Consciousness Live! with Gualtiero Piccinini (S6 E6)

0:05 Okay, we are live. Uh at least I like to believe that we are live. Um We'll see shortly. So, joining me today is an extra special guest. Um and I wonder if you just might say something about who you are and what you do for those who might not be familiar with your work. Uh hi, I'm Gualtiero Piccinini. Um I've worked a lot on computation and cognition. Uh but I have broad interests in um philosophy of mind and

0:39 related sciences and related areas of philosophy. Uh so, I started out primarily with um investigating what does it take for a physical system to implement computation because I wanted to get to the question of whether the brain and the mind have something to do with computation and um and and went from there. Very cool. So, also before we go too far into any of those interesting um topics you just brought up, you also started the Brains Blog. Um is that correct that you started it or you I know you were

1:13 managing it when I discovered it, but did you actually officially start it? It it started Yeah, I I think it started in 2005 and it was a Initially, it was just my blog and then um Dan Ryder, I believe, suggested turning it into a group blog because there wasn't really one in philosophy of mind. So, I did that and I started inviting anybody who wanted to to contribute. So, it turned into a group blog and then after a while, I gave it to John Schwenkler to manage for

1:48 some time then he gave to others and yeah, it's definitely become a really nice um venue, I think, for exchanging ideas in in our area. I agree. Yeah. No, it was very cool. I was just going to say it's where I got my start blogging. I guess there was kind of a golden era of blogging in philosophy for a certain period of time. Maybe it's not quite as at the peak as it was anymore, but I think it's I think brains are still going strong. It's definitely making a contribution. Yeah. Yeah. Yeah, they it's more structured now, you know, they have

2:21 editors and they organize, you know, symposia and sessions and but it's good quality. Yeah, I've been happy to see also their collaboration with the Neural Mechanisms project. Uh and kind of online mechanisms conference they have. So it's been very cool to see the direction that that's taken, but also like it was very nice that you you invited people and I remember when I first reached out to you and you're like, yeah, come on, go ahead. So you were very welcoming even to junior uh people such as myself who didn't I felt even at that time I was like,

2:53 should I even be posting online? So it was very much a welcome to be felt included in in something um at that level. So that was definitely appreciated. Okay. Absolutely. I mean, I've always tried to create opportunities for more people to exchange ideas and to be included in the in a profession that historically of course, you know, it can there's a lot of underrepresented groups. There's still work to do in that area in that respect. Absolutely. You know, I don't know if you're familiar Are you familiar with the International Society

3:25 for the Philosophy of the Sciences of the Mind? I've seen something about that actually, but I I haven't really paid attention to figure out what it is. I've been doing really busy, but it's on my list of things to figure out what it is. Can I tell you? Yes, please. So um you know, there are these uh more regional and more specialized societies like the Society for the Philosophy and Psychology, but uh as far as I could tell there was no society for the sciences of the mind in general, like the philosophy of the sciences of the mind in general,

3:58 including all of them, you know, psychiatry, neuroscience, um you know, the biology of behavior. Oh, interesting. And um also, there wasn't one that was truly international. So, um so I um got together with Inês Hipólito, who's a uh originally uh Portuguese philosopher. Hm. And we founded this new society, and uh the the goal is also just to keep it virtual. So, it's all going to be virtual conferences.

4:29 Ah. And we've already done the first one last fall, and it went really well. And it the call for papers just came out. So, if you look it up, if you Google, you know, International Society for the Philosophy of the Sciences of the Mind, there's a new call for papers. Okay. Uh the deadline is going to be in a little while. I don't remember the exact date, but there's time. There's time. virtual, too, so it's online. Oh, interesting. Um And the reason is that way anybody from anywhere in the world can participate without having to pay for a plane ticket and accommodation,

5:03 even if they're disabled, even if they little kids, even if they're from the global south. That's true. Um so, even if they're just students, I mean, obviously, there's the selection process, but we try to keep it inclusive. The quality, I thought the quality was really high at the conference. still available online from that first conference? Uh I don't think that they recorded everything. They might have recorded some keynote Oh, so it's actually like web conferencing. Yeah. Cuz like I when I did the online conference, I remember

5:35 which you were part of, too, but it was mostly like, you know, like post on a blog and then discussion kind of the comment section. But, this sounds more like actual face-to-face. Yeah, yeah, it's it's it's a web conference and I um and we you know, we try to have slots that are compatible with different time zones. So, some things you know, is are easier to follow in North America, some sessions and some sessions are easier to follow in Asia. Okay. Um It depends the session and the participants, but we try to keep it accessible to everyone.

6:09 Oh, I'm all for these kinds of ideas. So, I'll have to look more into it. I know last fall I was uh working on alleged book of mine on higher order theories and um and neuroscience of them. So, I wasn't really I even took research leave. I mean, I don't want to we don't need to discuss all that. But anyway, I wasn't even teaching. I was just trapped up here just uh working on that. So, I was really out of touch, but now I'm kind of getting back in touch seeing what's out there. So, this is a very interesting. I'll make sure to put a link to it in the um description of this once it's all up and stuff like that. Yeah. Yeah. So, the ISPS International Society MPS International

6:43 Society for the Mind Sciences. I The philosophy of the science of the mind, so PSM ISPSM. There you go. Okay. Very cool. Uh and it's good to see this kind of stuff. You know, just in just to plug the more written format, I still encounter the graduate students or persons uh that I talked to online who say that they've been to the online consciousness conference. The material is still up there and you know, they'll ask me about this comment when David Chalmers said this and you responded like that, what did you mean by that? I was like, "Whoa." So, it has been interesting to see that the that having that written

7:17 material and the exchanges, the comments and everything up there, people still use it and uh um I find that gratifying. So, uh well as I I value face-to-face interactions as well, I think that there's you know, something to be said for the other. Um but that anyway, I don't want to get distracted by that. But uh people should check out this. Is there any way that people who are not presenting at the conference can attend it? Of course, yeah. And it's free, too. because, you know, it's online, so the costs are very minimal so far. I hope it stays either free or close to free, so

7:50 you just have to register and you can attend whichever portions you want. Okay. Well, I'm very glad to know about this, so I'll pay more attention to it and um keep it in mind. So, now turning to your book, uh which I guess is this book a couple years old now? Or when did this book come out, the physical signature of computation? No, the physical signature of computation is coming out this summer. Okay. How did it come out? is is brand new. Yeah. I I also have a a book called neurocognitive mechanisms that came out

8:23 in 2020. Right. That's the one recent, but at this point not super duper recent. I see. Okay. Yeah, I got I was getting them mixed up cuz I was reading both of them recently. Um It's a It's a lot of stuff, but I find this stuff very interesting. So, I mean, just generally, the idea that the brain is a computer or has some information processing or computes or something like that, I think people like you and I find that commonplace, but maybe you less surprising than you used to, but I I still teach like an

8:55 introductory neuroscience class with for for non-science majors. So, um and when you and I teach it from kind of a cognitive neuroscience standpoint, um when you tell them this, the idea that the brain is a computer, they almost a lot of them find it almost um unbelievable or like comical or, "Oh my gosh, what has science got itself into?" or something like that. So, I I wonder if if you could start by motivating the idea uh generally that the that the brain computes. Yeah.

9:28 Why Why do you think this is an interesting thing to think about, in other words, is what I'm asking. Yeah, yeah. Well, you know, it's a long story, but it does start with a few uh observations and analogies. Um there is a There's a myth that it starts with Alan Turing, but it doesn't really start with Alan Turing. It starts with uh Warren McCulloch and Walter Pitts, who were what we could call now neuroscientists. And um McCulloch in particular was the senior one. He had he had training in philosophy as well as psychology and neuroscience. And um and

10:01 he noticed that that there's this tradition of mechanical manipulation of symbols, including logic, uh going back to at least Leibniz and then going through Babbage and Turing, and he knew about Turing's paper on computable numbers. So he thought, "If we can map that kind of logical machinery onto the brain, we might understand how the brain does it." So in his case, he really did think that it has it's a kind of digital computation that involves manipulating

10:34 language-like symbols, okay? What came to be known as the language of thought later on. This We're talking about early 1940s. Um and from there, once you have full-blown digital computers, you have um you know, there the fact that they're general purpose, that by exchanging a program, you can add and subtract abilities. And you know, our brain is very flexible, capable of many different things. So how is that possible? One possible

11:06 explanation is, well, it has it works kind of like a digital computer, you know, switch programs and you can do different things. Now, the problem with that idea is that even though it's it's uh it's it's plausible uh at least on the surface and um it provides an explanation that we didn't have before, it's actually not very plausible once you look in the brain more closely and see how it actually works. Um so, from that idea, you know, there's there are different routes that were

11:39 taken. One was just focus on the programs, that's classical computational psychology. You know, figure out what the programs are and then you'll have explanations of behavior. But then another route is the one of computational neuroscience, cognitive neuroscience, computational neuroscience that says, "No, actually let's figure out how the brain does it." Um which doesn't mean that it's not doing something like computation, but it might be a different kind of computation. It might not be digital. Uh and it might involve, you know, a lot

12:12 of training and retraining and learning uh of how to um build internal representations and process them at the same time. So, I think the second route is the is the one that um is both plausible and promising. Um and I've defended that, you know, in my own work. Um but there's still a lot of people who are attached to the first way of thinking. Right. And they haven't really quite figured out that we know enough about the brain to realize that

12:46 yes, that's not really the way it works. So, the brain does really work, you have you have to have a more expansive understanding of computation that's beyond just the digital computer. Mhm. And so, what uh Well, there's a lot in what you just said that I want to ask. So, I'm I'm wondering about um going back to these these early guys, uh McCulloch and Pitts, and and the general idea that the way you try to think of computation as you have a string of symbols and some rules for how to

13:18 manipulate those symbols. Um as far as I understand the artificial intelligence movement so far, that's been very, you know, lackluster Uh in producing results. And in fact, I was actually I don't know if you saw this recent talk by by Hinton where he was basically kind of doing what I consider a victory lap over the kind of language models that they're producing and the style of computation actually that they're producing. So, is that are those the two paths the kind of things that you were that you meant to distinguish the the symbol processing

13:51 on the one hand versus the kind of um neural networks on the other hand and the the project of your book is to try to give an account that covers both of them but shows how they're different. Is that is that basically the idea? Well, this new book this new book is entitled the physical signature of computation. That is um something for the most part even more basic. It's Okay. Uh it's not so much about the brain does anything. It's about what it takes for any physical system to implement any sort of computation. And it focuses

14:26 primarily on digital computation. Um cuz that's the arena where this debate has taken place primarily. Um and also there's a lot to say right there. Um but that's not to suggest that that is how the brain works. It's just that we're trying to give a more in-depth account of what physical properties must a system have in order to implement a computation. And and and you know, I'm happy to talk about that but I'm just saying um that's the project in this case.

14:59 There's one chapter about the the mind brain uh and it does talk a little bit about the kind of computation that's going on there. But for the most part, it's really about these like more more basic questions about how many physical systems implement computations, what properties do they have to have and what follows. Yeah. Right. So, I know get into something really, you know, fun and you can find in the popular press sometimes questions about do we live in a simulation, that kind of thing. We know, we end up talking a little bit about that, too.

15:31 Yes, I know. I want to talk about that stuff as well, but so I since we brought this up and we're talking about this, I want to kind of ask you about the PCE, the S, the P in the PCE, which I think are the main most of the actions about the PCE part, which that that the physical computational equivalence requirement that you have. So, since we are talking about this, I wonder could you say a little bit something generally about what you take it to be for something to be computational and for a physical system to implement a computation.

16:04 Yeah. And then I want to ask you a question about it. Mhm. Yeah, so historically there's kind of two camps, those who think that computation requires representation, so you in order to have a computation physically speaking, you have to be manipulating some representations, whatever that means. And then the other camp is, no, you don't. You can have computations without representation. Now, you know, there's all kinds of subcamps, but in this book, we're definitely in the second camp

16:36 and even more significantly, we uncover what we take to be some very important well, especially one, this PCE, physical computational equivalence, it's a condition that is very important and needs to be satisfied to implement computations and has nothing to do with representation. Mhm. Okay, so it's really a vindication of the non-semantic type of accounts of physical computation, which doesn't mean that computation does not in fact manipulate representation more often than not. It does. That's why we build

17:10 computers and our brains definitely, in my opinion, represent things and manipulate representations. it's just that to understand what it takes for a physical system to implement a computation, you have to look at some, you know, intrinsic physical properties um that have nothing to do with representation, okay? Um and I think you were asking me, so what are these conditions? Yeah. Um so, you know, one part of this this project was to get even more clear and precise about some conditions that have

17:43 been proposed before, but maybe fully explicit and and fully precise as they could have been. Um those are we call uh S for state and P for process. S for state basically says that for any computational state of a system that's putatively in a putatively implemented, for any state, any computational state, there have to be at least some physical uh states from some uh lower-level microphysical states of the system that map onto each of the computational

18:19 states, okay? Um and and they don't and and, you know, the physical microstates that cannot map onto more than one computational state, so, you know, each each physical state that maps onto a computational state maps onto only one, but for any computational state, there's something that maps onto it. And then, that's the the state condition. The the the process condition is that for any computational state transition between computational states, you know, in in the automaton that's putatively implemented,

18:52 anytime you're in a physical state that maps onto one of the computational states, and you transition onto another um physical state that maps onto another computational state, it maps onto the right computational state. So, the the the transitions also are mapped to each other. Right. Um those two things are sort of present in previous literature, but they weren't specified as carefully as they could have been. Those that's S and P you just went

19:25 Yeah, S and P and there and there's like more details. So one other very important thing that um maybe other people have hinted at in some degree to some degree, but it wasn't fully specified is that this mapping needs to be done basically digit by digit. Um so if you're implementing a certain automaton and it has multi-digit inputs and states and outputs you it's not enough to have like one physical state mapping onto the whole input or the whole state of the system, you have to have

19:59 sub states or sub states of sub systems of the of the physical implementing system map onto individual digits of the automaton they're inputting. So that's also very important. Okay, but still all of this is not enough. There's one very crucial condition that all that has to be satisfied in addition to all of this and that is physical computational equivalence and that is in you know put simply that um any physical state that maps onto a computational state

20:34 must carry the same amount of information uh about the computational trajectory of the system as the computational state that it maps onto. So no more information, no less information about the computational trajectory. And I can explain a little bit better what that means if you want me to, but that's the statement. Okay, that's the condition. Um And so the basic idea is that you should be able to I think as you put it infer from the physical state the next computational state in a it so that there's a kind of equivalence between

21:07 them. You can treat them as the same basically. predecessors to the extent that you can know them. But no no no worse and no better than you could if you just knew the computational state and the computational description meaning the automaton that's being implemented. Mhm. So I think this idea is very interesting um and a large part of it seems like you're at least in this book you want to distinguish it it from from other versions that try to ward off the spectre of pan computationalism the

21:41 idea that like everything computes. And you know there there been basically people who have posed this as a challenge and then various try people try to salute pose a solution and you find that none of these solutions really cuts at the right level so to speak cuz they can always find a wiggle through it. But then the PCE shuts that door so you can really say that I mean I guess the idea is that if you have a state in the wall that's transitioning between like a a shadow or something you say ah that's the negation function. It's like well you can't really

22:14 infer the negation function from the state of the wall so it fails the PCE criterion. Is that roughly the idea? Mhm yes. Okay. Yeah cool. I I mean I I find that to be a very interesting idea and and there are things in the you know the details that I would ask about but just more generally in this book what I was wondering kind of big picture was what I didn't see a response to the objection from kind of the other side. Um and what I have in mind so to speak is kind of the Kripkean um objection. Uh you know Kripke kind of

22:50 argues that may maybe you could put it that you know even your calculator doesn't compute. Um so that if the famous argument that if the calculator fell from the sky you could wouldn't be able to tell what what function it was computing um because the state that it's in is always consistent with different computations that might have been implemented by different designers and without access to knowing what they specifically had in mind when they designed it you can't say. So I I'm just curious if this is

23:23 something that was meant to be left out or something that's uh um you didn't think needed to be addressed. I mean I don't mean this as any kind of like attack on your view. I am just curious cuz I'm interested in this argument coming from Kripke and I used to wasn't in there. Yeah no that's a very good question and before I answer that it just became painfully aware that we haven't mentioned my co-author. So I Oh no right I forgot he has he has a co-author. I didn't write this book on my own. I wrote it with Neil Anderson who's a brilliant physical

23:56 information theorist at UMass Amherst. And I definitely could not have done this on my own. It builds it builds on his previous work in physical information theory. I'm happy to say a little bit more about it at some point if you want me to but I just want to make sure everybody knows that yes I co-authored this book. I definitely put a lot of work into it but I did not do it on my own. Absolutely you should have made that clear at the beginning 100%. So with respect to the Kripkean objection I if I remember correctly I'd say a

24:29 little bit about it in my previous book the 2020 book. Yeah that's what made me wonder why it wasn't in this book. Yeah um I don't know that there's like a very great reason but I didn't think that it was uh important enough to address in this book because what I think Kripke didn't fully uh appreciate is that computers are built from components. And so, once you go down

25:02 to the simple components, the logic gates, there's not really that infinite potential for, you know, infinitely many possible outputs from infinitely many possible inputs. Um, like any actual physical machine, strictly speaking, is a finite state automaton. So, and also, it can be decomposed into even if it has the, you know, the architecture of a general purpose computer, it can be decomposed into these

25:34 primitive computing components. And the primitive computing components can can be tested to death, and they get tested. They get tested, you know, um, carefully enough that that engineers can appreciate their reliability under various conditions. So, then they can build integrated circuits. And the integrated circuits, they have billions of these components. Um, but they are configured in very specific ways. So, you can, you know, if you wanted to, you could reconstruct exactly the exact logic of the whole design and which functions are are computed and or

26:08 supposed to be computed by each individual component and then by the whole processor, um, once it's all put together. So, um, you know, there's interesting epistemic questions there, but there's not really the skeptical challenge of, "Oh, what, you know, what does this really compute?" Um, you know, there's all kinds of questions about under what conditions does it work properly? Um, was it designed properly? Was it built properly the way it was

26:41 intended? Um, does it malfunction under certain conditions and not others? Is it being used properly? Um, you know, if something fell from the sky, you'd have to kind of open it up and see, you know, what are the components and then you test each component and then you figure out how they're hooked up together. You don't just like take it as a black box. Yeah, if you took it as a black box, you'd have more problems, more serious problems about figuring out what it does. But if you could open it and look at the mechanisms, I mean, eventually you would figure it out what the components are

27:14 and how they're put together and um reconstruct its own its whole um logic design, if you will. Yeah. So, Yeah. Yeah, I mean, I guess I see what you're saying that you just don't take it as a you take it as a skeptical challenge that's I guess you're saying epistemic. Um not not really a metaphysical challenge to say that this uh but I I I guess I that Kripke as I read him did interpret it as a kind of metaphysical challenge. Um but uh yeah, I'm never I myself did I never

27:48 know how serious to take these really skeptical challenges. But it it does make you wonder that when it comes to the brain, it is kind of something that fell out of the sky in a way. We don't have access to the um Yes. design scheme of its logic level the logic gate level description of what's going on in the brain. We're trying to discover that. So, I do think that when it comes to thinking about the brain and what if any function it is computing in any given case that it it's kind of more um

28:20 it presents more of a realistic challenge and and here's one way I would say say again. Oh, sorry. It's a huge challenge. I mean, it's very hard very hard to figure out what it does really. Yeah, exactly. And but but you don't feel the force of the idea that whatever function we think we're ascribing to the brain that's compatible with that series of states that uh there was some other function that it was computing it just so kind of overlaps with this one up till now which I guess is one way of kind of hi at a what I would think of as a high-level summary of what Kripke's point is is

28:52 that there are lots of functions that partially overlap. Um and so we may be observing a re- a region where there's this overlap and then extrapolating to cases, since it is finite, since the brain is a finite state automaton in some sense of that. Um we're extrapolating and and this is the idea that you always have to give an idealization of the physical item in question in order to make this mapping between the computational state and the physical state. And every time you do perform that idealization, part of which is what I see in the PCE, then you're

29:26 making some assumptions um about which function, which computation the system is computing, because otherwise you wouldn't know how to idealize it. Um and and what's uh what computational level structure you'd want the mapping to map onto. So there is a sense in which you kind of do assume. And I and I mean, yeah, I'm not sure what to make of that, but uh I wonder what you think about that that way of putting the challenge. Yeah, especially in the context of the brain. Yeah, so with respect to the kind this kind of like global

30:01 skepticism about knowing what function is being computed, um I I I'm not really moved by that. I think that it is a matter of figuring out the details, you know, start with the basic components, see what they do, how do they put together, you know, what do their com- composite systems do, and so forth. Um and PCE does help uh because um you do need to have a certain kind of uh informational structure in the physical system for it to genuinely

30:36 implement um a computation or another. But then, when it comes to um studying, you know, a super complex system like a brain, yes, there's going to have to be some uh idealization involved. There's going to be some abstract a lot of abstraction involved. You're going to have to identify relevant levels. And And in fact, even in neuroscience now, there are debates between people who focus want to focus more on uh computation at the level of

31:09 individual neurons and circuits and networks and people who want to look at uh higher-level variables. Um um they call them manifolds. And um so, these are abstractions over like larger systems or larger networks of um of neurons and maybe combinations of different systems to into larger super systems. Um And so, some people are more drawn to understanding neural computation at this

31:42 higher level, and other people are more drawn to understanding computation at more lower level. Um I think they're compatible ultimately. It's not like they're mutually exclusive and you have to pick one, you know, it will just like a digital computer has all these uh levels of organization, they're all computational. Some I mean, you can go below the computation and look at how they're implemented, but even with with once you reach the computational level of description or explanation, you start with the logic gates, and then

32:15 you build and build and build and build, and once you have billions of them, you have to talk about processors and memory components and um and there's lots of levels in between. Uh and this is all still machine language level. It's We're not even talking like operating system and application yet. Um so, the brain might well have lots and lots of computational levels within it, you know, like, you know, it's smaller uh, simpler computing components that are put together to make to build

32:48 larger and larger computations and representations, you know, representations are part of the story as well. Especially in the brain. Okay, I mean, since you brought up the idea of having these components realizing them, I did want to ask you about your account of realization and the way that you try to um distinguish that from uh, I'm blanking on the term, but I'm sure you'll remember it, but the uh The flat view and dimension view? Yeah, so the idea that well, can you tell us about how it is

33:20 that these computational properties are going to be realized? That's the basic question that I want to ask in a physical system. Um And how the intrinsic components of the physical system are going to play a role. So, I I'm not sure that I fully understand this question. So, there is a there's a If this is the same as the question of how is computation implemented, it you know, it's the stuff that we talked about before, but there is a broader notion of realization that's, you know, used in many debates in philosophy of

33:52 mind. Um, and in that regard, I think the best way to understand realization is as a relation between properties of larger systems and properties of their components. Um, and, you know, my account of realization, which I developed in uh uh, collaboration with Corey Maley says that a higher-level property is realized by lower-level properties just in case it is an aspect of the lower-level properties, the

34:26 properties of the components of the system that you're looking at. So, you know that could you have a high level computation like operating system level computation or um manifold computation in in the brain. Um it's might be realized by a lot of processes by individual neurons or circuits of neurons if it's an aspect of what they do. So, an aspect is sort of like a part, you know, like a like a proper part of this larger ensemble of properties. That's what realization is.

35:01 Yeah, that's what I meant to be asking about. So, I think that's uh sorry for putting it in the confusing way, but I'm curious about how this account of realization fits with the account of implementation. So, that's what I I am curious about how the two relate. So, when you say um that the that the brain is implementing a physically implementing a computation, but at the same time there are multiple computational levels. Um cuz I'm a bit and then each computational level has as an aspect

35:35 something from its previous level. I am a bit curious about how this is all supposed to fit together. Yeah, yeah, good, good. That's great. That's great. You're asking wonderful questions. Oh, sorry. I'm just confused. I don't mean to be asking good questions. No, it's because I mean there's so there's so many aspects to all of this and there's so many things to say and you can't say it all explicitly in every possible way. So, talking about it can really bring out some of the uh interesting details. So, the way I think about it is is let's start with like thoughts, okay? Like you're thinking about stuff and

36:09 reasoning about something. That's a very high level computation uh or information process or whatever. And um it's going to be you know, realized by lower level computations. So, you know, that's an aspect of what a lot of neurons are doing to get combined, you know, um collectively. Um but still, even though we call it neurons, they are sufficiently abstract that you could do the same thing in a large language model, or at least like something

36:41 similar enough. You know, it doesn't have to be biological. We're capturing something about the computational and representational aspects of the mind or thinking, um and it could be replicated, at least to some degree, in a model, in an artificial system, especially if you actually build artificial processing units, aka neurons, as opposed to building a simulation in a digital computer. Right. That's a whole other kind of worms, but Yeah.

37:12 Okay. So, we're like realizing high-level computations by lower-level computations. Then, at some point, once you have reached like let something that would constitute primitive computations in the brain, let's say, or in a in a computer as well. Now, you can realize that by implementing it into a technology. Now, you have to bring in more concrete properties of the physical system that's implementing the the computation itself. At that point, you know, in the in the

37:44 brain, you're looking at action potentials, ion channels, you know, electrical discharges going through axons and dendrites, and neurotransmitters being being released between the you know, in the synapses between neurons. So, you're going to have a bunch of electricity flowing, I know, I like charged ions going in and out of membranes, chemicals released in um synaptic clefts. That's one thing, but you know, if you were to build it in a silicon circuit, it would be made by other kinds of things, you know, it's just that it

38:17 would be maybe having some of the same uh computational properties, you know, for example, same firing rates or timing of individual spikes um might be might be analogous. And so, that's why you might want to say it's performing the same computation, just like you know, you can build a digital computer out of um you know, silicon microchips or you know, gears and um mechanical components like Babbage's analytical engine or electromechanical components

38:51 components or other kinds of uh physical uh media. So, yeah. So, But, there's realization realization realization, but at some point there's also implementation, which is a kind of realization, but it's a special kind because you're going from these um computational properties that are just basically degrees of freedom of the system to the uh constitutive aspect physically constitutive aspects, you know, the specific forces, the specific uh kinds of components, you know, kinds of

39:24 physical components that actually do the um implementing. Yeah, okay. So, um I see I there's a question from Johannes uh Kleiner in the in the comments over here. I want to ask that to you in a second, if you don't mind. But, I just really quickly, so which level does the PCE apply to? Because I thought that in order for there to be a computation, there had to be some kind of very strict equivalence between the physical state and the computational state um defined in a way that we talked about previously. But, now there's levels. So, where do I go to the physical state and

39:59 infer the process the computational properties of the highest level? Yeah, so it applies like I would want to go to the next lower level, not the physical level, but some like more abstract computational level. Well, so where PCE applies is in the connection between the kind of um primitive computational level and the implementation of that, you know, physical implementation of that. Um everything else is taken care of automatically. Meaning anything any

40:33 higher level computations are going to still basically reduce to that primitive computational level of like in the digital computers it would be logic gates. You know, you have like let's say AND gates, OR gates, NOT gates. So, once you have once you implement those faithfully in a way that respects physical computational equivalence, then, you know, all the higher level computations are already taken care of and they're all implemented faithfully. Now, this is an approximation in real life, okay? So, this might not be perfect perfectly faithful implementation, but

41:05 it's it's close to it's very close to perfect in digital computers. They're pretty um well made at this point. Okay. So, I want to come back to that, but really quickly there's a related question. So, Johannes asks, says, "Amazing conversation. Thank you. Uh is realization of one computation by another a matter of realizing qua implementation of both computations by the physical system?" Sorry, can you can you say that again? Yeah, so is realization of one computation by another a matter of realizing qua implementation of both

41:39 computations by the physical system? So, I think this is kind of a related question to the one that I was just asking actually. I If I understand the question, I think the answer is yes. Meaning I meaning yeah, there's higher level computations they're going to be realized by lower level computations and and those lower level computations are implemented by some kind of physical machinery. Well, when you do that, you're also implementing all the higher level ones as well. Can you infer from the physical states of the physical system that implements

42:12 the lowest level, the states of the highest level computation. Yeah, you you could, you know, it's just that when it when you have billions of components, and actually making inferences like that would would take a insane amount of time. But yes. Okay, but you are committed to that because it sounded like to me what you were saying was it's just that that lowest level that you make these inferences, and that at the these other higher level implementations um maybe you don't. No, you Well, you could. You could.

42:46 Okay. Um yeah, so like time and resources, you could. Yeah, that's Okay, that's very interesting. Uh and so this overlapping aspect thing I wanted to ask you about, and especially because we're going to be talking about consciousness here, so we may as well um Oh my I can't The time has been flying. I haven't even realized Keep my eyes I know I can't keep you forever, but uh um I do want to make sure we talk about consciousness before I run out of time here.

43:17 Uh so before we do bring up consciousness though, is there any aspect of the physical implementation of a computation putting aside the consciousness bit which is non-computational um that or that uh plays a role, but is not computational? Oh, there's a ton of that. Uh so, one of the things we stress in the book a lot is that um

43:49 to find, you know, an implementation of a digital computation in an ordinary computer, you have to disregard vast portions of the state space of the actual physical system that implements this the computation. So, um this is something that, in my opinion, has been really under-appreciated by uh previous uh philosophers or philosophical literature. Um So, I mean, some of it is really obvious. Um you know, computers have cases, they have fans, you know, they have um electricity flowing that needs to be

44:22 taken from the electrical cord. Um So, none of that is actually implementing any computation. It's sort of all supportive stuff. But even within a processor, there's tons of stuff that happens that is never mapped onto any computational states. Um So, there's a lot to say about that, but yes, the answer is yes. Okay. So, how do you know which to rule out? Um Cuz I I remember I'm trying to get my crypt key puzzle back in here

44:54 and then find another method. Right. So, I mean, obviously, we're talking about artificial systems, so they're designed for this purpose. Um so, they're to the brain though, I mean, cuz I think the real force of this is we don't know what level to focus at. Yeah, but that's that's another reason why the brain is not is not doing digital computation. Um and why, you know, classical cognitive psychology and cognitive science uh where to the extent that they were assuming that it's some kind of digital computation, you know, language-like or not is undermined by neuroscience. But it's

45:29 also undermined by a fully appreciating how digital computation really gets implemented in physically. Um So, Wait, are you saying you that you just Are you saying that these issues don't apply to analog computation? Well, they don't apply the same way or or even I you know, I think neural computation is it's sort of its own kind. All right, we haven't got to that yet, right? I I use the term neural computation, which I take from computational neuroscience. Um but yeah, Corey Nailer, for example, you know, he prefers to call it analog computation and he draws an analogy with

46:02 certain kinds of other kinds of machines. You know, I think that can be helpful. Um but you don't have this exact same uh kind of issue of disregarding some of the states that you do with digital computation. So, the beauty, the power of digital computation comes from the precision that you can achieve by stringing together all these digits. Uh but to do that, you know, you need to have have these states to be stable. Um you know, so you can recognize them, you know, reliably as you zeros or ones,

46:37 you know, if you have binary code. Um so, to do that, you know, you need to stabilize the state of these components. But components, you know, like physical components in a in an electrical or electronic circuit, um they're not naturally stable. So, you have to do a lot of fancy engineering to to keep these things stable enough during the right kind of tan time intervals. Mhm. And uh and therefore, all the stuff that happens to make this stable has to be disregarded as not really part of the computation itself.

47:10 So, um you know, the state of a capacitor can fluctuate, you know, in voltage and it only counts once it's stabilized around either like the zero level or the one level. Um and so, you're going to have this signal, it's called a clock signal, that goes through the machine um you know, like lots and lots of times every second that determines when you can check the state of these components and so therefore, when it counts for the transition to the next state. But but

47:45 when you're so when you're when you get the clock sign uh signal, it counts. When you're waiting for the next clock sign signal, it doesn't count. So, it's all built into the the how the machine works. Um that the the times when it counts and the times when it doesn't count, the places where it counts and the places where it doesn't count. So That's very important physically implementing digital computations. Yeah, I agree. But it's just so since we brought it up, can you tell us what is your view of neural computation that's distinct from

48:18 like what people like Cory want to call analog computation and also digital computation? Well, so I I'm okay with calling it analog computation in Cory's sense. I don't think that uh we have to call it analog computation. We can just stay with a neural computation. That's the term used in the field. Um and I think there is a more restrictive notion of analog computation that's also useful and and neural computation is not analog in that sense.

48:52 Um but one one one thing to point out to understand how different it is from digital computation is that there there is no clock signal and there's nothing analogous to a clock signal in the brain. Um there's a lot of synchrony, but it's a synchrony that comes in degrees. It It's very important. I mean, it's not fully understood exactly how much, but it's understood that synchrony between components is very important to neural processes and neural computation. It's just that it's not it's not synchronized by like a

49:26 centralized signal that says when things count and when they don't. Instead, it's sort of self-organized synchrony that comes in degrees and and it it counts even if it's out of phase, you know? So it's synchrony as in like, I do this, then you do this. I do this, you do this. So it's not like they're happening at the same time, but they're just happening in partial synchrony. So they kind of oscillate together. There's a lot of this in the brain and it comes in degrees. So, yes, it needs to be understood, but it's very different from the kind of

50:00 uh precise uh guaranteed synchrony driven by a centralized clock that happens in in a digital computer. So, uh is it just one difference? There's other differences, but um yeah, it's uh another important difference is that that the signals are also graded um but neither fully continuous nor fully discrete. So, you know, in in a digital computer, you you would have finally many possible states, period, you know, typically two, zero and one.

50:35 Uh in a analog computer in the strict sense or the restrictive sense, you have these fully continuous signals that you have to integrate, but in a typical neural system, you have spike trains, what matters is the frequency of these signals um of these the the the number of spikes per second and that can vary and they do need to get integrated to some degree, but it's a whole different um process of integrating these um inputs because you have to sort of combine

51:09 all these different input frequencies to decide whether they pass a certain threshold and then, you know, your neuron is going to fire again. And so, um so, you really need to understand the way this system works in its own terms, I think. And you don't find the idea of trying to map on the way that digital computation works onto this structure um would be fruitful or useful. Uh can you say that again? I'm sorry. Yeah, so, I mean, I I agree with you that So, it sounds to

51:41 me like what you're saying and what you've argued for in in your work is that um we look at the brain and we find well the things that you've been describing and they don't clearly map on to the notions of in digital computation for various reasons and you just listed some but there is a way of like making it the case that it does in a way if you disregarded certain states of the system and if you impose a kind of clock I mean maybe we'll discover one

52:16 maybe you know you said timing isn't exactly going to play that role but maybe there's some as yet to be discovered way in which these are uh clocked and many people have and I you know I just want to hear your point of view I'm not really pressing this but you know some people look at the binary nature of the action potential itself and at the reaching the threshold as a way of reading the binary states and so it's an open question whether it's synchrony phase locking or whether there's a kind of code in the in the spike trains themselves or at a at a

52:49 lower level maybe it's in the dendrites you know maybe Stu Hammerhoff was right man it's all in the dendrite so I think a lot of these questions are open but isn't it still also open that we could map on to the functioning of the brain or discover that it did implement digital processes even in all this messiness or do you think that's just ruled out given what we know there's no possible I think that it's we know enough to to conclude that that is not a fruitful

53:25 way to pursue or like that that trying to find digital computation in the brain is not going to is not going to work and it's not going to be a use fruitful way to invest resources Um, plenty more that we need to figure out about brains do work. But it's not going to be digital in any useful sense. Um, now now one important partial exception is that there are going to be

53:58 some high-level processes like a neural processes that approximate things with some kind of quasi digital structure. For example, when you process language, linguistic inputs. You know, linguistic inputs, they're still going to be encoded in these complicated uh, neural manifolds, but um, but at a high level of organization or abstraction uh, they're going to have something corresponding to the kind of quasi digital properties that language

54:33 has, you know, like the human language. Um, you know, it's not going to be written in some kind of digital code, but it's going to be encoded in in neural states. Uh, and you know, even oral language, even though like when we speak even though it's all continuous vibrations in the air, you know, our own brains are able to kind of parse that into words. Okay? So, words are have a kind of digital structure and even within them they have kind of phonemes and then and within sentences or paragraphs they have

55:07 they have their own, you know, distinction between, you know, word by word and also even higher level uh, grammatical structure. So, yeah, of course some of that is going to be encoded in um, in neural states. So, you know, you know, in certain cases there's going to be something that has quasi digital representational structure. Okay, interesting. So, Johannes is asking again in the um, discussion if uh, neural computation for you is a kind a case of Turing computation, but it sounds like the answer is going to be no

55:39 given what you've been saying. Uh well, it's it's going to be uh presumably Turing computable. So, in the sense that you could approximate it you know, with respect to the functions being computed, you could approximate it with a um with a digital machine like a Turing machine. But, it's not going to be digital like a Turing machine. Okay. So, so the brain doesn't implement some

56:11 version of a Turing machine either then on your view. It it can't be theorized in that way. No. But, you know, there've been these debates about whether the brain has like hyper computational abilities, can it you know, can it compute functions that are not Turing computable? Right. And I don't believe that. Although, there are a lot of interesting subtleties there. Um and we you know, we could talk about the mathematical objection to machine intelligence. This is something that Turing talks about. Uh but, um Yeah, I I do want to talk about that,

56:44 but I I want to get back to talking about the brain and how it you think it computes because um I still So, it it just to kind of sum up or I remember I was trying to get my understanding of this uh up to yours. So, what you're saying is you're happy to describe what the brain does as a kind of analog computation. Or in any ways, whatever you call it, it's not digital computation. And one of the reasons why it's not digital is because it doesn't

57:18 well, the pro the physical processes that the brain uses to do what we think of as its job don't seem to be the kind of processes that map clearly onto the structure we would expect to find in a digital computer. Is that a way of sort of summarizing the case? Yeah. Okay. Um yeah, very interesting. And then in the case of the brain specifically, now we're sort of turning to questions about consciousness, um I I know that you end up kind of saying in in this book, especially in chapter nine, which uh of the physical signature of

57:50 computation book, where you talk about the mind and consciousness in particular, but you end up saying that consciousness is not computable. Uh so, in other words, that a simulation of the brain wouldn't have phenomenal consciousness, or at least that's the way I understood it. Um so, is that First of all, is that the right way to understand the thesis of that part of your book? Yes, except I wouldn't say it's not computable, because that makes it sound like it's hyper computational. Okay. Um that's not what I'm saying. Um Well, I mean, it's not a computational process, fully.

58:22 Yeah, it's not just a matter of computational processes, yes. There There seems to be something else going on as well, meaning that's not just a matter of computing. So, is it the case that a simulation of your brain would have consciousness, according to your view? No. No. Why not? I mean it depends on what you mean by by simulation, but if it's a digital if it's a simulation Yes, I should have specified. A simulation in a digital computer, no, it wouldn't.

58:56 Um But you But you did just say that you could turn that these things could be, you know, approximated by I guess the key word is approximated, cuz you they can't be actually no, but I mean what I mean is that, you know, neural computations could be approximated by digital computer, but consciousness is is about more than neural computation. Right, no, I understand that. Uh what what I was saying was more that whatever else is going on, you know, it it's not automatically

59:30 reproduced in a digital simulation of neural computation. Can you give us your reasons for thinking that this is the case? Yeah. It's really a version of some of the classic objections to functionalism or computational functionalism. Um and you know, there's a Mary the color scientist and all that kind of stuff.

1:00:02 Um the the way I think about it is that there are um causal powers that objects have and there are qualities or categorical properties that they have and there's interesting questions about exactly how those those two things go together and fit together. But, we need both to account for our ontology in general. It doesn't even have to involve consciousness. Um

1:00:35 so, I think that ultimately computation is a matter of causal powers. In fact, they are abstractly considered causal powers. They are basically degrees of freedom and how they're organized together to generate more states from previous states, you know, inputs and and states generate other states and outputs. Um but, to get consciousness, it seems to me you are going to need the qualities, the categorical properties. Um you need that for to account for the

1:01:09 qualities of the world and you need to that to account for the qualities of our um experience. So, exactly how they fit with the causal powers is a interesting question that I haven't really thought through. Mhm. But it But it's not going to be just a matter of having certain causal powers, period. It's going to have to be combined with the right qualities. So, if you have the right qualities, great. But what it takes to have the right qualities, I don't know. So, I don't

1:01:42 think that just doing the computation, which is the causal power certain causal powers, in fact, very abstract causal powers, is enough to get you the qualities automatically. Uh maybe maybe under certain circumstances, you know, if the causal powers are organized the right kind of way, the qualities come come with them for free. Uh but the that remains to be seen exactly what that what those conditions are. And and maybe they're not just come for free. Maybe you need some something different, you know, you need certain chemicals in the

1:02:16 mix or you need certain degrees of synchronizations between the components or, you know, some some configuration of the components. Um It's It's actually very hard to think about that stuff. Yeah, I I agree. But uh what I'm wondering is why so um you So, for for example, David Chalmers is a is a dualist, um but he's a computationalist in the sense that he thinks the laws of physics guarantee that the right computations

1:02:51 bring around the right qualities. Um those qualities aren't identical to the computations, but the computations being instantiated in any way I I don't as far as I understand his view, um almost anyway, I think maybe has some some stuff to do with Russellian physicalism I come to play. But barring that stuff, um if the computations are instantiated in whatever medium, then that there the qualities come. So, I I wonder what what objection is to that way of thinking about pairing the two. So, you get the qualities and

1:03:24 you get the computation and you get a law that connects them. So, you get to be a non-physicalist. I mean, maybe you don't want to be, but it sounds like I don't. I mean, I'm a physicalist, but I also um don't see any reason to believe that once you have the computations, you have the qualities automatically, you know? What about your standard So, yeah, what about your standard dancing and fading qualia type thought experiments where you replace a part of your brain with a computer let's even say a analog computer or a kind of neural some

1:04:01 kind of, you know, artificially instantiated device um that that did what the brain does, but not wasn't biological. Uh so, so you could have different versions of this. You could have the Well, it's a digital computer you replace your neuron with or it's a weird, you know, neural computer that's not biological. Um So, so what is your response to the classic fading qualia argument? I am curious about that. Do you just think that these kind of a priori things don't matter or Well, I definitely don't think we can

1:04:33 figure this out a priori. Um but um yeah, with respect to fading qualia um the question is you replace you suppose you replace a a bunch of neurons with some artifact that computes the same functions. Then the question becomes does that computation is that computation enough to also generate the experience that we normally have you know, when we perform those computations or or does it not or is that not enough?

1:05:07 And as far as I can tell, we don't we don't really have a good answer to that question. Um, we don't fully understand what it takes to um, generate the experience itself. But, one thing we know is a lot of these computations can be done unconsciously. Even in the brain, a lot of computation goes on unconsciously. So, just because you're replacing the computations doesn't even to me suggest that you would automatically replace the experience. Maybe you're replacing the

1:05:41 computation period, and the consciousness goes away because it's it's involved something else that we don't fully understand. Maybe you need certain chemicals in there or something, you know. And so, once you have the the the the chip instead of the neurons, you do the same computations, but without the conscious aspect. Um, I'm not saying that's the case because I don't know. I'm just saying I haven't heard a good reason to think that just because you're replacing the computation with something that does the same computation, thereby, you also, you know, replace the experience with an

1:06:14 equivalent experience. I guess the standard reason would be that to deny that puts you in a very skeptical position with respect to your experience right now, having to admit that you could be having dancing or fading qualia and not know it as we're speaking right now. And that seems to be a kind of radical kind of skepticism that you could be cognitively out of touch with your conscious experience to that degree. So, so the non-skeptical conclusion is that the experience is the same between

1:06:47 the the replacement. Yeah, I mean, it's so I think I understand why you say that, but it seems to me that that begs the question in the sense that, you know, it's built into the way the experiment is, you know, the thought experiment is is is defined that there's no noticeable difference to the subject. Right. Or at least or at least like you they wouldn't be able to express it linguistically. Um but that is an assumption that is not really motivated empirically in any way. Like there's no reason to believe that

1:07:19 if you actually did try to build an artifact that replaces some neural computations and can be plugged in instead of the neurons that there would not be anything noticeable, any noticeable difference to the subject. So, I mean I don't I don't accept that part of the thought experiment as given. I think that that's questionable and um it might not work that way. Right, because in a way like it to accept that uh that that assumption is already to think that the um

1:07:53 the they have the same causal powers, but I think part of your thesis is that it's not clear that this is part of the causal power of it. Absolutely. So, you know, yeah, you can build the thought experiment that way, but then now we don't know if all it takes to keep that experience alive is to just do the computations or you would have to have an artifact with additional physical properties that are the ones that are required to preserve consciousness besides the computational ones.

1:08:25 Yeah. Okay, very interesting. And so, I like this idea that you could be um a physicalist and still think that there's some kind of quality involved, but I am curious about why you say that we need it anyway. So, uh we and you said not even about consciousness. So, when you're talking about physical qualities, uh can you tell us what it is? And I know you talk about this a lot, but uh can you tell us uh what you mean so we can discuss a little bit? Yeah. Yeah, so um I think I mean the same thing that somebody like John Searle means or even somebody like

1:09:01 David Lewis or um David Armstrong mean, you know, when they talk about categorical properties. Um now, they may or may not have been thinking about consciousness at all. But Well, Armstrong was. And yeah, Lewis, too. Oh, yeah. I mean, they definitely did think about it, but you know, they were also concerned with in this case, yeah. questions about yeah, metaphysics. And so, they actually tend to be categorical as they think that properties are categorical first and then, you know, dynamics or behavior come out of that and maybe in

1:09:34 combination with the laws of nature. Um and I, you know, I don't I don't go that far, but um You're not a truth-maker theorist. What did you say? I said, "No truth-makers for you?" No, truth-makers are fine. I'm just not a not necessarily a categoricalist, you know, about properties. I don't think I don't think that properties have to be just categorical and then, you know, behavior and dynamics come out. So, there's some structural dynamical properties. might be there I mean, I like I like this um

1:10:06 hybrid view that properties have quite categorical and dispositional features, both of them. Um So, but exactly how they go together, I don't know. Um But I don't think So, what I don't think is that dispositionalism is viable, that it could just be dispositions or causal powers all the way down. Um And I think that I think that maybe David Chalmers sometimes thinks that way or some of the people A lot of people in philosophy of

1:10:38 mind think that way. They They just discuss the things in terms of causal powers, you know, high-level causal powers, lower-level causal powers, how are they related? It's not just causal powers, though. There's There's categorical property somewhere. So, no, you're against structural realism or ontic structural realism as it's sometimes put. Yes. Okay. So, does this put you in the category of the Russellian monist? Is there a way to say what you're saying without being in that category? That is a rather excellent question. Um I think Russellian monism is often

1:11:14 described in ways that are Mhm. too consciousnessy. Maybe not entirely faithful to both what Russell actually thought and um maybe what I think is the best way to think about it. Um I No, I know there is this um there's a paper by um Barbara Montero called Russellian physicalism, I think. Yeah. And where she argues that Russellian monism really ultimately is a

1:11:47 is a kind of physicalism. I don't think she's the only one who says things like that, but I remember reading this paper a while ago. Yeah. And um I think that is the right way to think about it. And and and you know, the little I've read of Russell himself sounds to me like that. Mhm. Um so, to me it sounds like what all he's saying is there are these categorical properties in in the physical world. Science cannot tell us exactly what they are because all we can see is how they interact with each other. I mean, how

1:12:20 objects interact with each other. We can't just kind of go inside the object and then feel how it is inside, so to speak. Um but that doesn't mean that these things are conscious or mental or or they have qualia or even proto-conscious or proto-phenomenal, if that means anything. It just means they're categorical. They're not they're not just dispositions. Yeah, we observe the dispositions, but from the dispositions we infer that there's something

1:12:54 categorical there that's manifesting these dispositions in in certain under certain conditions. So, um Mhm. Yeah, interesting. That's all that Russell is saying. And then, you know, yes, once you put together a lot of different things in complicated ways, you might develop, you know, conscious experience, but that's not something that's sort of built into the fabric of microphysics. It's it's something that emerges under special circumstances. Yeah, very cool. And I'm a great fan of

1:13:27 Barbara and her work. You know, she was at the online consciousness conference from back in the day as well. So, I'm um a fan of that paper. But, I'm So, this is what I'm curious about because when you say that consciousness has a a non-computable aspect to it, and then also you have this kind of general Russellian picture in the background. It does sort of sound like there's some proto-consciousness in these in categorical properties or something. So, is it your view that these categorical properties are important leave related to consciousness

1:14:00 uh in any way? Because it it does sound like the answer is yes. Well, I think they're related to, you know, everything. To everything, right. They're related to everything. It's like the world comes in many levels of organization. Um and there's new stuff, you know, um being created all the time, you know, by organizing uh components. So, Right, but I mean in the in the Sorry, go ahead. Mhm. No, it's the elementary particles form uh atoms, and the atoms form molecules, and the molecules form macromolecules,

1:14:33 and the macromolecules form all kinds of things, you know, all kinds of structures, including under special conditions cells, and then the cells can be put together into multicellular organisms. So, you know, every time like you have more organization, you get new properties and and new qualities. And maybe at some point some of this amounts to conscious experience. Uh, but it's going to be something that's a special case. Yeah, much later and very special case, not not the norm.

1:15:07 And it's not going to be somehow distinctly relate So, this is sort of Okay, so a while back I asked about whether there was any kind of categorical or intrinsic thing that was wasn't going to be like computable a computational um that wasn't related to consciousness. Um and and part I mean part of the reason I had for asking that was what we talked about, but the other part was so I could talk about it here. Um because I I do it does sound like what you're saying is that there's a certain sense in which these

1:15:40 categorical properties are importantly required for any kind of computation. Um and that in that sense like you couldn't create a digital simulation not just of the brain or consciousness, but of anything in the physical world that we live in actually because of these categorical basis. So, any any digital simulation is going to leave that stuff out. And therefore, you're not going to be able to have this kind of concretely simulated environment. So, let alone whether we could be in there

1:16:14 and interact with it. So, maybe this is what you were hinting at earlier about the simulation hypothesis. This is the way you're viewing it impacts on that. Yes. So, right, right, of course. So, and this is this is actually our ordinary experience with simulations. So, simulation is a a dynamical representation of the evolution of a physical system without the actual categorical properties of the system. It's It's just a representation of what's happening and it's and it's not going to reproduce the the the the

1:16:47 the actual physical categorical properties of the system at all. Um Unless you So, here's where I always have the sticking point on this kind of way of thinking. Um and I saw some exchange with you maybe on Facebook about this point um which uh is interesting as a and it relates to the idea of the simulation of a computation. Um so, there's a kind of line of thought that's comes from you know traditional as quarters that if when you simulate the uh computation well, that is a computation. So, a a

1:17:20 simulator a simulated computer is a computer on this line. And your view is that no, not all simulated computers are actually computers or at least that's what it sounds like you're saying. Um well, what I was trying to say is that even in simulating a computation you you may or may not reproduce all the causal powers or the causal uh uh features of what you're simulating. So, this is kind of a much more specialized point. Um

1:17:55 and uh that's because you are now using say code to tell your computer your simulator how to manipulate representations of the states of the target computation that you're trying to simulate. Um and the reason this is relevant is that even if you were a computational functionalist about consciousness and you thought consciousness is just about computing the right things and the performing the right

1:18:28 computations um a digital simulation of that might not be conscious because a digital simulation of that would not necessarily reproduce the the relevant causal structure in the target computation. Uh cuz you know, you're just using code and representations of the states of your target and using the code to process those, but the states themselves are not connected in the right causal way the way they would be in the target. Mhm. So, that's not really my view, but

1:19:03 there is this new a brand new paper by uh Derek Shiller, I think. Yeah, that's who it is. The exact name. Where he makes a case that uh even if consciousness has to do with computation, we need to look at more uh uh specific constraints on what it would take to implement those computations for real as opposed to just simulating them, something like that. And I agree with him. I agree with him. I think that's a that's a breakthrough in terms of our um

1:19:37 understanding of computational theories of consciousness, you know, and I don't I don't subscribe to a computational theory of consciousness, but if I did, I would go along with um what he says there. Right. And the idea is that even if you simulated, just as you said, the same computation, that it might not the simulation might not produce the same causal structure. It might leave out certain elements of it. Yeah, and I think it's plausible that if consciousness has to do with computation, it has to do with like the computation actually being performed, not just being simulated in a in a digital computer where

1:20:12 it's it's manipulating representations of the computational states. Like the computational states have to be actively causing each other in the right way. Uh so so there's so much I have pushed up against reality here, but let me just ask you as as we're winding towards the end of this cuz I have kids that are getting off the bus at 3:00, so I can't be late for that. Whatever else happens. But, uh so do you take this to have any bearing on the it from bit kind of physical

1:20:47 theories Yes. Yes. Okay, so they can't. Okay. Yeah. It from bit Okay. is a goner. Um so, we have we have a whole chapter on this stuff, you know. Um I don't know if you had a chance to look at it, but it's chapter eight and we go through like every attempt that we have found to give any kind of substantiation to this it from bit idea and we don't really find a lot of merit in it. Um I mean, we do think that there are interesting analogies between

1:21:21 physics and computation and they can be illuminating and they can be used and that's totally fine, but if you try to take it seriously, that, you know, physics itself is at bottom computational, there's not much to that. Okay, so the the selling point of this book is it will rid you of thinking that we may be in a simulated world. Although, so it sounds like if you were like presented David Chalmers' argument that you have to, uh you know, wind your way through

1:21:56 and some level of credence in these, uh in the proposition that, um either I'm in a simulated world or there's some defeater for that. It sounds like the view that you've worked out is is trying to put its finger on what could be the defeater for that, um but also still take seriously the idea that computation is involved in cognition and all the things that we find are important. Um and kind of that's what I find very interesting about your work, whether you agree with it or not. And in large part, I find a it interesting and compelling. Not all the way, but I do think I'm a

1:22:30 fan of this work, but it's trying to thread this needle. This very interesting line between saying computation is important, it's real, the brain does it, it's an important part of our science of the brain, but at the same time it's not the whole story. We've got to be clear about how it's physically realized, what the brain itself is doing, that pays homage after It's so refreshing after all this long march of people saying screw the brain, it's all about this higher level. And now people are kind of coming to this. Right. No, it's definitely not the whole story about the physical world and probably not the the whole story about

1:23:04 the brain either, yeah. Very interesting. Okay, so I do have to wrap up here and let you go and live your life, but I do wonder just at the end if there's anything that maybe should have come up but hasn't come up, something that you thought maybe we we ought to have talked about. The answer can be no, but I always like to check at the end if there was something on your mind before you came. Well, I just hope that people you know, if anybody's listening up to now, do you get the idea that there were um there was like some substantive um

1:23:36 work to be done and there's still more to be done on understanding computation itself physically and it really helps us think more carefully about all kinds of issues about metaphysics, physics, uh and the mind. So um Yes. Very cool. Well, and on that note, let me thank you for joining me. I've I really enjoyed reading your book and having this discussion. I learned a tremendous amount, so I think I think it's well worth people paying attention to. Richard. Cool. All right, so just hang 1 second and I'll say bye off air, but other than

1:24:08 that, thank you very much. Okay, we're officially off of the air.