Watch Episode Here
Listen to Episode Here
Show Notes
This is a ~1 hour discussion with Richard Watson (https://www.richardawatson.com/) and Josh Bongard (https://www.uvm.edu/cems/cs/profiles/josh_bongard) touching on topics of agency, polycomputing, oscillations, and evolution.
CHAPTERS:
(00:00) Defining Co-Located Observers
(08:22) Energetic Observers And Self
(18:35) Harmonics, Independence, Polycomputing
(30:00) Frog-Pond Developmental Attractors
(35:40) Mirror Symmetry And Mathematics
(42:11) Double-Slit Observer Analogies
(51:59) Chaos, Coherence, And Cracks
(01:00:06) Nested Multiscale Agency
PRODUCED BY:
SOCIAL LINKS:
Podcast Website: https://thoughtforms-life.aipodcast.ing
YouTube: https://www.youtube.com/channel/UC3pVafx6EZqXVI2V_Efu2uw
Apple Podcasts: https://podcasts.apple.com/us/podcast/thoughtforms-life/id1805908099
Spotify: https://open.spotify.com/show/7JCmtoeH53neYyZeOZ6ym5
Twitter: https://x.com/drmichaellevin
Blog: https://thoughtforms.life
The Levin Lab: https://drmichaellevin.org
Transcript
This transcript is automatically generated; we strive for accuracy, but errors in wording or speaker identification may occur. Please verify key details when needed.
[00:00] Josh Bongard: So
[00:01] Richard Watson: Your question was, is it possible to see something doing two different computations without the observers being in a different time or place? My conjecture was no. If I understand you correctly, you don't like that answer. The truth is neither; it might become here or there, but it's neither here nor there right now. What intrigues me about that is I'd like to know more about why that's not the right answer for you.
[00:43] Josh Bongard: Got it. I don't like that answer, but I don't know that it's not true. It may very well be true that there cannot be two observers observing the results of two different computations, where those computations are arising in the same place at the same time, and the two or more observers are also co-located in space and time. I wanted to drill down about whether that is possible or not. If they're going to make an observation of these two results, and they're going to be in the same place at the same time, then it would seem that, yes, the results themselves have to be in the same place at the same time. The only example I know of with Etusa's work is in the case of vibration, where the results are arising at two different frequencies at the same place at the same time. So can an observer observe two different frequencies that are in the same place at the same time? Your answer last time, Richard, was no, because they need to be at different distances from the phenomenon, because vibrations are ultimately arising from rotations. Do I have that? I think I've garbled that.
[01:56] Richard Watson: I'm not so particularly attached to the why. I just make up the answer. That feels right. It feels like the presence or absence of a particular computation or some other computation is only a perspective. And to have a different perspective requires you to be a different thing or be the same thing in a different place or be the same thing at a different time. My gut said you can't have two different perspectives without being different. But when you said it just now, if two computations can be in the same place and time, can two different observers be in the same place and time? I'm not so sure because that feels like that might be a neat little symmetry that a capital A observes little A and capital B observes little B and they co-create one another in their symmetry.
[03:13] Josh Bongard: And if observers can also be observations and vice versa, then the symmetry is more likely because if it holds in one and observers can be observations, then it's probably going to have to hold in the other. But I still can't see through to how. What would this look like in biology where you'd have multiple observers in the same place at the same time? I don't know.
[03:39] Michael Levin: I was trying to figure out what we really mean by same place and time. I wonder if in biology — here's an example. Would you consider the two alters of a mildly dissociative human cognitive system? You've got two different personalities, same brain. I don't think you could say that they're just sitting in different parts of the brain. They see the same event. One says, "That's pretty normal." The other says, "I'm the one who can tell this is abusive. This is not right. It looks normal to you. I know it's not." Is it possible that the different place and time is not physical place and time? What's different about the perspective is the history that you've had or your cognitive architecture that enables you to see things differently, but it's not because you're in a different place in time right now and physically having a different perspective on it. Your perspective is different in some other space. Would that work?
[04:51] Josh Bongard: I could imagine that same place if we consider one brain the same place, but alters tend not to coexist at the same time. They turn on and then turn off again. You were saying in this conversation, one says something and then one responds.
[05:08] Michael Levin: To clarify, what they do take turns in is in taking over the physical control of the physical body. That doesn't mean they're not on non-verbally at the same time, where who's in charge isn't aware of the other conversations. And in fact, I've seen transcripts of conversations that the alters are having with each other about how to go about getting control and this kind of stuff.
[05:38] Josh Bongard: Interesting. Okay, yeah.
[05:41] Michael Levin: Clearly the brain is not indivisible, and so you might argue that they're not actually in the same place and time. I think we're getting close to being able to say that in the same medium, in the same chunk of medium, you could have multiple. If we're going to say that patterns are observers, then you could have different frequency signals down the same coax wire.
[06:10] Josh Bongard: But can you boil this down not to brains which are divisible, but to things that are increasingly indivisible? How far could you push this phenomenon?
[06:25] Michael Levin: There's another notion of same place and time, which is a hierarchy. In a cellular automaton system, the glider might have a perspective, the individual cells might have a perspective. Are they in the same space and time? But also not.
[06:49] Josh Bongard: This was going to take me into some ideas you were writing about this past week, Mike, about where is the agent? Is it in the energetic pattern or is it in the matter? Thinking about these observers, in our case, thinking about granular metamaterials and you have a vibrating particle, if that vibrating particle is vibrating at different frequencies and those are computations, there could be vibrations occurring at yet other frequencies that are damping or propagating one or two of those computational results on, and we could treat that vibration at that other frequency as the observer. It's doing something with the result of that computation. The only way I can think of multiple observers co-located in space and time is that they're energetic observers, they're vibrations. It's the minute you start thinking of the observer as made up of matter or as a thing, it feels very hard to think about multiple observers co-located in time.
[08:00] Richard Watson: You want relational processual observers, not substance observers.
[08:05] Josh Bongard: And that would support Mike's thinking this past week that maybe agency is better attributed to energetic patterns than matter. Does that comport with what, or am I mischaracterizing what you said, Mike?
[08:22] Michael Levin: No, that's certainly part of it. But now I'm wondering, maybe we need Chris or Don Hoffman or somebody, but do we have matter anymore? Meaning, isn't everything supposed to be some sort of residence of a field or something.
[08:39] Richard Watson: But it's at least a vibration, if not completely made up. It's pretty good, if not entirely imagination. I want to float the contrary conjecture. New conjecture: the only way to see two different computations in the same substrate at the same time is to have two different observers in the same substrate at the same time.
[09:16] Josh Bongard: Yeah. Okay.
[09:24] Richard Watson: I'm also cognizant of the fact that when you're talking about computations that are done with vibrations, there's no such thing as a computation at an instant. There's no such thing as an output at an instant. There's no such thing as a velocity at an instant.
[09:48] Michael Levin: Nor is there conventional cognition at an instant. You're talking hundreds of milliseconds in order to say anything about the human and his thinking.
[10:03] Richard Watson: Did you answer the question about why you wanted it that way, though, Josh?
[10:29] Josh Bongard: It would be more interesting if it was that way, but of course it's not up to us. I think a lot of the work that we're doing and others are doing — this traditional view of computation just seems increasingly wrong. The same thing with life, genetics, and evolution: we seem to keep stumbling over the fact that maybe we have it wrong. I keep coming back to the fact, well, how wrong do we have it? Maybe we should just explicitly try and push in the opposite direction. Let's start with the postulate that observers are not distinct in space and time, and where does that go? And then, what else? Memory, agency, observers, computation — these other building blocks of our assumptions about computation.
[11:18] Richard Watson: Yeah.
[11:20] Josh Bongard: True, we can fall back on quantum effects and then things get weird very easily. But are things weird in a Newtonian sense also? We haven't thought properly about how weird it could be.
[11:38] Richard Watson: I can see one reason that I might like it to be true, that two observers can be in the same place at the same time. Just because I like the symmetry of it: the idea that an observation or an observer making an objective observation might not make sense. Observation is always a computation; observation is always a process over time; it's always involved and not just watching.
[12:29] Josh Bongard: So if we're assuming they're separated in space and time, then it's easier to say that's a result, that's a computation, that's an observation. But once that goes away, then it makes much more sense to conflate these things we thought were separate.
[12:45] Richard Watson: One way to get some leverage on it would be to say, Mike was saying, "what do you mean by it being in the same place at the same time?" If you had two observers in different places or different times, they couldn't even agree that the two computations they were observing were in the same place or time. Relative to me, it's in a different place than it is to you. We'd have to have some common frame of reference in order to agree that the computations we're looking at were in the same place and time.
[13:38] Josh Bongard: This takes me into embodiment, what's an action? How can an observer say, "Here, let me do something and I'll prove that the result is coming from here at this time," and the observer—it's not so easy anymore.
[14:07] Richard Watson: If you're doing two computations at different frequencies in the same substrate, you can change one of them independently of the other.
[14:38] Josh Bongard: So a quote-unquote observer could modulate its amplitude or frequency or phase offset and muffle one of the two computations. It does a knockout study. It says, "I'm going to shut off the thing that I'm looking at," and that thing at that other frequency disappeared. Now I'm going to stop and it's going to come back on. Could that action, however it goes about modulating amplitude and frequency to muffle, to create negative interference? Can actions be in the same place at the same time?
[15:19] Richard Watson: Now I'm bothered about who's observing the observers.
[15:27] Josh Bongard: Okay, here we go.
[15:30] Richard Watson: Where's the frame of reference that can make a judgment about whether these two observers are in the same place or time?
[15:36] Josh Bongard: Yeah, okay. Infinite regress.
[15:40] Richard Watson: The only way to avoid the infinite regress is to say the thing that you thought was doing the two different computations is the two complementary observations about the two observers. That way you have a closure on the frames of reference.
[15:55] Josh Bongard: Okay.
[15:56] Richard Watson: Without. invoking another frame of reference outside it.
[16:01] Michael Levin: The same thing as making it, closing the loop, Hofstadter style, making itself, is that the same or different?
[16:22] Josh Bongard: The observer sees itself or sees itself through an intermediary observer?
[16:28] Michael Levin: The idea is that it's all observer-based, but you don't need others to tell you that you're real, because either all of them, or maybe it appears at some level, but some of them observe themselves. So the reason that you exist is not because you need anybody else to observe you. They do, and they make conclusions for themselves, but you can observe yourself. It's a self-lock. It locks you in place through a self-referential thing. And I think that's what Doug was arguing.
[17:09] Josh Bongard: Yeah, I don't know how that would work.
[17:16] Michael Levin: We have physical models of that kind of thing. There are knots and things that basically hold themselves together.
[17:34] Josh Bongard: I wonder if there's a way you could, I'll call it an observer. You have something operating at one frequency and it's generating patterns at other frequencies that are predictions of what it's going to do or its actions. And then it matches it up. So it's the free energy principle. It says, I think this is about to happen to me or this is what it's predicting, it's retrodicting. There are patterns arising at different frequencies just before or just after patterns arose at some other frequency. And that's demonstrating that it's observing itself. It knows what's about to happen.
[18:15] Michael Levin: That too sounds like reproduction, patterns spawning off other patterns. It is reproductions and maybe they're hypotheses about the environment or about your own future states.
[18:29] Josh Bongard: Yeah.
[18:35] Richard Watson: Another question that comes to mind is how separate do you want these two computations? If you have two computations that are happening at two different frequencies, and the detector for frequency one cannot see frequency two at all, that means it's not in any harmonic ratio. This means it has to be an irrational ratio. That's the only way the two frequencies can't interfere with each other at all, so it really is a two-dimensional computation that's happening, not two computations folded into one dimension. But I think we are interested in computations that do interact with each other most of the time, or at least they're not completely orthogonal; they share some common frame of reference.
[19:56] Josh Bongard: Why is that, Richard? Why is that more helpful?
[20:04] Richard Watson: Because otherwise, it's just a bag of computations. And I'm interested in building systems out of subsystems which are doing multiple computations, but also interacting with each other.
[20:18] Josh Bongard: But you're in the world and stuff comes at you independently on different modalities from the world.
[20:26] Richard Watson: I need to do some sense of fusion with them. Not just treat them as — they're ghosts that don't even haunt you.
[20:43] Josh Bongard: The lazy sound bite for this morning.
[20:51] Richard Watson: They're just other computations going on that you have no contact with at all.
[20:59] Michael Levin: I love that framing of it because I was putting in that outline of the paper that we were talking about. I put in this thing that Jung said, "People don't possess ideas; ideas possess people." It's the same notion of a thought pattern that's looking for a substrate and it can facilitate its own survival by either doing a little niche construction in that substrate or adjusting itself as needed, but that kind of thing. So there are ghosts that haunt you and ones that don't.
[21:44] Richard Watson: If we're two different observers observing different computations in this thing, and we want to compare what each observed, there has to be something commensurate about that. You have to be able to talk the same language. There has to be a common observation; you have to be able to convert those observations into a common observation that facilitates comparison. For truly orthogonal dimensions, you just can't do that. One is just not there to the other one. It's a measure 0. That's the thing I think is interesting about harmonic relationships: their semi-independence. They're not a direct dependence. It's not that this one tells you everything about the other one. But they're not entirely orthogonal either. And that's what gives rise to the naive idea that you can only have one computation in a given place at a given time: it's saying we live on a number line and you can only be in one place at a time. But if you live in a two-dimensional space, you can represent two different numbers at the same time. That's not what I wanted. What I wanted was something that was two numbers but one number. It was commensurate and independent.
[23:30] Josh Bongard: That's really interesting. I have a PhD student in my group who's doing some stuff with the computational metamaterials, and she's looking into logical independence. Can you actually knock out one of the computations that's at the same place at the same time without harming the other one? We've been trying to push to evolve, or optimize or train these materials to increase independence, but maybe that's not ultimately useful. Although in traditional electronics, that was the assumption. We want to shield computations and memories and storage from each other.
[24:12] Richard Watson: And then you bring them together when you're ready.
[24:15] Josh Bongard: It's super hard to engineer something where everything's not modular. Everything's affecting everything else.
[24:24] Richard Watson: But that's the normal engineering-y way of doing things. I'm going to keep these things separate until I'm ready.
[24:33] Josh Bongard: Because that's all we can handle.
[24:34] Richard Watson: Because I can't handle it otherwise. Whereas biology is exactly not that because every frame of reference that's involved was co-created by the products of the processes that were happening within that frame of reference.
[24:52] Josh Bongard: Yeah.
[24:53] Richard Watson: disappears up its own orifice.
[24:55] Michael Levin: But also, in that polycomputing paper that Josh and I did, we floated this idea that one way that biology can use this is: let's say you've got some complicated mechanism that you've evolved and you want to make changes and add new functionality. The traditional way is when you start to make changes, it's going to screw up a lot of what you've done before, because there are all kinds of other systems depending on it to be what it used to be. But instead of that, if you can, instead of messing with the machinery, what if you proliferated observers? In other words, you added a system that interprets what's going on in a different way. It doesn't need to change it and thus screw up everybody else's operation. But it provides a new perspective on the existing stuff that's going on and benefits that way. So you can make it less destructive. You get to keep more gains if you do that, I think.
[25:59] Josh Bongard: Yeah.
[26:04] Michael Levin: But I wonder. There's the Landauer, Bennett stuff about how much computation actually costs in terms of energy, and there's a limit per unit of space-time. Because it sounds like by proliferating these observers, you can get more and more out of it. Are we breaking some sort of assumption that all those calculations did, or are we?
[26:35] Josh Bongard: It would be fantastic to go back and revisit all that material. And exactly, are we violating an assumption in there?
[26:45] Michael Levin: Either we're going beyond an assumption, which is completely fine, or that actually is going to end up being some sort of hard limit on how many observers you can have, because it also doesn't seem necessarily true that there's an infinite number of ways to see something as useful computation. But it's pretty hard to see something that somebody gives you and say, I know how to see it in a useful way. That's hard.
[27:12] Richard Watson: I'm not sure about that. That might be limits of computation, limits of precision. I could give you an incredibly intricate program with lots of nested loops in it. You could fail to see the program as a whole and just see some of the nested loops or some particular subset of the nested loops. I think there's a space in which every program is just a number. By recording that number with finite precision, you're missing all of the programs that are in between the numbers. It's a good result. It's a bit like saying how much of the information of a phenotype is in the environment and how much is in the genotype. It's tempting to say you need to have the right kind of environment, but most of the information is in the genotype. But if you think of it as a sort of branching decision tree where the environment presents a sequence of left-rights as you go down that decision tree, then starting development at the right time, or starting development at a different time, would give you a completely different phenotype. Even though it feels like the architecture of the decision tree must have been in the genome and all the environment was doing was providing one variable at a time as though it didn't have much information in it. By changing the timing without changing the genome, you completely change the phenotype because you're changing the data that's flowing into the program, the sequence of data that's flowing in. I feel the intuition that it's that you suggest, Mike, that we can't have an infinite number of different computations going in there because we have an infinite number of different ways of looking at it. But I'm not sure. In the limit, there really isn't any difference between a program and a number. The precision of the number does give you different programs.
[30:00] Josh Bongard: I really like this framing, Richard. Does it map onto the triple helix — the environment matters here? I was thinking about this with the xenobots: is that the natural range of environments that brings forth a wild-type frog? Generally, frog DNA gives you frog.
[30:28] Richard Watson: add pond, you'll be fine.
[30:30] Josh Bongard: Because the pond is an ecosystem that's self-calibrating. Maybe it is this narrow range, this thing that draws out your precision. The set of environments that pushes frog DNA or frog bioelectric patterns into the wild-type frog attractor is actually very narrow when you think about it. There are a large number of other environments that push frog DNA and frog bioelectric patterns into other attractors. That narrow range of environment is this precision thing. I can feed in, I can poke at the frog genotype/phenotype developmental program to get frog, but it's got to be very precise. It's a precise number, so to speak. Once you step outside of it, other things start happening. But we're so used to thinking of the pond as this wide range of things, it's so deep, it's so broad, it might as well be everything. Frog DNA just leads to frog. There's just one attractor. Maybe we had it wrong all this time. I think the three of us are comfortable with thinking of the attractor landscapes when it comes to organisms or species. But what is the range of environments? What does that landscape actually look like? Is wild-type actually a very narrow, almost unstable, attractor or not?
[32:05] Michael Levin: I actually think we're looking at attractors in a whole other space here in the sense that you've got the genetics, of course, you've got the conventional environment that we're thinking about. But you also have got this platonic space of free lunches, rules, laws of mathematics and of computation and of geometry and all this other stuff. It isn't really; you can't pin it on either one of those things. You could call it part of the environment, but it isn't the physical environment. It's all this stuff about if you look at any of these mechanisms and you keep saying, "but why, but why is it that way?" Eventually you get down to math and to just laws or truths of number theory or category theory or something. That's in the environment, but not in any conventional sense.
[33:02] Richard Watson: Imagine if you start from an assumption that there's a symmetry between the amount of information on the inside and the amount of information on the outside, and that that's somehow in balance between the frog and the pond. Then one might ask where the information is held in the pond and how it got there. Think about how the pond has been conditioned by the lineage of frog. There's the sheer improbability of frog working — why the **** should that work? Not just how do you get a frog from a genotype? But now that you've produced a frog, how is that going to be a thing that's going to work out there in the world? It does need quite a lot of things to be just so about the world in order for frog to survive. But the fact that the ecosystem the frog is being born into has been conditioned by the lineage of frog that came before it has pushed information into the environment in a way that makes it just right for frog. And if it weren't, it wouldn't be able to draw frog out of frog genotype.
[34:36] Josh Bongard: Yeah.
[34:39] Michael Levin: That's very interesting. That makes me wonder what that situation is like with respect to the platonic space as well. Is it because I tend to think it's bi-directional in the same way that it's bi-directional with the environment. And another way of asking that is, is there an anthropic principle with respect to the rules of mathematics? People say, if the conditions of the physical universe were a tiny bit different, there wouldn't be any ponds, there wouldn't be any matter, whatever. What about, it's beyond what I know about, but is there a situation like that with respect to some of these mathematical truths that make the whole thing hang together? Is there some kind of improbable collection of parameters that make all that hold together? Or is that really an attractor that you can't get out of? It always just works.
[35:40] Richard Watson: I think it all follows. All of that Platonic space and all of that maths follows from imagining an entity looking at itself in a mirror. That if you can get that far, everything else follows. When there's already quite a lot going on in being able to imagine an entity looking at its own reflection, it sees it somehow. I don't know what all of that means. But once you've got that separation between a thing and a reflection of a thing that's the same in every way except in one respect, that it's reflected through the mirror plane. And now you've got things like there's the set of all things on this side and the set of all things on that side, and everything on this side has the property of being on the other side from everything that's on that side. Everything on this side has the property of being on the same side as everything else that's on this side. And now you've got things like the mirror image of a function, if and only if the mirror image is XOR, and then you say, are those two functions the same or are they different from each other? I don't know. Are they completely different from each other? But they're not completely different from each other because they're the exact symmetry of each other; they're the only two functions which are non-linearly separable over two variables, and one of them is the complement of the other. They know a lot about each other. They're equal in every respect except for the sign. And all of those nested levels of sameness and differentness, I think everything comes from that. And that comes as soon as an entity looks at itself in a mirror. As soon as you make some separation between a thing and a non-thing, and you're making a comparison between them: are they the same thing flipped around or are they different things? As soon as you do that, all of the rest of the maths and all of the rest of the geometry, truth, computation, and everything else follows. Bit of a leap.
[38:05] Josh Bongard: But you need a particular kind of universe to get to the point where you can have an organism that can look in the mirror and see itself, because that's a pretty long loop to go out and come back to and say, that's me over there, or that's me. To me, it seems you need a universe in which that going out and coming back, the recognition that is me, the Hofstadter loop, has a gradient that leads up to that. Because that can't come up de novo. I can imagine there's an observer that's sitting at frequency f plus epsilon and is looking at another frequency at f very close by. Whenever it manages to increase its amplitude, however it does, this thing next door in space and time does the same thing. Presumably that's going on all the time in very simple living systems. It's not hard to look at yourself in the mirror there. In the course of development, you hold on to that loop, but the loop itself gets bigger. It's always closed. But that requires a certain universe in which frequencies that are very close and frequencies that are irrational numbers from one another are independent. You need all those things to establish the fact that there is a gradient that points you eventually, if evolution does enough work, up to primates that can look in a mirror and still see themselves. They're still able to travel that longer loop.
[39:43] Richard Watson: Yeah.
[39:44] Josh Bongard: I could imagine another universe where that's not true. And then forget about it. It's not going to happen.
[39:51] Richard Watson: Mike asked me this question about resonators: I hit one tuning fork and it makes another tuning fork resonate. It puts power into it. How come when I damp the second tuning fork, it doesn't take power out of the first one? How come that isn't reversible? I think that it is. The power that you have to put into the first tuning fork to ring it requires that it isn't damped. When it isn't damped, that's the same thing as saying its power gets reflected back to itself so that it doesn't just dissipate away. When you have two tuning forks that mirror each other, that's actually the same as creating a mirror plane between them. The one tuning fork thinks it's just looking at its own reflection. If you were to take the second tuning fork away, then the energy that was coming out of the first tuning fork wouldn't have a mirror plane and it would just dissipate away from it, wouldn't be bounced back to itself. I take those issues of seeing your own reflection down to quite a low level. I don't need, I'm not thinking about an agent with eyes and awareness and a mind to think about it. Just some sort of ongoing physical process or it isn't. In order for it to be ongoing, there's a certain kind of closure or reflection that protects it from dissipation to the outside, which is the same as saying that the outside is mirroring it, which is why it looks like a reflection plane.
[42:11] Michael Levin: Is a super basal version of that? When, let's say, photons interfere with themselves in a two-slit experiment—when you send one photon through at a time—is that a basic failure to pass the mirror test?
[42:42] Richard Watson: It interferes with an alternate path it could have taken, which is the pilot wave explanation. The movement of the electron is creating a wave which propagates through the other slit that it then interferes with on the other side.
[43:07] Michael Levin: It's like if it's its own alter. So you got a camera facing this way and another camera facing this way. You'll recognize immediately that this one's you; you may not immediately recognize that one's you, right? There was a bomb or somebody had actually a picture of this, a fish and then two cameras from two directions. The one is incredibly obvious, but the other one was, wait, is that the same creature? Or is that me or isn't it? So I wonder, if under normal circumstances, what happens in the macroscopic world is that everything recognizes, "Oh yeah, that's all me. We're going to collapse it all into one coherent action." But by setting up these double slit things, we're breaking that and we're catching it at times where it's not sure that that's itself and it interferes with it the way that you might think that it would interfere with something else.
[44:39] Richard Watson: So if I'm an electron going through one slit or an electron going through two slits, but the other slit's very far away, then do I feel like a newborn on the other side of the slit? No, because my own waves propagated in front of me as well, and they're still there on the other side of the slit, aren't they? That's what causes me to diffract.
[45:15] Michael Levin: That fan-out is what it takes to break the normal cohesion and not interference. Once you've done that, you've punctured that self-world boundary, and so now you're not quite sure about these other things. When the dog scratches himself with his foot and then bites the leg, "Who's touching me?" That kind of thing.
[46:00] Richard Watson: The electron has got a bit of brain damage when it comes through the other side of the slit. It doesn't really know which way it was going anymore. It's bent by its own pilot wave, but not because of the pilot wave that made it through the slit, but because of the loss of the wave that didn't make it through the slit. It doesn't know its own trajectory anymore. And when it sees the other part of its pilot wave that went through the other slit, that's like reacting to its own tail, like the dog biting its own tail.
[46:41] Michael Levin: It's nice that that's the place where observers really became strongly into physics, right? What does the electron see and what do we see watching it? What does the electron say? We gotta get Chris on this because there has to be a macroscopic version of this, like in cognitive space, and it'll be something about the Markov blankets, but this is where that phase and the Markov blanket stuff could come together.
[47:36] Richard Watson: You don't need quantum effects to see the double slit experiment because you can do it with the water droplets.
[47:43] Michael Levin: Right.
[47:44] Richard Watson: It's not quantum stuff that produces that effect.
[47:55] Michael Levin: I bet we could come up with a polycomputing materials vibrating version of this.
[48:03] Richard Watson: Nice. That would be fun. Right.
[48:05] Michael Levin: You can do the double-slit thing. You can do that with; that's already been done with waves. But I'm saying specifically from the perspective of an observer. What's the integral agent?
[48:20] Richard Watson: So the two slits are like two different observers, and the interference between them reflects the fact that the two computations weren't really independent in the first place.
[48:37] Michael Levin: That's interesting. That completely fits into this quantum weirdness that shows up when something is observed. So again, you have this, you don't have non-destructive reads and this idea that as soon as you've added an observer, now everything's different. Things change. I think that's a common result in that field, where when you actually observe something, that changes what happens.
[49:09] Richard Watson: If the observation is a bidirectional resonance between you and the computation, not just a one-way thing, then that would mean that in order for me to observe something through one of the slits, I will alter the computation which is being observed through the other slit. It will appear that the two observations are not independent.
[49:42] Michael Levin: I think there's really something here. With materials polycomputing, we could formalize the notion of an observer and their interpretation feeding back on how another observer is seeing what's going on.
[50:00] Richard Watson: If instead of shooting at a wall with two slits in it, you had a circular wall around the source and the two slits were 90 degrees apart, then presumably the component of the wave that gets through one of the slits doesn't interfere with the component of the wave that gets through the other slit. I'm sure there are reasons why you can't do that.
[50:42] Michael Levin: Here's where the wave–particle thing makes a difference. If you think you're shooting photons, then they won't. If you've got a radio wave antenna in the middle, then it will, because it goes everywhere.
[50:58] Richard Watson: Yeah.
[50:59] Michael Levin: I don't know what happens after that. If you're at relatively low frequencies, then you do go everywhere. Does that matter? When people make radio transmitters, the fact that it automatically goes in all the directions and refracts and gets held up by various stuff, does that screw up anybody else's ability to receive? I don't think so. Is it just too small for anybody to notice? I don't know how that works.
[51:59] Richard Watson: It's interesting that when you shoot a signal down an optic fiber, it gets degraded when the fiber is bent because the reflections on one side are not going the same distance as the reflections on the other. But when you do it in that graded density fiber — instead of a glass fiber that's the same density all the way across and then reflects off the inside walls, you have a glass fiber that has a graded density so that instead of reflecting off the outside walls, it reflects internally. They don't do that distortion. It's not just bouncing off the walls, but the amount automatically compensates for how much curvature, deflection, and time distortion there is, so that the problem just goes away. It feels to me like the difference between the reversibility of the hydrodynamic models — the bouncing water droplet. Did you see that, Josh? The bouncing water droplet experiment: if you have a droplet that's moving around and its trajectory has sensitive dependence, it's a chaotic trajectory. You're keeping the droplet bounce going by vibrating the bath underneath. But if you skip a beat in the bath underneath so that it lands on the back of the wave instead of the front, the droplet reverses its direction. So instead of walking along this way, it walked back that way. Even though all you did was skip half a beat in the vibration. The chaotic trajectory of the path that the droplet has gone through also reverses. It's chaotic but reversible, which chaos shouldn't be. Instead of going the other way, that's also chaotic, and it goes back to the point where it started. That works in the hydrodynamic realm, but it wouldn't work if you think about chaos as discrete events and responses.
[54:45] Josh Bongard: So which of all these physical phenomena is responsible for eventually being able to develop the concept of self and other, the dog that bites its leg? What is the smallest change, what's the universe prime that's one epsilon over where evolution can't get going, it can't establish self and others and reproduction? Everything itself. You can't get enough distance.
[55:13] Michael Levin: I was talking to Chris one day and I asked him what would it take to make a universe in which there were no least action laws, no active inference? How could you get rid of it? Because that starts with his and Carl's framework, it starts very low down. And so I said, how would you get rid of that? And he said, the only way to do that is a universe in which nothing ever happens. So if you had a completely fixed universe, but as soon as you allow change, that's it. After that, you're off to the races.
[55:50] Richard Watson: Nothing particular happens at a particular point in time.
[55:56] Michael Levin: I didn't ask that.
[56:00] Josh Bongard: It seems a little bit of an easy out. Maybe there's a universe in which stuff does happen, but there was a slightly different result obtained from the double slit experiment. This wave-particle duality is not quite dual or it's biased towards one or the other, or then it can't happen. You are able to always retain your history. There's never a separation in time. Things never diverge from one another. Physics is over my head.
[56:31] Michael Levin: But my gut feeling is that kind of thing is going to be an anthropic principle here too, where all of that stuff could have been screwed up, except everything is balanced just right, so that it's not. That's my gut feeling. Although, just thinking back to my crazy question about the anthropic principle in math. In math, maybe it did fall apart, which is why we have Gödel and those things that tell you maybe that's exactly what happened there, is that it did not manage to land on the consistent and complete. We live in a universe where the anthropic principle failed. What's that?
[57:18] Richard Watson: A universe with cracks.
[57:20] Michael Levin: The platonic space failed the anthropic selection filter, and it does have cracks. It looks to us like the girdle and completeness results and maybe Turing limits and things like that as well. But the physical part seems to maybe Bernardo would say something like this: from an idealist perspective, we imagine the universe as nice and tidy because we have to, because it's our approximation of a failed deeper reality, but we've made a nice little neat vision for it.
[58:05] Richard Watson: You can only see the tidy parts, that's all.
[58:08] Michael Levin: Why does it look like everything balances out? Because otherwise it'd be too upsetting. That's why.
[58:15] Richard Watson: To the extent that I'm a tidy thing, I can only see tidy things.
[58:22] Josh Bongard: I would say it's Mike's cognitive frame rate. The cracks are all here, but luckily in this universe, they're small enough that I can skip over them in my cognitive frame rate. It's relatively easy to ignore them. They're in me too. I tend to feel that I'm observing at the level of the organism rather than at the level of the subatomic particles that I'm made up of. And if my comfortable frame of reference was the subatomic plane, then I wouldn't be able to hold on to this. Maybe that's why the subjective experience tends to feel to us at a higher level, because at this high level, you can comfortably not see the small cracks that are down at the lower levels. The agents that are down there, and there are agents that are down there, they can't get away with the illusion that things are neat and tidy.
[59:15] Richard Watson: All right.
[59:17] Josh Bongard: So it's an unsettling thought. They're there, they're aware, and they're suffering hell. They can't make predictions. They can't.
[59:25] Michael Levin: That's not bad. Basically the drive for cognitive complexity and the ability to coarse grain is an anti-stress response. Right? That it's basically "let's get the hell out of this." I can see reality. It's horrible. Let's get out, let's develop a confabulary, world-building capacity as quickly as we can, because it's really upsetting down there.
[59:54] Josh Bongard: That's right. Prey evolved to be big, so they don't have to worry about predators. Maybe this is a cognitive version of that kind of thing. You can escape by just being bigger.
[1:00:06] Michael Levin: I like it because literally to have a size, you need energy and time and energy become limiting pressures. That means you have to coarse grain. You don't have time to track microstates. You have to know coarse grain. That means you have to tell fake stories to yourself about the world and yourself. So you have to come up with these compressed models that abstract from all the upsetting details at the bottom.
[1:00:40] Richard Watson: Don't you think that being an agent in a world that doesn't make sense is the same as not being an agent? To the extent that you can't be a tidy thing.
[1:01:01] Josh Bongard: Richard, you're the expert on endosymbiosis, right? So I might be an agent way down at the lower levels and I still need to survive and reproduce, but I can't predict, there's a lot of things I can't do. So I'm gonna try and lean on agency being developed at a higher level of organization that does whatever I need it to do. I could just go away, but maybe there are agents down there, and they're parasites on us to do the things that they can't do in their messy world.
[1:01:35] Richard Watson: They're agents. They exist as agents to the extent that they are able to find an environment that makes sense to them.
[1:01:44] Josh Bongard: That maybe they can't push against the physical world directly because they can't predict how it's going to push back, but they can push on us and then we push in the Newtonian space and get reliable repercussions and somehow communicate that back down to them. Maybe we are a crutch for them. We, us big things are sitting on this Newtonian island in the sea of quantum effects. And they're in the sea, they can't get out, but they can influence us a little bit. Maybe it's the best they can do.
[1:02:15] Michael Levin: But it's probably calming to be part of a big system that feels like it knows what the hell's going on. And it's part of it and has a reasonable model of what's happening.
[1:02:30] Josh Bongard: But I wonder if there's a testable hypothesis here. Richard could be right. You just can't be an agent down there. You get snuffed out somehow. Can you survive barely, hanging on to bigger, slower things?
[1:02:45] Richard Watson: I'm just suggesting by symmetry that the entity is only coherent to the extent that it is a mirror with a coherent environment. I can't be a coherent entity in an incoherent environment. That's like an oxymoron.
[1:03:08] Josh Bongard: But if the host, the big slow host is 99% of my environment and I hold on to 1% of direct contact with the messy, painful stuff down here, it doesn't have to be a black and white thing. I don't have to give up.
[1:03:26] Richard Watson: The thing that's coherent is the upward visibility of the host.
[1:03:32] Josh Bongard: Yes, exactly.
[1:03:34] Richard Watson: If that's a mirror too, then the host is only coherent to the extent that the parts make sense.
[1:03:45] Michael Levin: Asking how to model it, isn't that what the biology is doing? So at the level of the molecules, it's extremely noisy. Any given molecule comes and goes, even cells come and go. But the ship of Theseus of the body persists for some amount of time. If you track the lower level stuff, it's pandemonium and everything dies and gets denatured and excreted. But on the large scale, you can hold on to a coherent story for a while.
[1:04:20] Josh Bongard: Right.
Michael Levin: And as Josh said, you don't lose the contact entirely. Your retina can sense individual photons, or at least some retinas can, and maybe there's some other quantum biology going on somewhere. So you have a little bit of that still going on, but overall, you've pretty much gotten out of that by raising the level to the point where you're not tracking microstates anymore. I can't see it anymore.
[1:04:53] Josh Bongard: We'd still have to spec out a benefit for staying in direct physical contact with the things at your level of organization. If we're going to say there's an agent down there, there's got to be something that benefits the agent down there. Otherwise, why wouldn't it just completely give up?
