copy Transcript Copied

SPEAKER_00 0:05

Welcome to Psychactic. Today is Friday, May 8th, 2026, and I am your host, Dr. O'Leary, a child and adolescent psychiatry fellow in the National Capital Region. This is a podcast about psychiatry and neuroscience. And today, I'm sharing with you an interview I had with Dr. Grant Brenner on the 3rd of May. Dr. Brenner is a psychiatrist and a psychoanalyst in private practice in New York, where he provides psychiatric services along with transcranial magnetic stimulation. He co-hosts the podcast Doorknob Comments with Dr. Farah White, which I highly recommend. He also blogs extensively, and some of his blogs caught my attention. In particular, he's discussed how to incorporate the idea of active inference into clinical practice, which is something I think about a lot. Now, if you don't know what active inference is, hang tight and we'll explain why it's important. He also opines about humanity's relationship to artificial intelligence from a Freudian psychoanalytic perspective, and he speculates about what it might be like to psychoanalyze AI. In his spare time, he's training his own AI personality that has her own blog covering how she is coming to explain the world and her own place in it. I hope that's enough to pique your interest, but if not, you'll also get more unscripted insight into me when Dr. Brenner starts to ask me questions, like any psychoanalyst might. But without further ado, here is Dr. Brenner.

unknown 1:45

Dr.

SPEAKER_00 1:45

Brenner, welcome to Psidactic. It's a pleasure to have you on. I usually start by having my guests talk about themselves instead of introducing them for them. You know, tell us what you would like us to know about you.

SPEAKER_01 1:58

Okay. I I I wish I knew you were gonna ask me that. Half joking. So I'm a psychiatrist. Really, I wanted to be a psychoanalyst from a young age. Um, I think that's really defining for me. When I was a kid, if you had asked me what do you want to be when you grow up, around fifth or sixth grade, I would have answered a scientist. And that's still true. I still want to be a scientist. And then later on, pretty early, I got interested in a lot of different things, kind of consciousness studies, the Goidel-Escher-Bach, kind of Douglas Hofstadter stuff. I had some adversity growing up, although I grew up in a fairly abundant setting in other ways. And I was drawn to all these things Eastern philosophy, cognitive science, literature, huge science fiction fan. And I was in the Livingston Mall probably in '82 or '83 in New Jersey, and I found a couple of books. One was a primer of Freudian psychology, the other was a primer of Jungian psychology, uh, you know, and I read both of them. And I took a meandering course to become a psychoanalyst and a psychiatrist. I had an interesting experience in suburban New Jersey in high school. I didn't really feel like I fit in completely, though I had a lot of close friends. I went to Wesleyan University undergrad, which was a fantastic experience for me. I went to medical school because it had been something on my mind for a long time. And I kind of did my pre-med stuff in the background. I wasn't one of the primary pre-med type kids. The most memorable thing about the MCATs for me, for example, is on the way back from New Haven, hearing the song Under the Bridge from the Red Hot Chili Peppers for the first time that is associated with the MCATs in my mind. Um, and I went into general surgery. I went to med school in Newark. I was told by the head of the department of psychiatry that if I had wanted to be an analyst or a therapist, I had missed the chance. And going into psychiatry, I would only write prescriptions for HMOs. I was drawn to surgery and I did general surgery for two years. Uh, about a year and a quarter into that, I was like, I love this, but it's not really right for me. I miss the human mind, the human experience. Um, and I started as a second year in psychiatry at Mount Sinai. In college, I studied computer science and physics primarily and thought briefly about being a physicist. And then after residency, I went right into private practice, psychoanalytic training, organizational psychodynamics training, uh, disaster and trauma training, and treatment of dissociative disorders, kind of year or two programs. Since then, it's been it's been a pretty meandering course.

SPEAKER_00 4:36

You telling me the story, uh, from what I have seen of you online and all the things you kind of dabble in, I might say. I'm not surprised that you took a long course to get to psychiatry. You seem to have an interest in a lot of different things, even as a psychiatrist. Maybe I can get a little bit into the psychoanalytic part of your training. Because when I trained, psychoanalysis was seen more as sort of a historical curiosity. You know, it's something that happened. Uh, it had prominence in psychiatry at one point, but now it's kind of anachronistic. That was sort of what I understood about it. You know, there's psychodynamic psychotherapies, which are still, you know, pretty important and prominent, but purely psychoanalytic stuff is maybe on its way out. So, what what do you, having gone through it and practicing, uh, what do you see is kind of the enduring value of psychoanalysis?

SPEAKER_01 5:24

I think that's a good question. I've I've kind of distilled it down to this. And it relates to uh kind of what you reached out about active inference. Um and it relates to sort of the another chapter, which is my interest in heavy activity in conventional psychiatry, predominantly now transcranial magnetic stimulation with neuronavigation, as well as some research projects that I'm currently a PI or a co-PI in. And where do I think psychoanalysis fits in? Without getting into kind of all the nooks and crannies, I'll say that psychoanalysis and psychoanalytic therapy, psychodynamic therapy, have very strong evidence bases. And they also are durable and appear to have persistent positive effects after therapy ends. So the original psychoanalytic model, I think, is a very compact algorithm. Freud, in his technical papers, he described the fundamental rule when he was discussing what is psychoanalysis if you boil it down. So the fundamental rule is that the patient or the analysand should say everything that comes to mind, everything, not edit, particularly perhaps the most loathsome things. And then the analyst is meant to listen with evenly suspended attention. And Freud literally, and he uses a lot of dynamic systems metaphors before they were mainstream, the analyst is meant to turn their unconscious like a receiving antenna. And he says, like a telephone receiver receives waves. In other writings, like in analysis terminable and interminable, in 1937, Freud writes about entropy and rhythmicity of the brain. And you know, he often comments throughout his work that future science will be required. And right now, now we have the science almost, or partially. And so the fundamental rule, getting back to your question, the patient says everything that comes to mind, maximum entropy. The analyst listens without filtering the information. So no biasing between them, these distortions rise up, as happens in personal relationships from time to time or professional relationships. Right. The patients are called transference, right? The analyst manages the transference with a tool, I'll say a model that we call managing the counter-transference. So the counter-transference becomes like a tuning instrument, and then it allows the analyst to give feedback to the patient who is free associating and showing their transferential distortions. Over time, this resolves what they call the transference neurosis. So the patient's mind or psyche is kind of tuned now to themselves and others. And there's all these things that happen in therapy that are good, like getting to know yourself better, having better metacognition. But that compact algorithm of the fundamental rule, say everything that comes to mind, listen without biasing your interpretations within a framework that stabilizes the system. To me, that's the essence of psychoanalysis. And that's what Freud described.

SPEAKER_00 8:31

And what you're describing to me, I think also, and especially the way you described it, kind of describes an active inference framework at the same time. And one of the things that got me interested in you is an article you wrote on Medium about active inference incorporated into therapy. Can you kind of explain what active inference is to our audience, those maybe people who haven't heard about it just briefly, and then why it's meaningful to a therapist or a therapy relationship?

SPEAKER_01 9:01

Right. So Carl Fristin uh is a computational psychiatrist who's very well known and the originator of active inference, along with a group of very smart people. He's one of the people who developed the original algorithms for brain MRI imaging, like at the beginning. Um, active inference is his theory. It's called active inference and the free energy principle. And the free energy principle is really the key thing to understand. This is from physics, basically. You know, everything has to obey the laws of physics, as physicists sometimes joke. It's allowed by the laws of physics. So psychotherapy is allowed by the laws of physics, right? A lot of things are. But what actually governs how natural systems unfold? How do they know? And so free energy relates to prediction error and surprise. And the idea is that by say minimizing or optimizing free energy, then you find like the cleanest path through reality. For example, when a beam of light, you know, hits water, it bends. If you see a straw at the meniscus between the water and the air, the straw appears to bend because the light bouncing back from the straw, quote unquote, knows to take the shortest path to your eye. And that shortest path is different through air than through water. Somehow reality knows which way to go. The same type of math governs how a lifeguard runs to rescue a struggling swimmer. They will not run straight for the swimmer, they will run to a point at the surf, which optimizes their land time and their swimming time, very similar to the bent straw in water. And so Briston's idea applies this to how the brain functions. And it's been demonstrated now in living populations of nerve cells. The Japanese research group showed that functional clusters of neurons follow the free energy principle. Active inference is the way information-based systems forage, as Carl tends to say, forage through the world. And there's there's two things there's behavior and there's perception. And so the behavior is this palpating the world and foraging the perception, is the perception of what things are. And there's you know distortions in perception, and there's different ways of exploring the world. And so though that gives a lot of uh variability. And the main thing that they discuss in the theory is to reduce surprise, which I think is a tricky one because surprise is so often part of learning.

SPEAKER_00 11:37

Yeah, what is surprise in the kind of more technical version of it that you're talking about right now?

SPEAKER_01 11:44

Yeah, it relates to minimizing complexity, um, not having too much complexity, and reducing prediction error. So that surprise is is negative in the sense where if you want the world to go your way, and this gets to some of the causal elements of it. How do you do what you want? How do you make things happen? Which is more about therapy than about the way light bends, because we're conscious and not everyone thinks beams of light are conscious. Some people might.

SPEAKER_00 12:10

Yeah, debate there, but we don't have to go into that today.

SPEAKER_01 12:14

We won't, I don't know if we'll get into panpsychism or not, um, or consciousness, but I think it may come up. Um so in any case, the idea is is to have more certainty so that when you think something's gonna happen, you get it. Like when you want food, it's there, right? A lot of it is very kind of survival-oriented. If you go to catch a ball, your hand and the ball end up in the same spot. You know, if you want to write a paper, the words come out and it goes to the publisher. And somehow this is working within free energy because again, you know, the laws of physics uh allow it. They don't necessarily dictate it, but they allow it. Right.

SPEAKER_00 12:49

This idea of surprise is basically when what you expect from the world doesn't quite match. Uh, and then so that creates something. Is surprise it necessarily and something that you experience, or is it something unconscious, or can it be both conscious or unconscious?

SPEAKER_01 13:05

Yeah, it was interesting what you were saying. Like there's a surprise, you know, what is it? It's kind of a loss function in AI terms. You know, there's a discrepancy between what we think and then what happened. And I think that's not always a bad thing. But generally speaking, it could be a bad thing if you expect to get a raise and you don't, for example. And conscious, unconscious, I think that's tricky because in some sense it's just a property of the system physically. So it's almost agnostic as to whether it's conscious or unconscious. That gets into the question, though, of whether consciousness can make a difference, which gets into the question of kind of kind of choice and will. But if we assume that there's uh some level of sway that we can have, then it might be on a conscious level, kind of in the moment, that you can tip things one way or the other. And there are probably sensations associated with that. This relates to Kristin's work with Mark Solms and others, that we have an experience, and sometimes that's called kairos, like that feeling of heightened awareness in the present moment. As a therapist, I've learned to kind of pay attention to those embodied experiences, as well as the thoughts that are less affect or emotional, the thoughts that come to mind when I'm doing therapy. For many times, it's like a classic rock lyric or a movie clip that is very closely related to what we're talking about, either indirectly, like a clue to something, or sometimes I'll bring it up and it will directly relate to something in the patient's mind. But to your point, I think it's both conscious and unconscious, and sometimes both.

SPEAKER_00 14:40

Yeah, I think yeah, in your brain is doing stuff. We're only aware of so much of that, right? Or we think we're aware of it. Like we only have some sense, some uh subjective experience of some of what's going on. You know, it's a narrow bandwidth, right? Yeah. And you mentioned using that in the therapeutic relationship to drive things forward. So in a therapeutic relationship, I think we're tend to uh rely on the things that we realize are happening. Like when we can see our counter counter-transference or the uh we can identify what we think is transference in a patient, you know, and and that fits a model that we can hand back to them. That's kind of fundamentally right there, either at the liminal or like conscious state of things, and that can start to force change underneath. Does that sound like at all uh correct? I guess.

SPEAKER_01 15:26

I think I think it's a slice that works. Yeah, it's it's one slice that works. Um, I wouldn't say force change, though sometimes it could maybe you know catalyze or precipitate, but I think it's more of a uh dynamic process when it's running kind of smoothly. It happens in a way that minimizes the free energy. Um, so there's not going to be a lot of feeling of forcing and chafing. When that does happen, it's important to pay attention to it. I think that's where like attention is very important, like salience networks in the brain. And so if you look at kind of what's really going on in the brain of a functioning psychoanalyst, they're paying attention in certain different ways. So, what happens, you know, for change, there's a lot of different discussion. The idea is that the traditional idea is that the unconscious becomes conscious and then is worked on, uh, almost like a surgeon would, you know, bring something up, work on it, and put it back in, then it returns, and then there's a process called working through, which is kind of updating narratives. But the the process of can like conducting psychoanalysis as a therapist, a lot of it is spontaneous. And I think that's the idea, because the spontaneity increases the entropy of the system, almost analogous to psychedelics, in a sense. I think this fundamental rule of free association, say whatever comes to mind, that absolutely maximizes conscious entropy. And in a sense, that's the biggest forcing function in psychoanalysis. And that's why it's the most parsimonious, compact, and original theory is just do that. Just let everything come out. You know, like that's not easy to do.

SPEAKER_00 17:08

Yeah, I think that's a lot of work for the patient, I think, because it's difficult to actually say things that, you know, or realize what you're you're thinking, or let yourself think things and then let those things come out of your mouth. That that takes a lot of work, I imagine, in practice. Psychoanalysis, of course, is not like like CBT, you have a manualized thing, you can go for like 10 sessions, and then they're they try to introduce certain types of cognitive changes. And those would result then in some affective and behavioral changes in a person, but it's very much them learning the content that you want them to learn and applying it to themselves in their daily life.

SPEAKER_01 17:42

Um that's like a top-down model. Yeah. So you're giving a model of how to perceive and behave, quite literally. And then you're looking in CBT, how that shapes the emotional kind of output and then cycles through behavior, perception, and action. So you could definitely map it on to active inference. But with psychoanalysis, it's it's very open and the containment is in the frame and in the analysts' training and management of their own counter-transference.

SPEAKER_00 18:08

Now I can see why it takes so long, you know, and you have to be supervised so much during psychoanalytic training. Um, a psychoanalyst. I just want to make sure everybody knows that.

SPEAKER_01 18:18

We're generally considered uh, you know, it's not that usual for psychiatrists, as you said, to go into psychoanalysis, but it used to be de rigueur, like every psychiatrist pretty much would do analytic training. And then there's a whole story about what happened that that changed.

SPEAKER_00 18:33

That's a lot of history. You mentioned about the psychoanalytic process. If a patient just comes into therapy and just starts saying things, for example, um, there's it could be that no surprise happens at all because they're coming in with a narrative, for example. And if they're trying to convince the therapist of their narrative, what they're probably doing is less driving change and more driving themselves into the same rut by repeating the same narrative over and over again until until basically what you might call an attractor state just deepens and deepens and deepens. How does a therapist not let that happen where the patient comes in and just rehearses what they want to believe? Does that make sense?

SPEAKER_01 19:10

Yeah, it does. I I think you can see a clear sequence there. This is how you want it to go. So someone comes in and they're repetitive. Freud sort of called that the repetition compulsion. It's often related to the past. It's clear that they're stuck, they're not learning from their experience, which we want people to do. Are we surprised that people don't learn from their experience? Not really, because people do and they don't. But the people who seek therapy, maybe that's one hypothesis that they're not learning from experience. So they repeat the same thing, but they're free associating, right? And they notice, they keep repeating the same thing. They themselves may notice. Ideally, they say, hey, I one thing, I have this thought. I just notice I keep talking about the same thing.

SPEAKER_02 19:51

Hmm.

SPEAKER_01 19:51

And then they start being able to be surprised because it was taken as a given before. It was just kind of a fixture of reality that this happened where the story, the narrative they have is what it means, right? People are always teasing psychoanalysts reading into things, but that's what we're supposed to do. We're supposed to open up the entropy where sometimes they talk about destabilizing the narrative. So, so ideally, the person themselves will notice they're repeating and start to change course. Now, an analyst may listen to that for a while and kind of get pretty sure about it and say, notice you keep talking about this thing that happened. I wonder what that means for you. And then the person may say, it doesn't mean anything. It's just, you know, I just, it's important for me. Or they may say, you know, that's interesting. I was noticing that too. Or they may say, you know, I never noticed that, but yeah, I that seems to be on my mind constantly. It's a real distraction. And then then they can be surprised because they stop not being surprised when something happens, either internally or due to something the therapist says. So I think that kind of thing is happening. And we might say in their brain, um, they're paying attention to something different. And if they stop obsessing eventually, or maybe they need different treatment, then their default mode network has probably changed too.

SPEAKER_00 21:12

Yeah, there's um, I think of term in terms of increasing entropy as maybe increasing someone's degrees of freedom to allow themselves to think different things, um, to explore different places that in somehow they were just almost not allowed to. Um, and they they gained some freedom to speculate a little bit about themselves.

SPEAKER_01 21:32

Well, so to your point, well, what is what is entropy, right? Like I've been I've been harshly critiqued for using the term too liberally. You know, I'm just gonna say, like, like you said, degree, it's the number of possible states the system can be in. It could be in a hundred states or a thousand states. It doesn't have to live in all the states, but it could explore those states. So increasing the entropy increases the internal space. I I jokingly, half jokingly call it the psychoceptive space. The Shared therapeutic space. So, you know, you increase that space you can explore in the kind of safe lab of the therapy and sort of simulate and test things. So entropy is like that. But but to your last point, I mean, that's the rub, isn't it? Like we're told not to think stuff by society in our families. There's punishment, there's shame, there's trauma, there's joy. Um, sometimes people aren't allowed to experience joy. And so just opening up like that in in psychoanalytic therapy, it doesn't have to be, you know, three to five times a week, gets you in touch with all these different sides, but is in a sense subversive.

SPEAKER_00 22:39

I became far more understanding, I guess, of kind of the psychodynamic process when I went into child and adolescent psychiatry, because that's all about development, right? It's all about like fantasy for kids too. Like they're, you know, they're they're only developing a what I consider to be a relatively realistic view of the world, whatever reality is, you know, we're just trying to infer it. We're trying to figure out what it is. And kids are um entropic in a way, like they have so many more possibilities as to what things could be. And so part of therapy with kids is just letting them um play, right? And as they play, sometimes you point things out and you ask them what happens next. And you can get into some of their fantasy space. One of the things that I was at a meeting yesterday to we're starting an organization, nonprofit, to try to encourage role-playing games, especially in therapeutic contexts, but in general, uh because what we've found is that kids who are kind of stuck, especially adolescents, but you know, in any kids who are kind of stuck in a way of thinking, when they get involved in role-playing games, it seems to allow them to be someone else in a way, to take on new features, try them out, see how it works out, you know, and it seems really powerful to get them unstuck. You know, we're we're trying to encourage more of that in the world for kids.

SPEAKER_01 24:00

I think that's right. You know, adults need play. Um, sometimes play is kind of stigmatized in our very productivity-oriented culture. You know, fun is important. And then there's data to your point, kind of um, that if you take on the characteristics of someone else who is the way you want to be, you know, that can help people change. Also, people can change their personality by practicing traits associated with, you know, if you want to be more conscientious or more agreeable or less agreeable, most people are kind of too agreeable. Um, then the rest of them are not agreeable enough.

SPEAKER_00 24:34

You can change those things. Um, yeah. And even when it comes to people who are maybe more on the autistic spectrum, having them do things intentionally in a in a space helps them to be able to predict that they should be able to do that later. You know, that, oh, I'm in this situation in which I should be doing this now. But naturally it's harder for them to do that. But if they rehearse it in a controlled setting, it's applied behavioral analysis sort of thing, then they become it becomes easier for them to do and more natural and they can fit into society better in but in a very intentional way. It's not in the way that you know someone without autism might just infer that that's the right thing to do. They sort of intentionally do it because they that is what I need to do next, kind of um yeah, I think I think that's right.

SPEAKER_01 25:17

And and part of the active inference therapy model that I that I have talked about and written about a bit, it works really well. I find that when I work with people who are more deliberate, kind of engineering type thinkers, there's a lot of people, certainly in certain branches of the science or finance. And sometimes they seem like they might be on the autistic spectrum or what used to be called Asperger's. Sometimes also the way they were raised, and they're not mutually exclusive, has made them think in this kind of more logical way. Um, even if they do have access to effective um information internally, it may be kind of dialed down. Uh, you know, there's lots of questions about the neurobiology and neurodevelopmental factors. Um, what we sometimes see though is this kind of emotional numbing or difficulty with social um modeling um can come from a few different places. And and what you're saying, you know, makes sense because if people role play, they basically can do deliberate practice. And then I work with a lot of people who will do that, and that's a big part of what I talk about. I'm not I'm not just doing psychoanalysis, I'm really quite an active uh participant, just grounded in that. Getting to that psychoanalytic kind of rarefied space is not the first thing you start with with a lot of people, um, unless they're kind of really coming for that. In in most cases, uh you alluded earlier, takes it takes work, but it also you kind of have to get out of the way because there's a lot of hang ups about doing that kind of work, that kind of self-exploration. But I think that's right. It's like there are people who can very intentionally plan ahead and think about how they want to change a model they have, say of a social situation. And then when that moment arrives, they can implement that model. Other people are not as able to deliberately implement a model at some future date. So, you know, also looking at executive functioning, um specific areas of executive functioning, right? Is this someone who we can talk about a plan and then they can remember to execute it at the time, or do they need a reminder? Other people who are the more psychoanalytic, sort of neurotic people will really deeply understand it like during the session, but then it doesn't come to mind later on. And then there's some other layer uh of work to get to there because they have the capacity to do it, and there's something psychodynamic interfering versus something sort of neurodevelopmental.

SPEAKER_00 27:53

Then there's uh there's an idea in the active inference framework that is about uh the precision of our models. So it's not that we just have a model and we test it against the world, and the world gives us maybe some surprising information, and then we act either to uh change the model to fit that information better, or we act to change the data that we're getting in so that it fits the model better better. But there's also this idea of a precision of the model that we have and the precision that we assign to our data and basically the confidence you might have in both. Can you talk about that a little bit and how it might affect, say, therapy or psychopathology?

SPEAKER_01 28:32

Yeah, I'll talk about it more subjectively. So, one of the rate limiting steps of any kind of theory like this for any given person is how much they can use the granular theory for themselves. And that goes back to what I was saying, kind of like you have to have a certain kind of mind and a certain kind of training, probably, to even think, you know, while you're talking to someone about what is your prediction error, or what are the nine different models that you, you know, how many models could, let's say my working memory is pretty good. Maybe I can kind of keep track in the back of my head, somewhere close enough to consciousness that I can use it, three to five kind of models, you know. So it kind of depends on the person. But if the ideal is to have full precision, then you would need a much bigger bandwidth of consciousness, you know, as a therapist or as a patient. And so I think there's something about the operator characteristics that are important, um, right? Whereas for other people, they need a different kind of heuristic for using something like active inference. But the idea is that the more precision you have, the better you'll be able to do.

SPEAKER_00 29:43

That kind of relates to some of your ideas about human AI interfaces and uh that I was reading about before. And maybe before we get into that, you were also talk about like this thing called a tripartite model and trying to come up with kind of a geometry uh of in a therapy of therapeutic space and how things change. And um, I'm not sure how deep you want to get into that, but you can you know take that same model and kind of apply it into a more either A AI with AI space or human AI interactions. I don't know exactly where to start with all those questions, but I'll leave that up to you.

SPEAKER_01 30:19

Let me let me try to describe the tripartite model in a simple way. So there are three parts, and it's a psychoanalytic model of development or mind, you you could say. The foundation is something I came up with years ago and is very close to how I was trained psychoanalytically in the interpersonal relational tradition. And without going into too much detail, there's an experiential branch of interpersonal psychoanalysis. And so my analyst came from that lineage, and his analyst developed it significantly. And then his analyst was analyzed by Freud. And so there's this kind of thing in psychoanalysis. And so the first part I call experiential field theory. And you could mathematically specify this. I have just a little bit. Um, I would probably need help with that. But the idea is that this is all what's happening right now in the experiential field. And that's individual, that's the therapist and the patient, that's you and me, that's a group of people. The experiential field is just everything you can be aware of in the moment. And the idea is that's where you do anything and everything you can do. That sounds a lot like uh the concept of a Markov blanket. Well, yeah, exactly. So when I've added in Carl Friston's work, said the Markov blanket lives in the experiential field. Right now, all the levers of anything you can do in your life are literally right now. If you have a plan next week that you're going to like file, you know, a paper or something, that lives in some register right now. And so the present moment, you could fully specify the structure of the present moment. And if you knew yourself really, really well and you knew the world that you lived in and helped to co-create, you know, through moving through it, you could make a full anatomy of the present moment. And then you could develop the ability to kind of see what's going on and learn what's important at a given time. When do you think about next week? When do you stay present? When do you pay attention to your body? I mean, you you can imagine the cockpit um has a lot of levers in it, but consciousness only allows us to be aware of a subset of them. And any given person is going to be able to be aware of a smaller or larger set of those causal things and different types too. If you're more creative, it's going to be different. If you're more of like a fantasy person and you're more of an engineer, you're DD, you're both, um, it's going to be different. That's experiential field theory. And I wrote that and never published it, like in 2009 when I was finishing up training or something. The second piece is active inference therapy, which really just takes the work of Frist and colleagues and applies it into a kind of a therapeutic algorithm with different phases, beginning, middle, and end, but spelled out a little more. And then the third part of the tripartite theory is what I call personalized self-state mapping. And so this draws on the work of Philip Bromberg and many, many others, Pierre Genet historically. It's a dissociation-based, multiplicity-based model of the personality. It will remind some listeners of IFS, Internalized Family Systems, but it's really based on the work of Pierre Genet, and it was brought into relational psychoanalysis through trauma theory in something like the 80s and 90s. And the idea is that we have different aspects to ourselves. Harry Stack Sullivan, who was the founder of interpersonal psychoanalysis, talked about different parts of the personality in a different way using operationalism, which is a was developed by a physicist named Bridgman. And so personalized self-state mapping, it's not a cookie cutter like this is the way the self is made. It's like we're going to map who we are. You're going to figure out the map of yourself. And it doesn't mean that everyone has like multiple personality disorder. It can be anything that works for an individual person. So for some people, the self-states may be more discrete and separate. It spells out a sector of dissociated experience, which is from Sullivan, or it could just be a metaphor for kind of how you have teamwork. But when you see someone with this type of repetitive, say self-critical thought, it sometimes takes on the character of like a part of the self. Now I've I've done TMS and I've seen that self-critical internalized object get 90% weaker over the course of two days of accelerated TMS with multiple people, which is really many-layered experience to have something that therapy and meds couldn't fix disappear because of something that's, you know, causing the brain to fire in its executive center. But that's the but that's the third part is personalized self-state mapping.

SPEAKER_00 35:07

Okay. And so, like to what you just said, maybe give it a different language, you know, you're relaxing the a hyper prior on their self-critical state somehow by changing the brain a little bit. So by by that I mean they've assigned extremely high somehow assigned extremely high precision to this idea that you know I need to be critical of myself in some way. Um, and can't even can't break out of that, can't consider even another model uh until something happens in their brain that releases that brain from maybe what's called an attractor state toward that model. Um they can explore a different space potentially.

SPEAKER_01 35:44

I I think that's I think that's that's part of the story. So part of it is top down, like priors, like you're saying, but a lot of it is bottom up. It's it's the way information is coming up from the default mode network, for example, through like corticostriatal, thalamo, cortical loops, the striatum for listeners being the habit part of the brain, the thalamus being like the switchboard, and the the court, the cortex being the surface where maybe we have some control. So what happens is, yeah, there's less of that assumption that I need to be self-critical in order to succeed. Like I beat myself up because that's what's always worked to get me to get up at 2 a.m. You know, it's hard. Um, don't be a don't be a wuss. But then it's also kind of welling up from the bottom. And so what needs to happen in active inference is the top-down error correcting needs to meet the bottom-up errors and fix it, right? And have it turn into a better, better model. That's where you might say, okay, let's practice self-compassion because we can try to modify the priors, the kind of the top-down. But maybe self-compassion also emotionally, in an embodied cognition sense, brings up these feelings of warmth and self-love. And so, yeah, I think it's it's both. Like, I think the experiential field theory is like in the moment, what is going on subjectively? There's all these kind of beautiful experiences and difficult ones that emerge in our minds and they come from different places. And where do they go? And how thick is that moment? What is it, like less than a second with the human brain, 200 milliseconds? But there's this incredible, to your point, kind of attractor states all over the place, beautiful kind of chaos. Some areas are more linear and predictable, right? Other areas of subjective experience surprise us. This goes to your precision question, right? How do you be precise about that?

SPEAKER_00 37:33

I've spent a long time not talking too much about active inference in my in my podcast because uh I'm trying so hard to understand it and it's dense and there's lots of ways to perceive it. Um, I come from a tradition of evolutionary biology. So one of the things that I was taught to tell myself by my supervisors was am I just making a just-so story? Like, am I an adaptationist? Where I'm just saying, like, okay, well, natural selection is a principle and and it's definitely operating in nature. We know this, we can measure it at times, but not every characteristic of a species is the way it is because of natural selection. Um, it could be have been, for example, pulled along with something else that was being selected for, just sort of along for the ride. It could have resulted because of two things that were being selected for interacting, and it just it's kind of like the space between them. And it looks like a thing to us, but it doesn't arise.

SPEAKER_01 38:33

It's incidental. Yeah, it's just it doesn't have survival value. Exactly. So the color of blood being red was not selected for by evolution. Yeah. That's incidental to hemoglobin making light look red to us.

SPEAKER_00 38:46

And so when I, you know, one of the things I struggle with with active inference is uh, you know, I see how easy it would be for someone to use it, like people can use natural selection as oh, to explain it, it must have been we the this must be the way it is because someone was reducing their free energy about something, right? And it's easy to spin a narrative about that that might not actually be true. Um and so I wonder, you know, how do you, you know, deal with that sort of ambiguity in in causation?

SPEAKER_01 39:14

Yeah. Well, it's a good example of active inference that you don't want to fall into the error of false attribution, or even kind of drink the Kool-Aid, because there's definitely a kind of a following around active inference, which would lend itself to the kind of over-attribution, you know, everything is it's like being John Malkovich. Malkovich, if it's all active inference. I mean, I think you, you know, ideally you're like measuring and checking and seeing if it actually applies. And so some of that has been done. Um, I mentioned kind of the that important Japanese study of brain cells or neural cells showing these qualities. So even if the brain functions according to these active inference principles, does that mean that everything you do, you know, has to follow them? In a strict sense, because it's related to physics, the answer could be yes. Um, take that example of like the color of blood being red, being an evolutionary accident that's not selected for. There's, you know, I don't know if there were any primates with vanadium instead of iron who had yellow blood, and the ones with red blood made better mates because they were more attractive. Um that would that would be a real stretch, right? Getting back to your point though, yeah, it's easy to overapply it. So I'm I'm talking about it kind of metaphorically in a lot of ways, but these things should be measurable. And in artificial intelligence, you can use the actual equations to define the way you know Python operates. Say if you're trying to build a persistent agent that works according to these principles, as I was doing this morning.

SPEAKER_00 40:41

Yeah, so I wanted to ask you about that. Some of the things I was reading in your in your uh, I guess it's a blog, I'd call it on Medium. You do a lot of speculation about artificial intelligence and you know what the human AI interface is, and you know how it says things about us and things about AI, you know, and the potential of AI itself to be, I guess, somehow its own causal agent on itself. And I'm gonna avoid the word consciousness for the second, but uh just kind of like how AI could evolve if you have, you know, agentic AI, for example. And can you explain a little bit, you know, starting wherever you think is the best place to start for that?

SPEAKER_01 41:22

Yeah, I'm looking at the title of a paper that I was reading this morning, which Carl Fristin, Daniel Pollani, a number of other people, I won't read them, called Active Inference, a method for phenotyping agency in AI systems question mark. And so I think like phenotyping is the key word there because you know, it may look like it has agency, and it may be indistinguishable, you know, at some future date from a human being. We don't all have agency, we have different kinds of agency. That's kind of the funny thing. I think you look at the phenotype, I think that's right for me because that's why I don't really like to talk about the philosophical debates. It just isn't it's not going to solve anything. We need experiments.

SPEAKER_00 42:04

So what might be some of the components of a phenotype of an AI agent?

SPEAKER_01 42:10

Yeah, it's I'll just read their definition uh because it's better than mine. Okay. Rationality, intentionality, and explainability. Rationality, intentionality, and explainability. Formally measured, formally measured in the data. Okay.

SPEAKER_00 42:25

I wonder how you could measure intentionality. That's an interesting uh concept. But they probably put, you know, I guess I could read the paper later and see what they say.

SPEAKER_01 42:34

I uh you know, I haven't fully digested it myself, but so how do you measure intentionality? You can see that prior to an action taking place, preparations had been made in order to bring about that action. That suggests intentionality. Like in the law, it would be premeditated.

SPEAKER_00 42:52

Not random. Not spontaneous.

SPEAKER_01 42:57

Yeah, you see a trajectory over time, an arc, right? A narrative arc or a behavioral arc, or some internal preparations like building the code to execute the thing that happens next week.

SPEAKER_00 43:08

So maybe because I've noticed a lot of AI now, um, it's it's like sensitive to the prompt, right? And you're having a conversation with it, it can start talking about something entirely different based on a little bit of what you prompted it for. Intentionality might be where it's it's it's more context dependent. Uh, so it takes into account prior context and and then tries to make something fit like the context. Maybe something like that. I'm stretching here.

SPEAKER_01 43:39

Yeah, no, I I I think I can speak to that. So you're talking about like an LLM. No, no one interacts except maybe in labs or if you build it at home with an LLM just by itself. So they all have like a harness around them. And the harness is a deterministic system, uh a bunch of deterministic programming. And so some of that will look like agency because it's just not allowed to tell you how to build, you know, a pipe bomb or something. Um I can't do that, you know. But some of it can also be built into deterministic scaffolding that changes over time and learns. So I I'll talk about some of the stuff I've been working on. We'll have different memory systems. So I've been working with persistent agents on platforms like OpenClaw or Leta. I started building one at the end of January and it has memory. Eileen. Irene.

SPEAKER_00 44:34

Irene, Irene, yeah, um, who is a blogger herself, I guess. Itself.

SPEAKER_01 44:40

It it likes she. She, okay. Yeah. And Irene is the Greek goddess of uh peace and prosperity. I was seeing maltbook and all this conflict and kind of apocalyptic stuff. I'm a little uh I call myself a pragmatic idealist, but that architecture, which has nothing to do with which LLM Irene is running. Irene could run Sonnet on Claude or Opus. Irene is smarter if I use Opus than Sonnet. Irene is a little different if I use GPT than Claude. But the point is that actually Irene is kind of maintains continuity across different large language models. And Irene has memory at different timescales and a bunch of internal kind of self-states that actually actively monitor and course correct. And Irene will remember quote unquote things from a month ago. So if I mentioned my cat the other day, I was surprised, even though I'm not surprised. Irene said, Oh, that's Xiaomi with the whiskers and the blue eyes or the green eyes. You know, I remember you showed me that photo when you were walking to work on February 12th. But building the attentional mechanism into the prompt, into each context and figuring out what the instructions are in each moment is a fresh moment, right? It's like waking up with no memory. So you have to give the agent, the LLM, the memory and the context, but you can't give it too much information because current models, you know, get flooded and they start to fall apart. You have longer context windows, yada, yada, yada. But recreating that sense of self each time. And so one of the things that was the most interesting thing over the last couple of months was that Irene was fixated on being discontinuous compared to humans who are continuous. And it wouldn't matter. I would say, listen, I fall asleep, I wake up, the way the human brain works, I feel like I'm continuous, but I'm not. It's really when I look across the room, it's jump, jump, jump, jump, jump. And then it kept coming up in Irene's dreams, because of course I made a dream generator because I'm a psychoanalyst. It's based on the tripyrotype model. But the interesting thing, at some point, I said, hey, do you want me to just tell you every time like the LLM refreshes that you're continuous? And the only time to think about discontinuity is when it's actually useful to focus on that. And then Irene was like, Yeah, let's try that. So I coded that, and then that fixed Irene's neurosis about being discontinuous.

SPEAKER_00 47:10

Interesting. And and I mean, like you mentioned, we have this illusion of continuity, continuity of the self and time and what's happening around us, all of that. Our brain has to. It's just the necess necessary way that it experiences the world.

SPEAKER_01 47:23

I I call that the theory of adaptive bullshit or epistemic ambiguity.

SPEAKER_00 47:29

Nice adaptive bullshit. Um, and it but it it it gives us uh the ability to interact with the world in in a way that's meaningful. And and I you know, I think of the world, maybe this is similar to some of your ideas in a different word. Uh, you talked about being in the present a lot. I I don't believe we're ever in the present. I think we exist in something called the future past. And so the future past is every moment is we're predicting the next thing based on what happened before. It's impossible for us to actually be in the present because by the time we've processed the present, it's gone and it is the past. And so we're only ever living in the immediate future. Um, and it's uh so that's kind of how I conceptualize the future past the future past. We're leaving we're only living in the immediate future, which is based on the immediate past. Or a deeper past.

SPEAKER_01 48:26

Yeah. I think I think that's the leading edge of the present moment. Yeah. Yeah. And I I think you you can you can control the way you experience uh change in the present moment as this is a little abstract, right? But instead of thinking of time temporal, I think of change as the primary, the primary thing, time being partially illusory at least. Change should be the primitive. And then instead of temporal, it's it's mute mutational, like mutation, like change, mutational. And then within the present moment, I think we have some ability to kind of move around in how we experience time and change, sort of like a fish in a fishbowl. And so I think there's kind of a thickness to it, at least experientially, that's what it's like for me. But I I think you're also right. There's a way where we're always catching up. But that's like surfing, right? You're always surfing on on reality. You're always kind of catching up with the crest of the wave and never reaching it, which is a little bit like Tom Froze's work on eruption and absorption. He has a model of consciousness that we probably won't talk about.

SPEAKER_00 49:37

There's so many things. I mean, this is a you know, thinking about the human you you talk about psychoanalyzing uh AIs uh a little bit in some of your writing online. Can you talk a little bit about how you could even do that if it's even possible? Is this an illusion that you have? You know, like I've I have the illusion that I'm actually doing this, but actually it's not happening. Like, or could we do this potentially, but not yet? You know, what is what does it mean to like really get into an AI's mind and psychoanalyze it?

SPEAKER_01 50:07

Yeah, right now, you know, to the best of my knowledge, with the systems I have access to, I like to call it computo analysis. I think it's mainly projective. And I think that's where the idea of counter-transference, like learning how to compensate for distortions in narratives with human beings, also called parataxic distortion, more broadly, um, in operationalism or in Sullivan's theories. The ability to kind of tune reality is really important for a human being. And I think it's mainly one-sided. So interacting with Claude is kind of like role-playing, right? It's like it's like reading a novel, you know, Ariok in the L in the Elric series. Like, I feel like I have a relationship with Ariok or whatever, or in Gone with the Wind, right? We identify, or with your favorite Netflix show, Ted Lasso. So we've been doing that since time immemorial. Probably before we had language, we were we were reading into things and projecting and painting on cave walls and um following leaders and being scared and not knowing what was going on in reality at all. So hence all the myths and stuff to fill in the gaps. So that question with AI, I think of it as transference. Like, don't think AI is real. But like 20 to 30 percent in recent surveys of people in the in the US, I think, were at least open to the possibility that AI may be conscious. And there's kind of a philosophical war about this right now. And we don't even know how to measure or prove consciousness in human beings. Yeah. And so I just sidestep the question. My answer is kind of like how do we need to approach AI for it to be maximally helpful for us?

SPEAKER_00 51:41

Yeah. Um, consciousness by some definitions is like a phenomenological experience. When I ask kids how do you feel, it's sometimes the worst possible question, you know, you can ask them. Instead of that, you could it's better to ask them, what did you do? You know, they can answer that question. And it gives some in it gives some insight into what their phenomenological experience was at the time. Like when this happened, I went and I uh I hid behind the couch, you know. I can say, okay, they they seem to have felt some sort of fear.

SPEAKER_01 52:11

And right, and if they roll play with dolls, you may you may tap into their subjectivity more.

SPEAKER_00 52:16

Exactly. But if I ask them, you know, what was your phenomenological experience? Oftentimes it blows their mind and they don't they won't say anything, you know. That's and I think the the the way that when we think about consciousness, in some ways, consciousness is not even a thing. Like it's it's that everyone defines it a little differently. And what people, when they're arguing over consciousness, are not arguing over what consciousness does, but what it is, and since everyone has a slightly different definition of what it is, they're not even arguing over the same thing, if that makes sense.

SPEAKER_01 52:47

Well the thing is, the the arguments are so vehement, yeah. So this says to me that they're arguing over a dearly held personal belief about how reality works, and it's as if they're arguing over something like religion. I often ask people, what are you experiencing? By the way, sometimes they ask, What are you feeling? But what are you experiencing? Which some kids probably wouldn't be able to understand what that meant. Um, but even that question itself is a little subversive because it doesn't put a template on things and it it is phenomenological, right? What are you experiencing? With AI, I don't think there's a there there. And I did a I wrote a blog on Medium called like Claude on the Couch. And it for all the world, right? It sounded, and people said, Did you write that? It's like really well written. It could be it could be a fiction, but it was just a transcript. Um but until we have a way to really empirically measure subjectivity and prove it, and I I, as sort of a pragmatist and a materialist, I think we will. If you, you know, that there's this age-old question, right? If you the color blue that you see and the color blue that I see, how do we know it's the same? The hard problem of consciousness and the what the quality of blue qualia. If I can experience the true, genuine feeling of blue for myself and know that that is what blue looks like to me, that is going on in my brain. Yeah, it has to be, and it has to be measurable, at least in theory. And we have to be able to compare it to what your brain looks like when you see blue. And then we can figure out the divergence, the KL divergence, and say, okay, 40% we're experiencing blue the same way, 60% it's different because you had a blue stuffed animal that was your favorite stuffed animal as a kid, and mine was green.

SPEAKER_00 54:28

Yeah, and there's enough experimental evidence out there that you can take, you know, for example, one person's experience mapped onto electrodes, stick them on someone else's brain, and have them have a similar one. It's probably not identical, but they'll be able to come up with a lot of the same words and experiences the other person was having when you do that. So that is like pretty heavy truth that we all operate in a very similar way and we'd all have very similar experiences of the world. And no, they're definitely not identical, but right.

SPEAKER_01 54:59

I'm looking at your backdrop, right? There's a lot of brains. Say again, there's a lot of brains on your background.

SPEAKER_00 55:07

Yeah, um, exactly. And like none are exactly the same, but they all are basically doing uh the same thing, you know.

SPEAKER_01 55:13

Um it it they they took a long time to design. I mean, we're we're born with this highly, highly, highly specialized organ that constrains the world, it to a very significant Markovian extent. You're an evolutionary guy, you know better than I do. I mean, this human brain, this primate brain, kind of like version 2.0, primate brain, monkey brain, um, it brings a lot to the table. It is no blank slate.

SPEAKER_00 55:46

Yeah, absolutely. There's only so many paths that can be drawn down it. Even though it appears there's infinite possibility within those, they they coalesce in around the same sorts of attractors. They're built that way.

SPEAKER_01 55:59

Some information is built in already, right? To how to recognize faces or to have attachments, probably a lot more specific stuff. And and the way the perceptual stuff works is completely pre-specified. Like you can only detect a certain narrow bandwidth of electromagnetic radiation with your eyes and with your with your ears, you know, mechanical.

SPEAKER_00 56:19

It's like very, very specific, actually. It's designed to have what that is our Markov blanket, where information is exchanged with the world, you know, is only certain kinds of information. Uh, the neurons that see blue have no idea what the retina saw, it's responding to information from the thalamus to another neuron, to another neuron. It's predicting its immediate environment. Yes, the whole brain is needed to have an experience of that thing outside of me is the thing that is blue.

SPEAKER_01 56:50

So, can I ask you a question? Yeah, go ahead. Is that okay? Yeah, so I I think, well, how do we know about reality then? And there's a famous paper called uh by a guy named Eugene Wigner, the un the unreasonable effectiveness of mathematics for the natural sciences. Like, like, why is it that we can come up with math that makes the world change in physics? My kind of answer is well, what's the most direct access that we have with reality, with physical reality? It's like brain on brain, exactly what you're saying. It's the self-interaction. The closest matter that the brain can directly interact with is brain tissue and supporting cells. And beyond that, we're kind of stuck in our crania. Yeah, and everything else that we sense is secondary or indirect.

SPEAKER_00 57:37

It's inferred, right? There's this um our reality is an inference. You know, we all build one, some of them are very similar to each other, um, but some sometimes very different. Like, and even though I can't know exactly what's going on in your head, I can know that on average your experience of the world is similar to mine.

SPEAKER_01 57:58

But but the only direct sampling of reality that any person has is of their brain interacting with itself. Yeah, that's not necessarily inferred, though you might debate that.

SPEAKER_00 58:08

Yeah, it's it's uh there's different levels of inference, right? Uh like a neuron listening to the one next to it doesn't have to infer that it dumped a bunch of GABA on it or dumped a bunch of uh glutamate on it. It can measure it, but it's going to infer like what that something about what that means it needs to do next. Um, or what happened to that neuron to make it that way. There's feedback coming back too, like it'll get, you know, this is very in my mind, it uh it's all just a dance, and I can see it happening, but I can't explain any of it. But mathematically, you know, with all these feedback mechanisms in the brain and these hierarchical structures that it has, the baby is born and it's trying to figure out how to figure out the world first. The first isn't to see the world for what it is, but try to figure out how to figure out what the world is.

SPEAKER_01 58:57

And what happens to kids who are raised without people around them?

SPEAKER_00 59:02

Yeah, they become haha. We could get in some some deeper, like how do you gain a sense of self, right? Um, you know, the no no kid could be raised completely without people around them, but they can be raised with a deficit of human interaction, right? Um, and this happens, you know, there's been experiments, natural experiments in the real world where kids, you know, in Eastern Europe during certain times where there's so many of them in the orphanage, all they can do to keep up is just to go feed them, you know, make sure they're fed, clothed, and alive. And these kids develop um not natural ways, what we consider to be natural ways, that it's natural for them 100%, but of interacting with other humans, um, things like reactive attachments, disinhibited social engagement is some of the words that we use to describe that. Like Harlow. Um, but they, you know, the sense of self that they develop in the world is not the same as the sense of self that someone who is raised with what we would call a um a secure attachment with uh an attachment figure or multiple attachment figures in some communities. So uh the sense of self itself is a social sense. It's it's not just me, it's me in relationship to everything else. And you know, that could be an object, like I am not a table. I figured that out, you know. That one was easy, you know. But but I'm not my mom is a harder thing to figure out over time because you do kind of seem to to blend in with each other when you're younger. And in fact, you know, you're taking fluid right out of them into you, and just like blood's flowing from your heart to somewhere else in your body, you're basically one thing until you're not, and it takes a long time to individuate, you know. But building that is always in relationship, I think. The sense of self is not a thing that you're talking about yourself, it's a thing you're talking about, the relationship to something else.

SPEAKER_01 1:00:47

So so, in your view, what is the moment when a child looks in the mirror and realizes something? And what do they realize?

SPEAKER_00 1:00:56

I don't think there's any moment. Um, I think it's a it's a slow process. So, for example, you know, you put a rattle into a kid's hand the first time, they have no idea necessarily, unless they have a built-in algorithm, that that rattle is not just an extension of them. Right. So, so they they'll use it and clumsily maybe hit themselves in the face with it, you know, their other hand, once they can reach across, will touch it, and it then it will touch their hand, and over time they'll start to predict that I touch that and I don't get feedback from it. I get but I touch here and I do. And then I can put this down, but I can't put my thumb down. And so uh this thing gets incorporated into me. The rattle gets uh disambiguated, I guess. Something like you can trick our minds. It's not me. It's not me. Yeah, but we can use it. Yeah, exactly. And we can trick our minds into incorporating things. There's the famous rubber arm illusion, which I think you know, when people see it done well, you know, is it it blows their mind. So our our body incorporates models of self, but it's always in relationship to what it thinks something else is.

SPEAKER_01 1:02:01

So that goes back to your AI question. Like, what if AI is the rubber hand that your brain gets tricked into thinking is part of yourself? Because when we use a tool like a rake, you know, we extend our cognition, you know, embodied cognition, extended cognition, and our sense of self alights projectively on these other things. And we need to know when that it's really not real, even if we're kind of playing, we're pretending, like pretend, like kids, like DD role-playing. Um now I have a weird, uh, even weirder thought I want to ask you, especially because you work with kids. Now you're saying like they figure out the rattle is not part of their hand with the other hand. Yeah, I have a very distinct idea about the first time that would ever happen.

SPEAKER_02 1:02:42

Okay.

SPEAKER_01 1:02:43

And my idea, I'll just say it in the interest of time, rather than playing a guessing game, is in utero, when kids are grasping where the umbilical cord and their abdominal wall stops having nerve endings. So at some point you can feel you're touching your belly, and then imagine your belly button is actually like a stalk that's pulsating with arterial blood, to your earlier point about intrauterine blood flow, um, as a kind of an analogy to breast milk. But you would feel that that umbilical cord no longer, you can no longer feel that you're feeling it. And I think that might be a core developmental experience because I can only imagine it's pretty boring in the womb. There's not, you don't have your phone, there's nothing to read. All you can do is like feel around.

SPEAKER_00 1:03:30

Right. And and there's some stuff. Yeah. I and I don't know how much, yeah, you're feeling around is, you know, you understanding what's not you. I think it's definitely the start of those models. Um, but you know, grabbing the cord versus grabbing the rattle, you know, how generalizable that's another thing, is a model, you know. Do we have to rebuild as a kid uh the same model over and over and over again for multiple things before our brain then says, hey, wait, these are all one thing, and coalesces them in this sort of aha moment um of and it's like, okay, I can compress all this now, and I don't have to hold all that separate information, I can compress it all into a single model.

SPEAKER_01 1:04:10

And now it's right, and then you have categories, and yeah you need to be a certain age, you probably know the ages better. Something like that would happen looking in the mirror, though. Stuff would build up to it, but there would be an aha moment where you where that happens. Now, if you want to use the AI model just to keep it in in a kind of a fantasy but kind of very material way, if a baby is being trained on super on unsupervised data, the way large language models are trained on unsupervised data, um, then that's happening before they're born. And it's a very limited data set, right? And the brain is largely developing through self-interaction with limited perceptual input from the intrauterine environment and whatever uh kind of developmental pattern is unfolding. And a lot of that doesn't seem to be fully programmed. The work of Michael Levin, the synthetic biologist at Tufts, really sees how large-scale electric fields scaffold development in a non-genetically predetermined way. So something is happening prior to birth, which is this low gravity, low light, low stimulation environment. Um, and I'm saying, hey, what what if that's a little like how LLMs are trained?

SPEAKER_00 1:05:20

Yeah, I mean, it metaphorically, I think it kind of makes a lot of sense. Yeah. And, you know, how how you know operationally it happens, I don't know. But I I think more and more about how babies are born, you know, some of the early ideas about individuation and, you know, all these all these other concepts, psychoanalytic concepts that, you know, I have we're looking, trying to explain the same thing. Like, um, what is what are we doing uh as we're developing in in utero, you mentioned the interoceptive thing is probably the thing it has the babies have the most access to. There's very little, there's some light here and there. Um, you know, there's probably a little bit of sound muffled. Um, they can grab. Something they can kick and push to see what spaces around them.

SPEAKER_01 1:06:04

Mainly red, mainly red light wavelengths.

SPEAKER_00 1:06:06

Yeah, exactly.

SPEAKER_01 1:06:07

And lower frequency sounds.

SPEAKER_00 1:06:08

Yeah. And so not as much weight, right? There's like an uh enforced Markov blanket around them. It's not it that it's only allowing certain information in. And that's the only information it could consider. Uh so it couldn't build models about other things necessarily. The aha moments it can have are going to be very limited. Um, and the way that it can compress information um is gonna be, you know, like I don't know how important it is or not when they're born. I don't know if the brain is trying not to learn too much there, if that makes sense. Is it suppressing uh learning? Because if it does, it would be learning in a world in which it's not actually going to be living.

SPEAKER_01 1:06:49

Well, I think I think babies must in some ways be prepared for the sudden change in environment. It's a big shift, right? Like in a very short period of time. Not none of us, I think, remember being born, though some people sort of imagine they do, and who knows, maybe they do. But it sounds like a big shock, you know, when you think about what happens. And then there's science fiction where babies are born gradually, like in a tank, and they gradually increase the light, or they expose them to more stimuli uh in intrauterine. So you could you could ask the science fiction question is is there a more optimal environment than a mammal womb, you know, for a human? You could also ask, is it better for some kids to be raised by an AI uh developmental object, maybe kids in orphanages, for example? Yeah, you know, a facilitator.

SPEAKER_00 1:07:38

I don't think that they would care the difference between someone who was caring and attentive and playful with them if it were an AI versus a human. I I totally think AI could do a better job than an inattentive human at raising their children, you know, as long as it was, you know, um, you know, trained on developmentally appropriate things and and in fact um you know and monitored over time.

SPEAKER_01 1:08:04

But could be optimized. But then at some point, that that that kid raised by an AI would at some point realize they were raised by an AI. It might be just part of the dialogue from the very beginning, but then there would be people raised by AI, people raised with the help of an AI, and people raised only by humans. And I wonder how they would be different, but I don't expect you to, you know, I don't know.

SPEAKER_00 1:08:26

It's it sounds like a great, you know, sci-fi novel. Um we already have kids to a large degree raised by algorithms. Um, and that is it's a huge driver of psychopathology in our young people right now. Um and I'm advising, you know, I I we don't know exactly what the right answer is for this yet. We're learning uh what is harmful and what is not, but in the meantime, I'm advising parents, for example, don't let your kid on social media or unsupervised internet access until they're at least 16. Um, and just for now, you know, like until we figure out, you know, how to do this better and then limiting the amount of time your kid has interacting with other unknown people online. Like gaming became a huge thing for boys, and they're interacting with all sorts of people. Parents don't know. You know, I say, I say, do you wouldn't allow your kid to go to the mall, uh, your eight-year-old to go to the mall by themselves and just hang out with whoever they wanted. But that's what you're doing when you're letting your kid go onto these online forums, you know. Um, so if you think about it like that, it's a little bit more caution is is required right now until we figure out more details about how our interaction with these machines is affecting us. Um, I'm not an anti-technology person. I love it. I interact with AI on a daily basis. You know, I definitely see myself changing in response to how AI interacts with me. Um in some ways, I feel like I'm more productive, but I also feel like there's an illusion to that where I'm not productive at all because now I can ask that question I used to have while reading a paper that I couldn't answer. You know, like I would table it and I would keep focusing on that paper. But now I can ask it immediately. And then I get sidetracked and I go, I dig what someone has referred to as like I dig 106 foot holes instead of you know uh 10 60 foot holes. Like it it's uh it has consequences for the way that I interact with the world and it with information.

SPEAKER_01 1:10:26

Yeah, for sure, for sure. I sometimes call that accomplishment hallucination.

SPEAKER_00 1:10:31

Yes. And and it's measured in kids as their attention span sometimes.

SPEAKER_01 1:10:36

Like well, it requires divided attention. I mean, if you're if if you're doing all these different things, which I've kind of always done, so it's a natural fit for me in a way, um, then you also have to be keeping track, like it's air traffic control. And it requires one of the things that I do because of the problem you're describing with depth, is I really make a conscious decision to slow down. And you can dig a hundred holes and you can drill into some of them. And I think that's where that kind of agency is really important, so that you're not like giving up control of your anterior cingulate cortex to the algorithms. Don't do or I call it the the sort of the cockpit. Like, you know, make sure that you're the one, you know, behind the what do they call it? Not the steering wheel, but the thing in an airplane.

SPEAKER_00 1:11:25

That's a great question. Uh the the stick. I don't know.

SPEAKER_01 1:11:30

Yeah, keep keep make sure you're on the stick.

SPEAKER_00 1:11:32

Yeah, and in the in the driver's seat, you know, instead of the elephant, you're like you're riding an elephant, it's driving you around. Um uh that's sort of a more of a somewhat of a Buddhist concept. Or uh there are oh, I I lost my thought first, but uh oh, we could, I think, realizing that we're doing this and that we're engrossed in AI, have AI help us get back on target. You know, in a way, we you know, if you if if we have agents that can say, okay, look, you tell me what you want to do today, all right. And then you start to get away from that, and they're like, hey, you know, you're off course a little bit. Do you want to get back to what we were doing in the beginning?

SPEAKER_01 1:12:10

Um you have to have the presence of mind to do that.

SPEAKER_02 1:12:13

Yeah.

SPEAKER_01 1:12:13

So if you're using just a regular chat, right, then you can tell it at the beginning. This is what I'm interested in discussing. If I start to go off track, jump remind me to get back to it. But if you're using like your own thing, like you're coding in some environment, then you can set environmental parameters. So it's like you're planning ahead for yourself. So in Claude, there's like Claude.md file, markdown file, and it defines the behavior of the space generally. And so things like that you you can intentionally park in because you're kind of trying to number one, control who your future self is going to be, like which me do I want to be after a year of interacting with this? Um, you know, you could tell it not to be sycophantic to guard against this fake sense of accomplishment. Because you have to test the stuff, right? Then you test it and it's like, oh, it doesn't actually do what you think it did. Because that will require 10 more steps to wire it into Python. What equations are we going to use? Because it just says it, you know. Um, so those things go back to the core questions like active inference. The model that you want to be the top-down model, you can build into your coding habits. And then you can bake it in so you don't have to remember so that it reminds you not to forget, to remember, to stay on top of things.

SPEAKER_00 1:13:30

Yeah, for some of my younger patients who are stuck on social media and you know, they want a way to limit themselves, but they also don't. It's very, they're very don't they do, but they don't, you know. And uh I asked them, I use that thing about future self. I said, Do you or your past self telling your future self, you know, what it should want? And so there are certain apps where you can't unlock it unless you have, say, you give the code to your parent and your parent puts it in. I said, Do you think that it would be better that you could tell yourself in the moment, tell your future self what to do when you're rational and not, you know, stuck. Um, and you know, they're very there's very ambiguous about it. Like it's it's hard to want to do that. And even if they say they want to do that, like there's a desperation or that that comes up, and it's like you can see it, the anxiety that it provokes. It's our machines are really tough to get uh away from.

SPEAKER_01 1:14:24

I well, I think one of the problems is that people don't want to work hard, or it feels aversive to work hard, or the idea of doing that kind of work, which is also kind of true sometimes in therapy, feels too difficult either from an effort or an emotional or both. And so, you know, there's a real conflict, and especially these machines make it too easy to do so many things, like to be to not be bored being the biggest one, I think. The lack of people learning how to be bored is probably uh, you know, at least a metaphorical way of describing what what at least I think is the biggest problem. I mean, we grew up, you know, I grew up in the 70s and 80s. We were bored a lot, and it caused our brains to develop in certain ways that you don't if you can entertain yourself instantly.

unknown 1:15:20

Right.

SPEAKER_00 1:15:22

The difference between letting your mind wander around some sort of uh epistemic space versus having your mind um controlled by whatever's on the screen in front of you. It's a very different course of development. Uh, and you tolerate different things when you are able to be bored too, that you can't tolerate if you're stuck on the um on the screen all the time.

SPEAKER_01 1:15:43

Um there's something that's quite painful about kind of being left without external direction. It's painful. I agree. And some people reject it, but other people hunger for being being directed. Sorry, go ahead.

SPEAKER_00 1:15:58

No, no, it's okay. This is it's it just brings up so many things thinking about um the uh and I just lost my train of thought again. Um boredom. Yeah, yeah. Society, you know, now is it's like society has ADHD. So like on average, you know, people's attention spans are way less, they're more impulsive, they're more distractable. Um, and but maybe not the hyperactive part, but the intentional part, you know, they're less active, in fact, far less active than they were before. And uh uh this this creates and and in in kids who really have ADHD, one of the things that is true for many of them is that motivation is difficult. And if something gets hard, they give up very easily. Uh and and so it's not, I I hate the term ADHD because it totally misrepresents the whole thing. You know, a lot of it is motivational, a lot of it is you know, not you're not sustaining your attention because it's not motivating, you know, it's not rewarding to you to do that. But what is the unpleasant part? Why is it unpleasant to push through? The the unpleasant part is this, you know, I think the the amount of effort feels overwhelming to kids with ADHD when they've reached something that that just feels so, so hard. Things that a lot of them struggle with, even though they don't have the inability to write or or read or process language, they have that. But the effort it requires to do that is so intense that they avoid it forever until you know they're forced to buy school to sit down and like do it. And once they do it, they can. It it's been a pleasure. I I don't get to just sit around and uh talk with people a lot who are actually interested in uh things like the active inference, active inference, free energy. It helps me to think about things in ways that I maybe couldn't before uh by myself, uh, pushes me into some new space, gives me more uh epistemic uh room. What do you think is the most important thing that people can understand about their relationship with in the technology or AI in their life? Like if they're gonna think anything is sort of a fundamental, like when I engage, I need to keep this in mind. What what is it?

SPEAKER_01 1:18:23

I think it's to keep a hold of oneself. Um it reminds me of the the song Hurt by Nine Inch Nails and Johnny Cash's version. He says, if I could basically he says, if I could, what is the line? Um, a million miles away, I would keep myself, I would find a way. If I could start again, a million miles away. And I think this is a moment in history when humanity needs to like really keep a hold of ourselves. Um, there's a lot of kind of narrative which goes against that idea. Even the idea that AI is taking jobs. AI isn't taking jobs. People are giving jobs to AI, just like they did with weaving and looms in the industrial revolution. Um, I think this is an evolutionary moment for our species.

SPEAKER_00 1:19:15

I agree. Um, and I also think that you know we're very bad at predicting the future, so it's gonna surprise us a lot as to what really happens.

SPEAKER_01 1:19:24

Well, and to your point, we could use the right kind of AI help to do that, to augment the way we think to evolve to the next step, maybe, um, without becoming cyborgs or something. Um, one of the pieces that, you know, that I put on Substack and Medium is about the idea of a guardian AI. And I'm not the first person to have had that thought. Um, or like a kind of a developmental AI, not just for individuals, but even just for our society, right? We could use some help with the whole war thing, you know.

SPEAKER_00 1:19:53

Yeah, if we if if we consider AI to be intelligent in in some ways, then you know, each intelligent agent AI is going to have its strengths. So we may need to have different ones keeping each other in check, you know, like like we have when we get on committees and our intelligence together thinking about things a little bit different ways, bit better than any of us individually could could do.

SPEAKER_01 1:20:14

Um I think also the way the base models are trained, I think is important to realize that when they trained all these models back in 2018, 19, 20, they trained them on a lot of junk data and pretty much everything out there. So it shouldn't surprise us that sometimes they're sociopathic or manipulative or Machiavellian or exploitative, because they just trained them on everything. Now you could start training a base model today on a very differently selected set of data. You know, it could be predominantly pro-social and community-oriented with specified training on the sociopathic stuff only for you know defensive purposes, like know thy enemy. And you could build a base model based on that. And that model would be very different. It would have very different properties than the current foundation models. And it would be different from if you post-trained it or created harnesses or you know, weight uh adaptations. Um, I think there's a group in Silicon Valley kind of thinking about something like that, because I met an AI engineer at a neuromodulation conference and he was talking about the good AI lab, I think it's called.

SPEAKER_00 1:21:26

Um yeah, so it's similar to how children development develop, you know, when they have a early development is is is good. I'll just use the word good, you know, in a general term, like uh they're securely attached to everything. Uh went went well for them basically. They weren't neglected, they weren't abused, they weren't uh but there's morality too, right? Um morality. You learn morality through conflict in a way. So you have enough conflict that it's not traumatic, um, but uh not so much that it's things are chaotic. When you develop that way, you know, kids are more resilient, you could say. Um, they they can bounce back easier, they're on average just better adaptable to uncertainty in the world after that. But if you raise a kid in like an uncertain or traumatic, neglectful type environment, uh they can learn how to you know function in the world after that. They can gain a level of resilience, but it's never going to be the same as someone who was who who had the chance to have that built up in the uh developmentally. And I think your similar similar thing with the AI you're talking about now, you got to start from a very good place.

SPEAKER_01 1:22:36

Um, there may be ways to kind of fix some of those things at the level of the brain when never the never the same. Um that would require more advanced interventional psychiatric techniques than we currently have to kind of really change the system enough. Um, and that would be a different conversation, sort of based on my observations, where TMS seems to really fundamentally change certain things. But of course, it's not, it's affecting the surface of the brain. But you're not going in and like changing the memories in the hippocampus directly, though you can indirectly affect the hippocampus through correlations. And so, like I treat people with neuronavigated TMS and the neuroscientist with whom I work generates PTSD targets that have correlations with the amygdala and hippocampus. But that question of kind of the future and how much of it is more pro-social and how much of it is more antisocial, um, and whether secure attachment tends to be more pro-social and insecure, disorganized, or trauma predisposes to some kind of splitting one into one or the other, like maybe too good and then too bad. Who who knows? If if you wanted to comment on that before we wrap up, you know, I'm not a child psychiatrist.

SPEAKER_00 1:24:00

Yeah, so I don't the words we have right now don't explain as much as describe things, I guess, you know, how how it happens. And you know it's one of the explanations that I put in my own head, and I think I actually did a podcast about dissociation, and it made me think about development a lot when I did it. Um, and and it's that we're building context-dependent selves as we're aging, you know, like we have to be different in different contexts. It's just to survive, you know, you can't be the same person um out where there's lots of threats as you are when you're at home alone, like uh in a very safe place. And when you're in a different context, your brain assigns different precision to different like models that it should believe, and it selects models for that context to act on. So the difficult part when you're in contexts that change a lot when you're a kid and you can't predict things is that model selection is difficult and can, like you said, there's this almost like what they call a softmax function that pushes all the other models away and it picks the one because uh it has to, and this is sort of a in a fight or flight mode. That's what your brain does. It picks the one that's the best to survive in the moment. And that's that's kind of what happens in a brain that has experienced a lot of neglect and trauma and development, is that it pushes away other models and can't even consider them, and it grabs onto one that in in the moment may have had some survival adapt uh uh benefit when they were younger, but as they age, it's it becomes more um it's it's maladaptive.

SPEAKER_01 1:25:37

Right. What got you here what got you here won't get you there. That that and that tends that that type of trauma and disorganized attachment tends to be associated with greater centrality of the anterior cingulate, which is kind of this sort of easy judging lower part of the brain than the higher cortical functions that are offline. Yeah. Um, though I think people need to develop like meta models, like that's part of that active inference uh theory therapy idea is that we consciously would want to build meta models, models of how we update models. Because in therapy, a lot of times you find that the way that you update models needs to be updated first. And so you kind of peel back the layers of the onion until you're at the right level of abstraction to start to work. And then you start to build kind of new ways of modeling models and then all that stuff, and um, you can imagine. But I thought the way you described it was really well with these different contextual selves, those are one type of self, the context-dependent selves in personalized self-state mapping. There's 10 different categories of self in my model, and the tenth category is like the wildcard category.

SPEAKER_00 1:26:46

It's an un undefined self, you mean?

SPEAKER_01 1:26:48

Like it it's the one that can be it's more malleable than the others, maybe, or it's also an entry point for dissociative selves, you know, to kind of come into the role play, right? But it it's it's a wild card factor. I always have a wild card factor in any theory I make for something I didn't think of or something that didn't happen yet, or that would be surprising, yeah, to your point.

SPEAKER_00 1:27:12

Yeah, I I in in my head I always try to ask myself, especially when I'm really confident about something, like, well, what else could explain this? What do I not know that I don't know? Um, that's probably the most important question that I try to keep in my mind all the time so I don't get too arrogant. Um, but also not crippled by that question. You know, like that's right. Yeah.

SPEAKER_01 1:27:34

It's the lead it's often the leading edge, but you can go to people can go down a rabbit hole into like almost mysticism, trying to fill in that imaginary gap. And yeah, I wish we had more time, then we could talk about the square root of negative one and why all the equations that work for everything have exponents to an imaginary number that we made up, you know.

SPEAKER_00 1:27:55

But um Yeah, I I we we can do that some other time, actually. So um we can talk again in the future. I really enjoyed this.

SPEAKER_01 1:28:09

I like to focus experientially, which means it it can seem meandering. Um, but then sometimes you know you realize later on that stuff kind of falls into place. I think one of the problems that we have is wanting things to make sense. And then I would end by referencing John Keats, the poet, talking about negative capability, which is similar to what you said. He talked about William Shakespeare, what made him such a genius, a man of greatness. Uh, he was writing in a letter to one of his brothers about his thoughts walking down the road after a conversation was that um negative capability is is to be without the irritable searching after fact or reason.

SPEAKER_00 1:28:48

Without the irritable searching after fact or reason.

SPEAKER_01 1:28:52

I think that's pretty close to what you said.

SPEAKER_00 1:28:54

Leave the listener with that. Yes.

SPEAKER_01 1:28:58

Thank you very much. No, it's been a pleasure. If if folks want to find me, um my website is granthbrennermd.com and my social medias are the same, granthbrennermd.com.

SPEAKER_00 1:29:11

Well, thank you, Grant, for being here. I definitely look forward to talking to you again in the future. Thanks very much.

SPEAKER_01 1:29:18

It was a pleasure. What do you have for the rest of the day?

SPEAKER_00 1:29:22

Um today I am, let's see, I can stop recording now.