Tanha with neuroscientist Michael Johnson

Transcript

00:00:00Hello and welcome to Imagine an Apple, the podcast about our inner mental worlds. My name is Francis and today our guest is Mike Johnson.

00:00:08Hi everyone.

00:00:09Also with me is Vin, my co-host. Hello Vin.

00:00:12Hey Francis. Hi Mike. So Mike is, uh, formerly worked at the Qualia Institute but now is currently doing other stuff.

00:00:21But today we will be interviewing Mike about his internal experiences, specifically about the stuff that he specializes in.

00:00:29Tanha, qualia, and active influence among some of the things we'll be talking about.

00:00:42So

00:00:44So what we're interested in talking about today is inner experiences and how you've

00:00:51might experience some of these, um, phenomenal, uh, your phenomenology, whether it's more visual or auditory or, or, um, olfactory.

00:01:01And I was wondering if you have, um, special kinds of phenomenology that you think are uncommon in the general population.

00:01:08Hmm. Thanks. That's a, that's a really good question.

00:01:11Um, so I would open with, uh, you know, no one exactly knows how common or uncommon their phenomenology is.

00:01:20but we can sort of analyze base

00:01:24base rates and different sort of physiologies. And yeah, I was actually talking about this with

00:01:33a friend today. And I was saying, well, I think that I have a somewhat unusual

00:01:42phenomenology. And they were saying, okay, cool, Mike, why do you think that? What makes you say

00:01:49that? And I actually wrote a decently long article about something very similar. It's called

00:01:59autism as a disorder of dimensionality. And it basically makes the case that people vary in a

00:02:07very interesting way. So part of physiology is understanding dimensions of natural variation.

00:02:16And kind of the interest,

00:02:19give rise to interesting personality types, or in this sense, interesting phenomenologies.

00:02:25And my thesis in that piece was that people vary in terms of literally how many neurons they have

00:02:37in a unit volume of brain mass. And this is based on some research, some autopsies of

00:02:47children with autism.

00:02:49And basically, that, and this is actually a pretty crazy, crazy result. They did autopsies

00:02:58and found that autistic children had on average, 67 more, 67% more neurons in the same volume of

00:03:08prefrontal cortex, which is a lot. It's not just 1% or 10%. It's 67% more.

00:03:16Um, so that got me,

00:03:19I'm sort of thinking about, okay, well, what does that mean? And can we sort of derive

00:03:26interesting things from this sort of one factor model? And my thesis was that, yes, we can.

00:03:33And I guess I'd sort of frame that as, if you, if you take a, so everyone's kind of familiar with

00:03:41how LLMs work, at least a little bit these days. And, and sort of the, the challenges of,

00:03:48if you sort of, if you sort of, if you sort of, if you sort of, if you sort of, if you sort of,

00:03:49sort of have a certain number of parameters in your LLM, then you have certain alignment

00:03:54challenges. Certain things work to align the LLM and certain things maybe don't work as well.

00:04:01And in this case, basically you'd be jacking the parameter count by 67%. And so you get into

00:04:11a different kind of, different kinds of failure mode. And I promise this is answering your

00:04:19question. Yeah. So the observation that I made in the piece was that the natural sort of circuitry

00:04:31that evolution gives us sort of has pre-built optimization into it. That this is a

00:04:41very finely tuned evolved circuit that produces a specific result. But basically if you make

00:04:51these networks thicker, if you take an optimized circuit, I don't know if you're familiar with

00:04:56in computing, there are ASICs,

00:05:00application-specific integrated circuits, basically a custom chip for a custom task. And it does one thing, but it does this one thing really well, so like a Bitcoin miner.

00:05:11And then you have FPGAs, field programmable gate arrays, that basically, it can be anything. This is a chip that you can reprogram at the hardware level.

00:05:23You can make it look like any other chip, or you can sort of solve any problem with hardware, but you actually have to do the programming.

00:05:31And so, if this sort of autists have more thicker networks, we can say, more neurons and more connections between neurons,

00:05:45then their circuits are more like an FPGA than an ASIC.

00:05:49More like a general unoptimized circuit.

00:05:53More like a generalized network than sort of a sleek, highly tuned, thin network.

00:05:59Right.

00:06:00So, yeah.

00:06:01So, in this, so I'll pause a little bit. Does that all make sense?

00:06:06So, I'm a little bit lost at the moment.

00:06:08I'm not entirely sure what an ASIC is, or the other thing that you mentioned, an FPG.

00:06:13And not entirely sure how emotion, how this relates to emotions and qualia.

00:06:23So, expand a little bit more what you mean by like dimensionality.

00:06:26What does that mean for autistic brains to have different dimensionality?

00:06:31Yeah, yeah, you bet.

00:06:33So, the, so an ASIC is a specialized chip.

00:06:38It's very optimized.

00:06:40And an FPGA is a much more generalized chip.

00:06:43It's not optimized.

00:06:45But you can program it yourself.

00:06:46And the sense is that if a brain has sort of extra neurons in it,

00:06:53it's more like an FPGA than an ASIC.

00:06:56It has a lot of potential in it.

00:06:58But it's not particularly optimized.

00:07:02Right.

00:07:02But what is it optimized to do?

00:07:04Yeah, yeah.

00:07:05So, we generally, like a lot of circuits are pretty optimized in the brain.

00:07:10Like we have circuits to detect faces.

00:07:13We have circuits to detect noises and scary noises.

00:07:16We have circuits to figure out which foods are tasty and which are not.

00:07:21And the, the ASIC is a very well-known tool.

00:07:24people on the spectrum, on the autism spectrum,

00:07:27might have to do a little bit of post-production tuning on their circuits.

00:07:34They're not so optimized because it's just they have more neurons

00:07:41than what evolution was planning for.

00:07:44Right. Okay.

00:07:44So does this translate to a different inner experience, so to speak?

00:07:48Do they have qualia that might be different from someone who's not autistic?

00:07:54You know, I want to say yes.

00:07:57And I think that part of that is every sort of every ASIC is going to be

00:08:04every sort of highly evolved circuit is going to be very similar.

00:08:08But every FPGA is going to be wired up a little bit differently.

00:08:15And so you're going to get a lot more diversity in sort of the solutions

00:08:21that the brain finds.

00:08:24Oh, okay.

00:08:25So I guess this would like, I don't particularly make autism a part of my identity

00:08:32or not part of my identity.

00:08:33It's just the word.

00:08:34But I do think that like this, I sometimes interact with the community

00:08:42and there's this focus on neurodivergence.

00:08:46And I think that's interesting because it's like everyone with autism is going to be,

00:08:53you know,

00:08:54different than people without autism,

00:08:56but also different from other people with autism.

00:08:59It's like if you have a very small circuit,

00:09:03there's one way to make it work.

00:09:05If you have a very big circuit,

00:09:07maybe there are a hundred different ways to make it work.

00:09:09And maybe you found solution 57,

00:09:11whereas someone else found solution 62.

00:09:15And I think that that may sort of show up in sort of how sensory experiences

00:09:21present to people.

00:09:24Right.

00:09:24So here you're talking about the phenomenology of direct sense.

00:09:27So how people, when they're actually seeing something, experiencing it,

00:09:32is that what you mean would vary?

00:09:34Yeah.

00:09:35I think that you actually get different results in phenomenology.

00:09:39Okay.

00:09:40And in terms of the kind of more imaginative, internal phenomenology,

00:09:45so like inner imagination and inner voices, that kind of thing,

00:09:50do you think use of those would vary?

00:09:52Or is that like a separate question?

00:09:55I think it's all linked.

00:09:58So yeah, I would also

00:10:00apply this to that frame as well.

00:10:02So imagination often has a practice component as well. And for many people, they engage in some kind of attentional training. Meditation being a pretty good example of this, but other techniques, alexander techniques, various other things also come into play.

00:10:19I'm aware that you are a meditator. Am I correct on this?

00:10:23Yeah, that's right.

00:10:23Do you think that your meditation practices lead to different inner experiences, whether it regards to like your imagination or how you experience qualia differently?

00:10:35Yeah, I would say so. If I had to say some words on that, I do think that just practice looking at things makes them more clear.

00:10:46And I would also say that just like you mentioned alexander technique.

00:10:51And I guess,

00:10:53especially lately after my latest piece on vessel computation, I've been sort of practicing this.

00:11:02There's this phrase, opening the hand of thought.

00:11:06And I think that practicing that is a very powerful move.

00:11:12And often there are clenches and clamps in our phenomenology and just sort of practicing, okay, like, what would it be so bad?

00:11:22Would anything bad happen?

00:11:23Would it be so bad if I just sort of released and just tried to be an open experience?

00:11:28And I think that's very powerful.

00:11:30Yeah. Opening the hand of thought.

00:11:31I really like that phrase.

00:11:33You mentioned vasocomputation, which is something that I've seen you written about.

00:11:37Can you tell us more about what vasocomputation is and how that relates to TANA, which you mentioned quite a lot in your blog post?

00:11:44Sure. Yeah.

00:11:46Well, I'm a bit afraid that I'm kind of throwing listeners into the deep end here.

00:11:53That's okay. We'll unpack it once we're done.

00:11:55Nice. Nice. Good.

00:11:58So I wrote this piece called Principles of Vasocomputation.

00:12:02And it's basically a unification of three or four domains.

00:12:07And the first domain is Buddhist phenomenology.

00:12:11And especially this term that the Buddhists call TANHA.

00:12:16And this is often translated as grasping or thirst.

00:12:21Or clenching or desire.

00:12:23And according to the Buddhists, it's responsible for maybe 90% of all sort of moment by moment.

00:12:33And so it's kind of, it's a big deal.

00:12:35It's important.

00:12:36Just to make it clear what you mean by TANHA.

00:12:40When you experience it yourself, can you describe how that comes to you?

00:12:45Do you get like, is it like a verbal thought or do you see things?

00:12:51Or is it just the experience of having TANHA?

00:12:53Yeah. So there's, I would call it a contraction in awareness.

00:12:59And there's both the general contraction and a contraction around a specific part of awareness.

00:13:06And you can kind of gauge how flat your phenomenology is in a way.

00:13:13And when TANHA is going,

00:13:19your phenomenology feels more lumpy, I guess I would say, for lack of a better word.

00:13:25Cool. Okay.

00:13:26So how does TANHA relate to this thing called free energy and active inference?

00:13:33And can you help unpack those terms and what they mean?

00:13:37Sure, sure.

00:13:38So principles of vessel computation connects three domains.

00:13:43And one is Buddhist phenomenology.

00:13:46And the second domain is TANHA.

00:13:48And the third domain is the active inference framework.

00:13:51Or the more full mouthful there would be the free energy principle active inference framework.

00:13:58And so this was developed by Fristen and others.

00:14:03And it's basically a framework how the brain is a prediction machine.

00:14:09And we sort of hallucinate our reality.

00:14:11We're always trying to predict our sensations.

00:14:13And we live in that story, we can say.

00:14:19And active inference is another layer on top of this that says,

00:14:25not only do we make these predictions all the time,

00:14:28not only do we try to predict our sensations,

00:14:32but sometimes we predict false sensations and we hold these predictions until we act in the world to make them true.

00:14:42So this is a very subtle, important thing, I think.

00:14:45That, for example,

00:14:48you know, if I'm thirsty, maybe I'll predict, oh, I'm not thirsty.

00:14:54I have the taste of water in my mouth and down my throat.

00:14:58Of course, that's not true.

00:15:00That's a false prediction, but I'll hold it in my mind until I go, I pour myself a glass of water and take a drink, then I can release the prediction. Or if I turn the burner on on the stove, I'll hold the prediction that I need to turn off the burner until I turn off the burner. So I sort of keep track of the state of things by active inference.

00:15:23InScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScriptScript

00:15:24So yeah, this is a really cool, cool framework.

00:15:29Sorry, I was going to say,

00:15:31so the Burnham one was super interesting.

00:15:33So there you said, once you've turned it on,

00:15:37you would have like in your present mind,

00:15:40like a prediction that you need to turn it off

00:15:42rather than predicting that it will turn off.

00:15:45Yeah, there's some kind of accounting system in the brain

00:15:49where, okay, you know, I know now I'm turning this on

00:15:54to sort of hold this state in my brain.

00:15:59And I know that I can stop holding the state

00:16:02once I turn it off.

00:16:04So we can sort of frame that as what I'm holding

00:16:07is the prediction that I, you know, the burner is off.

00:16:10Gotcha.

00:16:10Right, yeah.

00:16:12I have a question about this

00:16:13because you say that making a prediction,

00:16:17for example, that you are not thirsty

00:16:20is said to be a false prediction

00:16:22according to this framework.

00:16:24Right.

00:16:24Whereas I typically interpret my desire

00:16:27to grab a glass of water to be a type of wanting.

00:16:31It's a type of desire

00:16:32rather than thinking of it as a false prediction,

00:16:36which I hold true until it becomes true

00:16:37by way of my acting in the world.

00:16:39And is there a difference?

00:16:41Is there a difference between like falsely predicting

00:16:43and simply wanting or desiring?

00:16:47Or are these like functionally the same thing?

00:16:49So it's the same phenomena, same mechanism for achieving.

00:16:52Yeah, that's a good question.

00:16:54I would say that with it,

00:16:56it of course depends on the mechanism.

00:16:59And with some mechanisms,

00:17:02it's implemented in such a way that they would be the same.

00:17:04And in some mechanisms, it would not be.

00:17:07So I guess this would be an argument

00:17:09for sort of dropping down

00:17:11and paying attention to implementation,

00:17:14which is the third domain, of course.

00:17:16Right, cool.

00:17:17So can you tell us more about Buddhist conceptions of Tana

00:17:23and how that works?

00:17:24And how that might have influenced your development

00:17:27of your framework in Veso computational theories?

00:17:32Yeah, sure.

00:17:34So I want to give a shout out to both Romeo Stevens

00:17:38and Nick Camerata.

00:17:41They've really done a lot of foundational work

00:17:44on thinking about Tana, novel translations of Tana,

00:17:49or like sorting through, okay, what's a good translation

00:17:51and what's not.

00:17:52And in first person,

00:17:54exploration that, you know, can we see this?

00:17:59Can we put a timeframe on, you know,

00:18:03does it happen within 25 to 50 milliseconds

00:18:07or a hundred milliseconds or whatnot?

00:18:09So they've done really, really great work.

00:18:14I think, so my focus is generally

00:18:19sort of connecting these domains and saying,

00:18:22okay, this Buddhist,

00:18:24Tana is sort of using this active inference system

00:18:31in an unskillful way,

00:18:33trying to control sensations in ways that don't make sense

00:18:39in some sort of like there are in programming,

00:18:43we can talk about type errors, trying to, you know,

00:18:47add two letters together or multiply two letters together

00:18:51or something like that.

00:18:52It's just like, that's not the sort of thing that we're,

00:18:56the tanha as unskillful active inference frames as the brain engages in a lot of that stuff

00:19:03when it tries to manipulate sensations that it's it's often unskillful in the ways that it tries to

00:19:10make predictions about sensations and that in a in a moment by moment sense this happens a lot

00:19:18and this really adds up to like uh a lot of un sort of submerged unpleasantness in your moment

00:19:27to moment experience and that i mean the the buddhists say that and if you fix this if the

00:19:33the vipassana frame is you can see this happening and once you see it you can't really unsee it

00:19:39and you'll start to not do wrong things right so when you say see it what do you mean by that do

00:19:48you mean like a

00:19:48literal seeing um i would say that uh when you observe what the mind is doing i wouldn't say uh

00:19:57it necessarily flashes into your visual

00:20:00field, but I do think that everyone's going to be a little bit different in how it presents.

00:20:07But I would say that I think the takeaway would be sort of a bit of a shock, like, oh, wow, I'm doing that all the time?

00:20:17This feels suboptimal. And I think that progress on the path, on sort of getting better at not doing this, is sort of developing a taste for what it feels like, and a distaste for like, okay, no thanks. Like, I don't need to do that to myself.

00:20:39In your blog post, you use the word stress, just in a passing part of it.

00:20:44Is that potentially one of the ways that you're going to be able to do that?

00:20:47Yeah, I think that maybe someone would experience this. So if you're grasping a lot, partly that might add to your stress.

00:20:52Yeah, so I sort of break down the Tanha as unskillful active inference into three buckets.

00:21:02The first bucket is just if you try to control your sensations or control your environment more than you have the energy to.

00:21:12And, you know, we all know people that try to control their sensations.

00:21:17And we can all kind of feel, okay, you know, that runs into problems.

00:21:25And you just get stressed trying to control everything, and it doesn't work.

00:21:30I think what the Buddhists are pointing to, though, is a little bit different.

00:21:35It's that we try to apply active inference or predictions to sensations in ways that don't make sense, could never come true.

00:21:47Or just are sort of bound to cause us a lot of suffering.

00:21:51A couple of examples?

00:21:52Yeah, so a couple simple examples would be like if a sensation is good, it's a nice, pleasant, tasty sensation.

00:22:02For example, if you're eating some cake and you really like it and you make the prediction, I will have this taste in my mouth forever.

00:22:12Of course, you're going to run into trouble because, okay, you're going to have this problem.

00:22:17And then you're going to eat all the cake, trying to eat that taste in your mouth, and then you're going to be out of cake.

00:22:23And then you're just going to be left with tension.

00:22:25You can't make that prediction true.

00:22:28And you don't have sort of any easy way to release that tension.

00:22:33The word forever in there is very powerful because, yeah, I wouldn't have thought I feel like I want the taste forever.

00:22:39But maybe there is an aspect to that of the desire of the taste.

00:22:43Yeah.

00:22:44So Romeo Stevens says this.

00:22:47This nice frame that we try to make our sensations stable, controllable, and satisfying.

00:22:55And, of course, the Buddhists would say, you know, we can never ensure any sensation stable or controllable or satisfying.

00:23:05So we're sort of doomed to discomfort when we assume that.

00:23:10Right.

00:23:10Going back, do you...

00:23:12So you mentioned before that there are three unskillful active influences.

00:23:16Yeah.

00:23:17And the first one was that if you try to control the environment more than you have energy to, that leads to stress.

00:23:22What are the other two causes of stress or unskillful?

00:23:26Yeah.

00:23:27So the first is just trying to control too much.

00:23:30The second was trying to control in ways that don't make sense.

00:23:35For example, you know, I'll have this taste forever.

00:23:39Or, you know, if you drop a drink on the floor and you make the prediction, that didn't happen.

00:23:45Right.

00:23:47So it's not going to make it true.

00:23:50You can hold it, but it's not the appropriate prediction for the event.

00:23:56And then the third bucket or category of sort of ways that we can be unskillful in using active inference is...

00:24:05And I think this is really a cool, important thing.

00:24:08It's context desynchronization.

00:24:11Mm-hmm.

00:24:11So if you're doing a very hard math problem.

00:24:14And you're sort of holding, you know, using tension as medium-term memory and sort of holding a problem in your mind by sort of let's pinch here and let's hold here and so on.

00:24:28And then someone knocks on the door and asks to, you know, get a cup of flour, a cup of sugar, whatever is the Xanadu scenario.

00:24:39Or, you know, someone says, oh, you know, we have an emergency.

00:24:44There's a fire alarm or something.

00:24:45The context gets switched very rapidly.

00:24:49Mm-hmm.

00:24:49The tension that you're holding in your mind to remember the things in context one no longer has sense in context two.

00:25:00it's just stress. It's just tension. It's just suffering. It only has meaning within the right context. And it's like the, you know, if you're working on a computer program and the memory allocation changes, then your pointers are not going to point to the right things.

00:25:20And I think this actually happens a lot. And that a lot of our stress, it's sort of these fragments of tension that made sense in one context, but now we're in a new context where they don't make sense.

00:25:35Okay. So what is the, that's what unskillful, active, and looks like. What does skillful management tension look like?

00:25:43Right. Not using it a lot.

00:25:47Which means letting go?

00:25:50Just kind of, you know, a little bit, a touch here, a touch there. We still need to navigate the world. And we may need tension to do that. We may need predictions to do that. We may need active inference to do that. But don't, like, make as few as possible. And don't hold them.

00:26:09So skillfully choose your predictions and then like update them more often or replace them?

00:26:14Yeah, yeah. Just hold them gently.

00:26:17Right.

00:26:18So how does this now like lead to the next step?

00:26:20to predictions?

00:26:23Yeah.

00:26:24And what is like the optimal level prediction or active inference to do in order to have a healthy relationship, internal states, whether it be?

00:26:32Right. That's a good question.

00:26:35I think that, you know, as we were saying, you know, being sparing with the predictions that understand that a prediction is a commitment to tension and a little bit of suffering until it comes true.

00:26:51But you also need them to navigate the world.

00:26:53Yeah.

00:26:54I think that, so we were all at just camp together where we met. And Vin and I were talking about the self and, you know, the self as kind of the sum total of tension in the mind.

00:27:09And I think this sort of nicely dovetails with some of the ways the Buddhists talk about the self and tension as sort of not the optimal way to navigate the world.

00:27:25Um, so I guess I would just say that, uh, you know, everything is contextual.

00:27:30There may not be a, you know, this is always good.

00:27:33This is always bad, but there will always be a dimension of, is this skillful?

00:27:38And is this not?

00:27:39Can you talk about what, what good and bad mean in this concept?

00:27:42I know that's like, right, right.

00:27:44Um, yeah, that's, that's a hard problem.

00:27:47Um, I mean, so in my more formal philosophy research, um, I study pain and pleasure.

00:27:55And I'd be, I'd be hesitant to, to say, yes, you know, pleasure is always good.

00:28:00Pain is always bad.

00:28:01Uh, because now sometimes these sensations are trying to do things for us.

00:28:06And, um, I think that, uh, I mean, good and bad could be framed in terms of the free energy

00:28:12principle.

00:28:13Um, and sort of, it could, you know, be framed in terms of, uh, adaptiveness to the environment

00:28:20and so on.

00:28:21So I'm not sure I have a, a sort of clean.

00:28:25Like this is what good is.

00:28:27This is what bad is, but just an appreciation.

00:28:30There's, there's a lot of theories.

00:28:31Yeah.

00:28:31This is a lot of context.

00:28:32Yeah.

00:28:33That's a pretty tall order to deliver.

00:28:35Um, anyway, uh, just moving on from that, I suppose, um, I really want to talk about

00:28:40qualia and when you worked in a institute that had qualia in its name, and I was wondering

00:28:46what you thought, um, what, what do you, what do you take qualia to?

00:28:51And can you tell us more about that?

00:28:52Yeah.

00:28:53Um, so I always.

00:28:55Uh, introduce quality as the components of subjective experience.

00:29:00And it's a little bit, you know, uh, the, the easy, uh, explanation would be, you know,

00:29:06there's red, there's heavy, there's, uh, hot, et cetera.

00:29:12Um, but what the actual natural kinds of qualia are is a very interesting unsolved question.

00:29:18Um, thanks Mike.

00:29:19Yeah.

00:29:20You bet.

00:29:20Uh, thanks Ben.

00:29:21And, uh, thanks Francis.

00:29:22Thank you.

00:29:23It was super interesting.

00:29:25Yeah.

00:29:25They're connecting to some really important things about how we experience our life every

00:29:30day.

00:29:30Um, but also a theory that connects them up, which is great.

00:29:34Yeah.

00:29:36Yeah.