06:11 Scaling Intelligence: Evolved & Artificial, with Blaise Agüera y Arcas and Chris Kempe58:13 What is the Artificial? Artificial Embodiment, with Olaf Wi…
Transcript
0:05 · Okay, thank you everyone. Thanks.
0:06 · Welcome to um the the Anticther uh conversations um which we will uh very first of all just say thanks very much to to Olaf uh and to the AIFE conference for uh inviting us and for hosting us here at this. We’re very glad to be part of the uh uh part of the the convenings.
0:28 · We’ve met so many interesting people and we’re very glad to to be part of this whole thing. Let me begin by saying a few words about what Antthera is uh and what we hope to accomplish with with these conversations if if accomplish is even the the the right word. Um Antthera is a uh R&D institute uh that focuses on planetary computation and the evolution of intelligence and what the two of these have to do with one another. Um we have a a book series with MIT Press.
0:58 · Um the one of the first major titles of which was whitest intelligence by by Blazigu Earcus. Um we have a journal with MIT press. Um we run R&D studios um throughout the year and do a number of other things. We’re very interdisciplinary institute in many respects. You know the the kind of work that we do is philosophy with a small P.
1:23 · Um we we we are hope to sort of zoom out a little bit with the presumption that many of the disciplinary structures that our modern universities are built upon are largely a kind of inadequate taxonomy to actually deal with the epistemic ontological and technological problems that we really should be discussing. Um and so our our hope is to um try to find a different arrangement by which um making and thinking can can work together in some more more productive way.
1:48 · Um one of the questions that we think a lot about in the program um that brings us here to a life um is not only is is as you’ll see in this particular discussion is the relationship between life intelligence and technology and the ways in which those three things have co-evolved in very intense ways and and also the status of the artificial.
2:16 · And so unlike perhaps some other panels around philosophy of artificial life that you may host here at a a life or through the years which which are focusing quite appropriately on the philosophical status of life in artificial life. Is this really life? Is this a simulation of life? How would we know what are the sort of the conditions which are all good questions to ask.
2:36 · Today we’re going to focus again instead on this on the question of what is the artificial and what is it that artificial life as a research agenda can tell us not only as an applied space to think about how artificialization works but how is it that what a life as a research program over the past uh the past generation has allowed us to discover about the ways in which the process of artificialization um has worked.
3:02 · And so by that what I mean is that for our perspective we don’t see and I’m speaking as you know at Antrica school of thought not for Chris and blaze necessarily um that the dichotomization between artificialization and the evol and and the evolved is actually probably wrong.
3:25 · our the the the thought that I’d like to sort of begin with is that artificialization is actually something that is intrinsic to how evolution works. It’s intrinsic to how life works.
3:36 · Evolution selects for forms of life that are good at artificializing their environment such that it allows them to have access to more energy, information, and matter that allows them to sustain themselves and importantly it persists over time and to scale their population. In some cases, this is what we call niche construction or or forms of adaptation.
3:59 · But if this is the case and that artificialization is actually something not something that begins with modernity with artificial life, but in fact something that that has um that has been there all along. Um then the project of artificial life in fact becomes arguably more profound. Um and the work that we all do together becomes more profound.
4:18 · It becomes a kind of process by which life which art always artificializes its environment. Artificializing itself thereby is now artificializing the process of artificialization itself. A kind of second order dynamic that that begins with this. So that’s just a sort of beginning to sort of set the ground of a little bit of what we’re what we’re on about here as well. Um a few two more then sort of key points that she’ll go across both panels I should say.
4:44 · First we’ll have Chris Blazarkus and Chris Kempus and then we’ll have a second panel which will focus um on some of the artificialization questions in in a slightly different way is that um a ter a concept that we in antither what we call convergent artificialization and that is it’s easy to notice I think that over the past five 10 years that
5:08 · there has been a both epistemic but also practical convergence of research agendas within a life AI and robotics that would have been unusual to to observe even just in a few years pre previous to that. Traditionally the people working on one don’t necessarily have a lot of don’t necessarily collaborate with the others and they were asking what what appeared to be different questions.
5:29 · This was far less the case in Asia and particularly in Japan would argue where the where these areas of research were always much more uh intermixed uh and and co-constitutive. Um and in this sense if we’re observing something like a convergent artificialization at a at a global level, it’s in fact the um the Japanese uh epistemic approach of this that turns out to be the right one arguably.
5:55 · And so part of the the question is what can we learn from that context and how it is that we may want to think about this convergence itself um as something that becomes the the um uh uh an important part of a more deliberate part of the research agenda rather than um an epiphenomenon.
Scaling Intelligence: Evolved & Artificial, with Blaise Agüera y Arcas and Chris Kempe
6:12 · All right. So enough from me. Let me then um ask uh our two our two uh panelists to to to join in on this as well. Um, the way I usually like to do this instead of reading you their institutional affiliations and where they went to school and and all the rest of this is to actually intro have them introduce what they’re interested in and and we’ll and we’ll go from there. So, if I could ask you to start with Blae and then we’ll go to Chris. Um, first um, introduce yourself in any sort of any way you like.
6:39 · But what I’d really like to do is to introduce what are the questions that you that you are really interested in as as research questions as philosophical questions um as ones that um if you feel these if you could answer these questions it would be a big deal and then we’ll make sure that we can talk about those and then I’ll ask the same of you.
6:59 · Okay.
6:59 · Well, one question that um that is certainly on my mind a lot nowadays is what it means to thrive. Uh and in particular what it means for uh for an ecosystem, an ecology to thrive. Uh I I think everything is an ecology. Uh you know, from from a cell to a planet uh or at least a planet like the earth that that is full of living stuff.
7:19 · And um you know I I I think that you know with the the one of the goals of um paradigms of intelligence which is the research group that uh that I founded a couple of years ago and uh and that has grown since and you know includes includes both of you as visiting uh faculty which is which is a great a great honor. uh is that you know we all question some of the sort of basic tenets of of AI uh including uh
7:47 · that the current generation of computers are the right physical substrates on which to run it uh including the question of whether or not uh a life is a part of AI or vice versa whether those two are are are connected as you said uh and including whether optimization uh or utility is actually the right way to think about what uh intelligence is uh and In particular, uh if if one thinks about symbiosis and symbioenesis as as
8:15 · equally important force in both uh the evolution of life and the evolution of intelligence as uh competition uh or selection uh or you know or sort of zero sum games, then um you know the idea that things helping each other uh and working together is part of what scaffolds the next stage of of life’s evolution or intelligence’s evolution uh means that uh the whole concept cept of a score or winning uh at a game uh is really thrown in into question.
8:42 · So uh so those those are the questions that that I think so that I mean follow I mean it sort of teases this but just to bookmark it that that that you know the open-endedness of that game suggested certain things why the the research of open-ended evolution actually makes really important fair.
9:00 · Okay.
9:00 · So Chris T, what what why do you um what’s really most important? What are we doing here?
9:07 · Yeah.
9:07 · So I’ll I’ll use the same small P big P idea that you did. And so I’m I think there’s a distinction between small L life and big L life. Um and I think through the history of biology, we’ve learned an enormous amount about small L life, right? Which is just the life we have. Um we know unbelievable things about the vast diversity of amazing species we have on our own planet. Um things that are almost uh unbelievable.
9:31 · As I said, I mean, once you realize that the platypus is a mammal that has toxin sacks and lays eggs and all of that, it’s just amazing that we’ve uncovered all of that and we know quite a bit about all the life we have. I think we know less about capital L life, what life might be like anywhere in the universe. Um, which which things that have been created in computers are in the right direction or not towards a lifelike system. We just have endless debates about what life is, what the right definitions are, what axioms, and so forth. And so I’m really interested in um what can we call life anywhere?
10:04 · What can we count on for life anywhere?
10:06 · What is what is a good definition of capital L life? Yeah.
10:10 · With the question of of of of if it were possible to answer that or it possible to even know exactly how to ask that question and frame the question properly which you know I think your work has contributed enormously to in that kind of god. It would seem that that would have at least two very important implications.
10:28 · One in terms of the way in which we as life would understand the larger spectrum of which we are an in particularly amazing instance but also in the sense in which it would allow us to remake refashion or artificialize that dynamic. Is there a way in which in your mind in your thinking one is sort of a you prioritize one over the other as a precondition essentially that sort of epistemic self-reflection that would leads to a condition of self reinstrumentalization or or vice versa
10:59 · or how do you see the knowing versus making dynamic in the answering that question?
11:03 · Yeah, I mean I’m I’m really not committed to making over knowing first. Um and I think looking at the history of other disciplines, there’s lots of cases where knowing leads to making and lots of cases where making leads to knowing, right? And so I think um you know I disagree with the the statement um I think it’s Fineman who said you know I don’t understand anything I can’t build.
11:25 · Um I think you can build things and then learn from them. I think you can understand something and still have a big gap in being able to build it and so forth. And so I think um I don’t really have a primary thing there. I but I think the end goal for both is to try and figure out what are the principles, right? So what really are the the foundational components to something like that?
11:45 · Um so for example in physics there’s lots of principles we have that even tell us of devices one could make to measure something um for cases where we don’t yet know how to make that device. Um, LIGO was a great example, right, where it uh the principle was there, the idea was there, and then there were huge engineering challenges around how you actually build this thing to the right specifications and so forth. Um, theory guided all of that that thinking around what the specifications had to be and so forth.
12:12 · So, I’m um I think our ultimate pursuit is the principles. Um, I don’t know whether knowing or making will come first.
12:20 · Okay.
12:20 · So, you’re the book you just the book we just published um what is intelligence? Um, speaking to the thing you said before about you’re not it’s we’re not sure whether a life is a subset of AI or AI is a subet of a life or whether they’re actually just two words for the same thing. Um, you published a book called what is intelligence that includes a little book called what is life which maybe at least implies an answer to this question in a certain sort of sense as well. Do you see the the what is the L in the what is life title the small L or the large L in the way in which Chris is framing this?
12:53 · Um and do you think the book proposes an answer or do you think it proposes a better way of asking the question in relationship to other questions or how would you how would you enter that into this into what Chris has proposed?
13:05 · I propose an answer. Um so u you know we drum roll. Yeah, we we just um you know with with um uh Mike Leaven and Reed Bender, we actually just wrote a book you know called or sorry wrote a wrote an article called what lives which is a taxonomy of of dozens of you know people
13:23 · in in various interdisiplinary fields uh some of whom have been you know pretty big figures in a life uh writing their definitions and none of the definitions agree with then use like LLM of life of life that’s right and intelligence is even more contested of course nowadays um you know you both have alluded this this this issue of whether technology is ahead of of science of understanding or theory versus theory being ahead of technology and those things have you know switched places various times um until uh very recently uh you know I
13:54 · would have said uh that with respect to intelligence theory was well ahead of of um practice now practice is well ahead of theory we have intelligent systems but we’re still arguing about what intelligence uh is and on the life side of things of course Our understanding of the little L is ahead of our theory about the big L, but we certainly understand, you know, how how little L works a lot better than uh you know, that we’re able to make it. Um we don’t have open-ended uh you know, life systems yet. That’s a big part of the goal of of a life researchers.
14:25 · But uh yeah, so here’s my definition of life. Uh life is embodied autopoetic computation uh arising and complexifying through symbioenesis.
14:37 · Um so that one has gotten refined a little bit uh you know over the last over the last year but the the key is that it’s a functional definition uh meaning it’s not based on on the details of what particular chemistry were made out of you know in Chris’s super elegant talk this morning he you know described uh what some of the constraint space looks like that has led to you know the solutions you know for how life works on earth the implication being that those
15:04 · same constraints could be met functionally in different ways using different parts on another planet in a different ecology in a different environment and um uh and and so that leads to a functional I think that that has to lead to a functional definition.
15:17 · This this also uh I think is closely related to uh to ectoenial’s uh really nice uh talk yesterday in which he was talking about you know even causation uh not making any sense outside the context of computation. So this is a functional and computational definition. Um and and vice versa arguably. I mean I think the fact that Yeah. you can’t talk about Yeah. And and vice versa.
15:37 · And and the fact that we have metabolic pathways that multiply realize various functions uh you know or that wings you know are multiply realized by by bats and by insects and so on you a hallmark of the same thing or you’ve got aerobic and anorobic respiration you know in your cells. So you know those are all um multiple realizations of a functional need for generating ATP or for flying or whatever it is. Uh and and then the the other part of this that it you know it’s autopoetic meaning it’s self-constructing. That’s kind of the key. Uh things that are that are living uh I think have to keep living.
16:09 · That’s their that’s their their goal. Uh and uh and that means healing um growing andor reproducing. Uh all living things do that. And uh and finally um complexifying through symbio genesis is kind of the insight that that that really fuels that you know that that what is life part of the of the book which is that uh rather than just uh Darwinian selection in the standard way
16:35 · um these things compose themselves through um through symbioenesis and that’s that’s why they get more complex over time.
16:43 · So I want to ask you guys a little bit then to turn to the question of intelligence because both of you have a very you know deal with this in I compl in very complimentary ways but I switch things as well but the question of artificialization in terms of the definition that you give I mean in theory you could go through each one of the words in this definition and sort of put the word artificial in front of it and say it’s you know it’s artificial autoois using artificial symbioenesis or you could just put it in front of any one of these words and it would still be meaningful in some sort of sense.
17:10 · And so maybe this is this is this is a whole another discussion perhaps when we sort of go here. But I wonder if you that if you you see where I’m going with this and if you have a particular place you would put that word.
17:21 · So I I don’t I don’t think I I I may think about artificialization a little bit differently from you do uh Benjamin. I I so far so far um my my take on on um artificial is that usually when we use the term we mean something that we have engineered uh as opposed to something that arose. Um that that is how we use the term. Yeah, but we’re wrong.
17:43 · Yeah. So that’s kind of standard.
17:44 · Yeah.
17:45 · Um and uh mean fake, right?
17:48 · And that way we’re wrong.
17:49 · And that’s I think that is wrong. Uh if we if we’re going to be functionalists, then there is no fake. If you uh if you have a a you know, a code emul, you know, a code emulator, uh you know, that that you’re running calculator inside of then it’s still a calculator, you know, right? So it’s not like a fake calculator just because you know you’re you’re simulating the code or you have an extra compilation step or something right.
18:11 · So um so you know if we believe intelligence for instance to be about you know what it does uh as opposed to what what it you know is in some essentialist sense there is no such thing as artificial intelligence. Same with life actually. Um but we we can talk about life running in a engineered computing environment as opposed to on a physical substrate that that that sort of evolved the oldfashioned way.
18:32 · And there is something different about engineering than about that oldfashioned stuff which is that uh it wasn’t simulated in a in a mind you know like life life doesn’t have a a prospective process of simulating various possible design choices and picking one over the other ones because it worked in simulation. Uh you know you just you what you what you get is is what is what survived functions. Okay. Right.
18:55 · So that’s artificial for so this question of function um is another one of bookmark to come back to it because it also I think connects to something that really struck me from your talk this morning, Chris, which was about the the computational efficiency of ribosomes, right? that in a certain like through the p you know through the sort of blind pursuit of sort of finding the function of this itself that we that we arrived at what is close to landower limit ideal ideal computers for for
19:24 · making the the building blocks of this this as well which kind of suggests I don’t want to go too far down this rabbit hole but just to sort of preview a little bit which kind of suggests in a way that there was a degree if not inevitability a path dependency towards that particular kind of solution which constrains a function space in in a interesting way But the thing I want to talk about first is intelligence because one of the ways in which both of your you guys work is so vibrant and and
19:48 · vivid is the way in which intelligence and the function of intelligence the quality of intelligence is understood in social mutualistic dynamic systemic terms rather than as a quality of a only only as a quality of a particular agent or organism rather is something that emerges from these from these kinds of systems itself.
20:07 · So I wonder if Chris if you could then maybe perhaps in terms of scaling dynamics in particular which is a lot of where your your work kind of focuses on if you could think about as way you see it at a really kind of a functional or fundamental level that relationship between intelligence scaling and life like where is the causality within that computation.
20:29 · Yeah.
20:29 · Yeah. So I think uh from that perspective you could always say what intelligence is about is solving problems right and so there’s sort of two components to that one is uh what problems matter to an entity and then the other is uh what does it take to solve those problems right and so if you take something like bacteria they certainly solve many simple problems right so chemotaxis is a certain sort of problem solving where you have a gradient of nutrients you’re trying to follow it around there’s a simple algorithm that bacteria use for doing that that involves some amount of sensing, right?
21:01 · Um Dan Dennett and many other people have written about how that is a form of intelligence, right? It’s a certain sort of problem solving. It’s it’s much simpler than the sort of abstract thinking about thinking that all of us are doing. Um but there’s sort of two points there. one is that’s the problem that matters to bacteria and then the implementations you get for that are what’s achievable in this tiny little system that are also matched to the
21:27 · problem it’s trying to solve so that’s sort of one issue of scale um and then as systems get bigger uh there’s both the capacity to solve more types of problems right I mean this is something simply understood for computers where the more transistors the more types of things you can solve the bigger problems um as memory grows you can solve different sorts of problems so there’s a very simple way in which just scaling up computation allows you to solve different problems. Um, and so that’s true for other organisms. As you get bigger, you have more types of components, more ways those components can interact.
21:56 · Um, and then also the problems you’re solving may be getting harder as well, right? Um, there’s all sorts of game theoretic issues that come into play as you get bigger. Do I have predators? Do I have prey? What are my prey and predators trying to do? If I have information about them that I can harness in some way that gives me advantages. Um for us certainly the ability to forecast seems like quite a useful thing. Um but that requires a lot of memory. It requires whole theories.
22:20 · It requires a bunch of abstraction. Um all of that is is a scale issue in a lot of ways, right? And then if you look at societies, economists have written for a long time about uh specialization and the importance of that. Uh that’s really a way to get a diversity of functions that interact in different ways to solve new problems. Right. So there’s certainly and that’s a function of population size too.
22:41 · Exactly.
22:41 · So that’s, you know, diversity scales with city size in terms of specialties, functions, professions, all of that. Um, and so those are those are really important scale questions, right?
22:52 · Um, and and there’s also just a scale thing about how much you’ve written down in the past, how much you’ve discovered, right? So a lot of what we’ve done is is have an expanding population that’s learning more and more about the world and recording it and transmitting it to each other in in really simple ways. Um my close collaborate David Krakow has talked a lot about external apparati apparati um books and and other sorts of
23:15 · things and and that’s really important to what we do and how we solve problems right so some of that is impossible for a bacterium um just in terms of what they have how much they could store some of that is also not essential to them they live in a simple world they have particular problems to solve and they found the right ways to do it I I’m really interested in in the evolution of those external apparati itself, right?
23:36 · That the whole one of the great things that you know, one of the things that SFI has been is, you know, among its many fruitful tracks is is it research in technological evolution itself from Brian Arthur to sort of work on this as well. Maybe that’s that’s another panel to sort of to do, but I just wanted to sort of flag that. One thing to sort of connect it with your workplace is is Chris mentioned um predator prey dynamics, which is really all about theory of mind among among other kinds of things as well, right?
24:04 · And so you have I think an equally rich and quite correspondent theory of intelligence and scaling and one that really is has theory of mind as its kind of at it at its core and its crux. I wonder if you could sort of lay that out for us.
24:19 · Yeah.
24:19 · Um nodding along very much to everything everything you know Chris just said. I I agree with all of that. Um you know intelligence is is uh has at its core the idea of modeling uh modeling what matters uh to you. So that means that there is a U. Uh the question of what the U is is also a model. Of course uh you know everything is models.
24:39 · I mean uh as soon as we go beyond I don’t know electrons or whatever electron is a model too. But you know something like a chair uh you know it’s it’s not it’s not a a fact in the world the way you know would have said or whatever you know chair is a is a is a model. We recognize a chair when we see it. It’s a functional model. It’s it’s there because that’s what we can sit on.
24:59 · Um you know is a random boulder a chair?
25:02 · I don’t know. Um, you know, and different cultures actually, you know, define, you know, what is what what works as a chair or a bed differently.
25:07 · You know, certainly what counts as a bed in Japan is not necessarily a bed that I would sleep on. Um, and um, uh, you know, it’s it’s uh, so these things are are are always model dependent. Uh, now the the most important aspect of of the Vvelt or of the model of a living thing, uh, even something as simple as a bacterium is itself. uh in order for something to um persist through time, it has to know uh what it takes to stay alive. Uh so, you know, if you’re a bacterium, you’ve got certain building blocks and sources of free energy that you need to pursue.
25:38 · Uh and that’s why you’ve got chemotaxis. So, you know, the bacterium follows those gradients and and you know, solves that problem of modeling concentration and and and so on because you know that’s that’s essential for it and it has a self model by doing that. The the other thing though is others uh you know and even for for something as simple as a bacterium other bacteria are an essential part of of the of the environment. Uh they have quorum sensing and changes in behavior that happen in bofilms and in other kinds of context depending on starvation that you know they might uh you know develop or or or retract a flegellum and so on.
26:09 · So um you know that there’s always this there’s always something like theory of mind even in something as simple as a bacterium which is a model of your of others like or unlike you in the environment that have their own uh desire you know deciderata and and in principle to your point about forecast some sense of if if if then statement like if I go over there good thing if I go over here bad thing yeah I mean again even something as simple as chemotaxis as an if then right
26:35 · uh you know if I’m going in the direction of increasing gradient uh you know then it’s probably a good you know then if I if I reverse my flegellum to tumble then you know I might screw that up so I should probably you know right so so um it’s it’s a behavior is always
26:50 · about an if then based on a model uh you know even something as simple as a thermostat is an if then based on a model uh so uh the moment you start to have uh anything like cooperation between agents uh that forces or competition or competition right so predation uh is you know is is you know also requires that the predator model the prey that the prey model the predator and and they can engage in an arms race once that’s going on which is how you get intelligence explosions.
27:15 · So uh you know the the reason it becomes favorable to get smarter uh is when uh you’re when when you modeling others becomes favorable uh whether to to eat them or to cooperate with them or to not get eaten by them. And um and in particular the cooperative aspect of that is very is very powerful because that often happens within species.
27:35 · And uh and if you uh if you get smarter to model somebody else in your in your species and and the genes for doing that are are you know are shared genetics then they become smarter to and harder to model and then you get into a friendly arms race. These are social intelligence
27:52 · explosions which have been observed of course in the hominin line but also in citations and in bats and in some species of birds and in a bunch of other uh places and yeah and I want to I want to note there that um you know the interesting thing in a lot of this space is that some of these things are so highly multi-dimensional right and evolution is always navigating this high highdimensional space and so it’s interesting to ask when we get pushed down these channels of greater intelligence and when some other solution does a good enough job. So, that’s probably really important though.
28:22 · Yeah.
28:22 · So, so for example, if if you have a new predator and you quickly evolve the right sort of defense, you know, say a a shell with giant spines on it, um that may be good enough, right? And there may not be a push for greater intelligence, right? And so, um I think there are real questions about why we often see so many lineages getting pushed in the direction of intelligence rather than some simple solution.
28:43 · Um, and I think part of it is that uh once you get to these higher levels of intelligence, things that are more general, more abstracted, they’re useful for so many different types of things that they solve your predator problem, you know, if you figure out how to hide, but then they also may solve how you find more food and all sorts of other things.
29:02 · The autois problem.
29:03 · Exactly.
29:03 · Yeah. So, it’s so it’s solving many problems at once. And I think that’s why we see it um fix so often.
29:08 · Well, and and when you of course you’re always limited by energy and as you as you showed you know so elegantly you know with with a bunch of your work since you know since 2012 at least um and uh one of the cool things about intelligence is that no matter what the initial impetus is you know whether it’s uh you know Mchavelian hypothesis or uh you know or or hiding from predators or whatever it is you know you if if you get intelligent enough you can unlock new new chests new sources of energy which then allow you to grow further.
29:35 · they they allow you to break some some new uh barrier. Uh and uh you know, so I think the the example of of octopus is is an interesting one for a couple of reasons. I mean, one one um sort of hypothesis about why octopuses are intelligent that I I think Peter Godfrey Smith put forward is that they lost their shell. Uh and you know, so they become extremely vulnerable and they now have to model everything else in the sea because they’ve got to both hunt and and not get eaten.
29:59 · And that becomes a hard problem when you have no defenses. uh you know or your your defenses involve complicated camouflage you know as well as all sorts of other uh funky strategies.
30:08 · So vulnerability is vulnerability might matter pressure for for those passes.
30:13 · Yeah.
30:13 · But but you know the other one of the reasons that I began thinking a lot about octopus intelligence uh is because they seem to be a the a weird exception to the social intelligence hypothesis since um they’re you know infamously solitary. You know they really they really don’t tend to like other octopuses. Um, and uh, you know, of course, there could be other, you know, they’re a very ancient lineage. They could have been very different 600 million years ago.
30:38 · But, but my my pet theory is that, uh, there there’s also something else weird about octopuses, of course, which is that they’re so highly modular. Uh, they’ve got, you know, eight arms. Every arm has, you know, hundreds of suckers on it. A lot of the intelligence is in the suckers and in the arms. They, you know, they only have, you know, two-fifths of their neurons even even, you know, in their heads. And as far as I can tell, the neurons in their heads, what we would call their brain, is actually just an optic lobe that is shared by the by the eight arms. Uh so another way of thinking about it is that it’s a community of eight arms.
31:09 · Uh you know, they’re so uh floppy, right, and and so complex and you know, they have a lot of intelligence if you cut them off. They have to be modeling each other constantly because there’s no central model.
31:19 · Multi- agent interaction.
31:20 · It’s multi. Yeah, exactly. So it’s like it’s like a rowing crew of eight uh rowers that have to constantly model each other in order to cooperate. Did you want just wonder? I I I did want to ask you though a little bit on the path dependency contingency thing which you I think you brought with the what when does evolution select for intelligence as adaptive versus when is this just too energy intensive to bother? It’s better just growing a spikes on this as well.
31:41 · Like there’s a certain sort of there’s a contingency dynamic there as well. But I mean the bigger I mean there’s there’s a there’s a cosmology scale question of of you know how deterministic are these sort of but you I think you get the the gist of the question. I’m giving you please as much rope as you need to hang yourself on.
31:58 · Yeah.
31:58 · Thanks. So I mean as as I said and as Blaze mentioned um I think about so many of these things in terms of energy budgets, right? And so shells take a certain amount of energy to build. Uh brains take quite a lot of energy to to maintain and so forth. And so I think it’s all about return on investment, right? Which is a very sort of pragmatic budgetary perspective, but you get quite a lot of traction out of that. And so I think one of the points about intelligence is that you get many things out of the same investment, right? And so I think that’s that’s the key. Yeah.
32:27 · Exactly.
32:27 · And then you also get things out of it that you haven’t already been selected on. Right. So if you have a truly general intelligence um or just what a general enough intelligence, you can deal in the moment with problems you haven’t seen before.
32:40 · And that’s really impressive.
32:41 · Preemptively adaptive certainly. And so I think if you had really good spikes um but then you also had intelligence that allowed you to evade the same predator and do a bunch of other stuff uh for the same cost you’ll take the general intelligence over the spikes right and so I think it’s there’s a lot to that it’s a very simple budget perspective but I think in many cases we can evaluate traits uh from that perspective okay based based on evolution that was essentially done 10,000 years ago uh you
33:06 · know when we were uh you know picking uh you know lice out of each other’s hair um I guess we still do sometimes but you know that that’s the same brain that that you know discovered how to how to solve nuclear fish so you know unlocking energy sources like that based on you know the the primate brain is quite an impressive achievement of course it’s a collective one too so I want to then link this question of scale uh and a sort of integration sort
33:32 · of cognitive symbioenesis to some of the work that you know both of you involved perhaps raised more plays most explicitly around artificial intelligence and the ways in which that a lot of the work through in pi that question that’s focused on AI um is built around dynamics of of social intelligence and multi-agent interaction and the conditions of theory of mind within multi- agent interaction of this as well and it it’s it’s not an uh you
33:59 · know given the sort of how much noise there is in the moment around AI and where things stand and what sort of term means it’s a um it’s a novel and not necessarily obvious way of even defining what artificial intelligence is and where we should be looking for it and how we should be thinking about building it. And I wonder if you could sort of take it from the top a little bit and talk about some of the the recent work that you guys have been doing in this area because um you know I think in ways are pretty clear.
34:24 · It connects really quite nicely to the the scaling issues and intelligence scaling issues that you know your work has been really sort of built upon and they’re really I mean they’re building blocks of each other it would seem.
34:35 · Yeah.
34:35 · Yeah. They’re they’re very connected which is why we you know why we why we vibe so well but uh yeah I mean there are various scaling laws of intelligence that have been well documented uh there there are the ones about uh cortical size you know or incilization uh there are ones about the uh the sizes of societies uh and you
34:52 · know we get into sort of Dunar number and you know the sizes of monkey troops and stuff like this which connects with cortical area uh so you know number of number of neurons in your head and um and there are also uh scaling laws that that some of the Santa Fe folks have figured out about cities uh you know that’s right right and so you know whole societies have have you know intelligence that you know again like a single person doesn’t have the intelligence to solve a nuclear fishision but you know get a few million of them together and they can they can get a lot done uh even though individually people are not that smart
35:23 · um so the the um the idea that that intelligence scales through cooperation of those of those uh agents so you know a bunch of agents work together to make a bigger agent is kind of central to this.
35:36 · It’s the artificialization of the cooperation as much as it’s articulization of the capacity of any one agent.
35:41 · Yeah.
35:41 · And uh and in that sense, you know, a lot of institutions that we’ve developed in the holene, you know, are key to this and have nothing to do with with the evolution of our of our brains.
35:51 · I mean, we we’ve evolved a little bit in the last 10,000 years, but not, you know, nothing compared to what, you know, what we how we’ve evolved socially. And um and that’s the key to all of our all of our you know big big eye intelligence ability to make computers and to artificialize and all the ways that you know you’ve talked about. So um so the the the question I mean the the question that some some folks on the team have been um have been working on uh hard over the last year.
36:14 · Um we have a paper that’s been like two weeks away from publication for the last couple of months about this but uh it’ll it’ll come out real soon now. uh is is about multi-agent reinforcement learning and uh you know how you can solve this classic problem of um you know of a bunch of selfish agents that you know possibly have something that they want to maximize themselves or want to solve for themselves um solving a collective problem you know how do they how do they cooperate and that’s you know that’s been a very hard problem in in RL um
36:44 · because normally you think about an agent as being outside the system that it’s playing, you know, it’s maybe playing a game and uh it’s not a part of that game. Uh that’s okay if if if you’re playing uh you know, a full information type game like uh chess or something where you know the next best move has to be determined by just looking at all the pieces on the board.
37:06 · But if if the game involves working with others, then you have to actually model them, too. And if you’re modeling them, you have to model yourself, as it turns out, and model them model you and model them modeling you and model them, modeling you, modeling them, modeling you. up to infinite order.
37:22 · Uh and and and once you introduce that idea that that you’re you’re doing inference over, you know, in the space of yourself as well, something really cool happens, which is that human users are part of that loop too.
37:32 · Yeah.
37:32 · Human users are part of that loop too. Exactly. And and um and and this is what allows you to cooperate in short.
37:38 · Um but it also means that you know I mean I think that’s where consciousness comes from uh in in a very functional sense in the sense that you know if if you modeling yourself is actually necessary for this kind of social cooperation business you know and having meta thoughts about it if you like you know model yourself modeling yourself etc then that that functionally has a lot of the things that we talk about when we mean conscious right and they connect with Gratziato’s work on so the encourage attention theory very closely related yeah um but but the the other the other thing about this is that u is that then if you are thinking about um about about scale
38:11 · uh and cooperation is necessary to generate scale then you start to see that parallelism uh is uh you know parallelism and cooperation and symbioenesis all go together as ideas. So this is why my definition of life and my definition of intelligence start to really blur.
38:27 · Okay.
38:27 · So um yeah please I just want to say one thing there about scale which is I think if you think about um the problems you want to solve as some highdimensional space um then as you start to push out the boundaries that
38:42 · surface area is growing very fast right and so there’s just a scale question to populate that surface area but you can solve more and more problems as as you have more people especially as they’re specialized to find each of those frontiers and then you have some communication between the frontiers right so for example for mathematicians, you know, which is fully structured knowledge, right? You could argue that mathematics is fully hierarchical in the terms of the way we teach it and the way people learn it and so forth. Um, so much of the game for mathematicians today is just learning enough to get to the frontier, right? Just knowing all the theorems that get you to the unknown. Yeah.
39:13 · And so that’s a real indication of where specializations and scale helps you is that you need to have some number of people dedicated to learning that trajectory, getting to this boundary, working on it, and then finding ways to communicate that across the surface. And so I think there’s just a geometricness to what scale gives you for unknown problems or unknown or things we don’t know yet.
39:36 · Where division of labor is the key, where division of labor and specialization is the key. And then with a huge amount of cross communication and I think that’s the piece we actually don’t have solved as well is how do you communicate across all the nodes that have found these frontiers but there seems to be I mean that in addition to the communication across these kinds of frontiers there’s also what you what you were calling earlier compositionality right there’s ways in which like the knowledge that comes before or like in your work with the the brain like scaffolding and that you
40:03 · know things building on other things the things that persist through time that become components of other things like you know the knowledge that someone builds upon doesn’t go away it sort of persists in this as well you know your your your work with assembly theory also you know with this scaffolding kind of dynamic I wonder sort of each of you just this this will by the way sort of be the last thread I’m going to ask him to go around and then we’ll have a little bit of time for some questions Q&A so just a bit of foreign to sort of think about what you’d like to ask and then we’ll we’ll spend some time there as well but please just let’s let’s talk
40:33 · scaffolding for a moment yes I think there’s there’s two points there I mean if we look at the history of life and the transition itions life’s gone through, it always takes existing components, puts them together into new arrangements. Often you take two components that are the same or roughly the same, put them together, and then you specialize them in different ways.
40:48 · Right? So that’s sort of um people talk about that in different ways. I’ll I’ll let Bla1 talk about his version of talking about that. Um but assembly theory certainly is one way to to say how that works, right? Which is once you discover a set of parts, um it’s easiest to put those parts together to make new things if the parts are general enough, right? um and therefore they and they persist thereby.
41:10 · And they persist thereby, right? Because it’s also part of the same synthesis pathway and so forth. And then once you get new parts, you can combine those in new ways and so forth. And so that’s why looking at the shortest path to build something has meaning because you’re asking um once I have a collection of things, it’s much easier to use the solutions I’ve already found in recombinatorial ways than to go to some totally different solution space and try and rediscover equivalent things. And so that sort of pushes you down a trajectory where you’re building parts and then reusing those parts and combining them.
41:35 · Um in evolution we see the same thing right where uh I mean a lot of that is sort of applied to chemistry but but in whole macro evolution whole body macro evolution we often see um say two of exactly the same cells come together form a multisellular organism then you start to specialize one of the cell types and many people in in this community have written endlessly about that process happening over and over again in both um real and artificial systems intelligence scaffolding.
42:01 · Yeah.
42:01 · So 100% all of that and um when we talk about general intelligence by the way you know the the ability to specialize any which way you know for one of those parts is what that is about right that’s the G in in in general intelligence uh and and what that means is that you can um you can form uh you can you can do compositionality with your uh fellow with your your con specifics and um and generate uh larger
42:29 · uh you know intelligences in an almost unlimited way when when when that specialization can take any route. It doesn’t require a genetic specialization. Um the the other uh so the other maybe brief comment about that is that um there you know the the idea of Moor’s law that Silicon Valley people you know prattled on endlessly about for many years and that I was very dismissive about for a long time that sort of scale solves everything. They were right about this and and they were also right about the speeding up that happens um again due to compositionality. Uh W.
43:00 · Brian Arthur who you mentioned earlier uh you know in his ideas about how technology evolves said the same thing right that when you have start to have a toolkit of technologies that can be recombined to make new technologies that process inherently speeds up because every time you put something together to make something new that’s useful your toolkit also gets bigger which means the number of ways you can put things together grows. So, and that same principle for life for life too for intelligence which something like they’re not so different from each other.
43:29 · Exactly.
43:29 · So, so they they all have that same speeding up that same exponential speed up dynamic and and that’s why and and that’s not a dynamic that you would predict from Darwinian evolution for instance in which everything just drifts uh you know through through gene changes by random mutation. This requires it’s not just increasing complexity or you know it actually there’s structure.
43:49 · Okay, it’s always recomposition. Um, we could certainly go we could go on for a while, but I’d love to sort of open it up to then to the audience. Um, do we have a microphone here? Um, and it it’s a bit first come first serve uh on that as well. So, anyone who’d like to we’ll we’ll take 10 minutes, 12 minutes or so depending on how on how animated you are. Uh, and we’ll we’ll please any questions you might have for Blazer.
44:12 · Chris, hi. Um, amazing uh discussion. And I have so many threads, but maybe I’ll sort of go through this one. Um, so I’m interested in your thoughts on um how the cooperation versus competition trade-off uh scales with the level of recursion of me modeling you modeling me. And sort of, you know, if you look in society for example, you can imagine there are certain strategies politically at least that model only the first layer.
44:40 · It’s like oh you know I will meet your immediate needs if you if I can convince you that you know your immediate needs can be met by removing sort of some another group of people’s needs right and so how do we sort of encourage people to sort of maybe you know I if it turns out that cooperation emerges if you go up the levels of recursion ultimately you think to yourself of course it makes sense for me to care about wider ecosystems because if I model people modeling me and so on you know I’ll be less selfish.
45:09 · So maybe first on that and then secondly sort of we’re in Japan and sort of there’s these sort of um you know Shinto and Zenh kind of influences and I’m wondering whether any of those could be sort of shortcuts to this recursion uh that you know if you don’t have the time and space to be a person with sort of um a lot of recursive capacity maybe some of these influences can help. So yeah, uh you also mentioned selfish agents cooperating.
45:37 · Maybe that’s sort of, you know, investigating those uh principles would be useful. Um and yeah, ultimately how do we encourage humanity to engage in a friendly arms race sort of, you know, that that engages in cooperations and raises intelligence for everybody.
45:56 · Yeah, that goes to your thriving thing.
45:59 · So um so something I’d say about that and I I think a lot of you know if someone were writing the 10 axioms of a life I think this would come up which is how do you avoid cheaters right um and I don’t mean cheating uh often that has sort of a very negative connotation but just when you have a cooperative dynamic how do you prevent one of the two from
46:19 · defecting and maybe taking the goods from the other but not returning anything or um how do you know how do you keep uh if you’re a symbiant with something how do you keep it from eventually eating you and how as the host, how do you keep the endo symbion from exploding and and eating you from the inside out as a parasite? Right? So, all of these things have potential for cheating dynamics. Um, I’d say from a history of life perspective, every time we get a new layer of life, there’s always a question about what is the negotiated steady state, right?
46:48 · Is that um some sort of uh neutral fitness between predator and prey in some way that leads to a steady state dynamic? Is that some sort of new form of negotiated cooperation and prevention of cheating?
47:02 · Um, and so I think to your question, which I think you’re, you know, at the highest level you’re really asking about as human society becomes more complex and we have more people and more different types of dynamics and so forth. Um, I don’t know what the answer is, but I think the question really is what are the new types of uh, prevention against cheating that we need? And I just mean cheating in terms of a dynamic that leads to something destructive and so forth.
47:24 · I think in the last 200 years we’ve seen many cases where there was a new dynamic and then there was a new negotiated steady state that prevents a certain sort of um compet the wrong sort of competition right say warfare and that kind of thing so I think that’s always a question for new dynamics new forms of life um higher levels bigger scales is just um how do you navigate prevention of cheating competition predation all those sorts of things um I
47:52 · don’t know yet if we have a good theory for telling us what the next one we need is. It would be nice to have such a thing because I think we do need new forms of of prevention of negative dynamics, right? Big big agreement to all of those. Uh but I I I will maybe mention two cool hacks that have definitely you know happened in the in the human era of life on earth. Uh one of them is solidarity and the other one is insolvent.
48:15 · So the solidarity one is basically the creation of anonymous societies. uh and this is what uh uh or anonymous associations these are what allow us to scale beyond the Dunbar limit of 150 or so people. So in in the other primates uh we you know they have
48:32 · individual recognition individual relationships but um you know the the size of the troop is limited uh by the by theory of mind uh you know and by direct you know one-to-one relationships uh kind of like Ronald Reagan um you know famously the the thing that you know he was a very generous man and a sweetie uh you know but but failed to be able to uh to generalize you know the the you know the the kindness right in onetoone uh relationship ships to groups uh that that were more abstract and um when whenever we look at the other way around.
49:05 · Yeah.
49:05 · Yeah. Exactly. But you know when when we when we uh you know form you know national identities, religious associations whatever it is right those those things uh bring with them a package of uh you know a cultural package and and assumptions about norms and about um solidarity uh that that allow human societies to scale far beyond the number of people we can model individually. So that’s a cool hack.
49:25 · And the other interesting hack is one that um that I I I really kind of got turned on to by uh by the writing of Robin Wall Kimmeer and relates to uh the projection of a of a soul onto uh things on onto other species, onto ecosystems, onto things that are inanimate and and that creates a sort of uh care ethics with systems that are that are uh that are not other uh human beings.
49:50 · And uh you know, it’s very practical. uh it it allows you to to um to care for ecologies in in ways that you know we certainly didn’t you know evolve our priority to do those instincts were are being mobilized in in a new and interesting way.
50:08 · Philip, I think you’re up.
50:10 · Yeah, quite a simple one. Bla1, I was trying to contrast your definition of life with the old NASA self-replicating chemical system and I was just wondering if you are adding any uh signatures via your definition for the kind of practice of astrobiology. I know you know we have our bio signatures technicious Chris was talking about stochometry earlier and I just wondered if there is any new lens kind of comes through it. I didn’t manage to write it down in time your definition but I’ll get it from you later.
50:38 · Well, there’s there’s some slightly earlier version of it in the what in the what is life book which is which is online as well. But um but yeah, I I you know it’s it’s not a chemical definition uh in that it’s functional. So it doesn’t matter what the substrate is. Uh it does add compositionality to the to the picture.
50:55 · Um that’s related to assembly theory in in key ways. Um but um at least if one reads the the tweets of Lee Cronin uh you know uh assembly is not computation and and and my definition is absolutely computational. So that is it adds that as well. Um you know and and computation goes along with modeling uh you know for the for some of the reasons again that you know that that um you know Ector talked about that that’s you know the moment you start to think about um about
51:23 · software about about a program you are in the realm of causal um modeling essentially.
51:30 · Soft match in the house.
51:32 · Yes.
51:32 · But I just a meta point about definitions of life. Um which is I think you know when we look at the history of different disciplines often the first definitions or first comp uh uh concepts have all of the right ingredients right and then it’s about sort of figuring out what the metrics for each ingredient are the waiting of different ingredients and then over time you get this projection of oh that’s the thing and it’s this much of that and this much of that and and we have a nice mathematical theory for it and so forth.
51:59 · So I think it’s it’s interesting to think about um you know I think the summary of the components or the pieces or the elements are all in the table and then there’ll be really interesting questions about what theories metrics uh mathematics emerge to sort of uh project all of that into some crystal that we all agree on eventually 100%. But by the way the computational view of of of life also explains why metabolism exists and why it costs energy. You know, you talked about the land hour limit.
52:26 · Uh, you know, anytime you are doing an operation that reduces the phase space, you’re both creating negative entropy in in yourself and you have to use free energy in order to do that and spentism.
52:38 · Okay, we’re a little wrong. So, but I want to hear from from David and and Wendy. So, please um if we maybe if we do if we have each of you ask your question and then you answer them in concert um and then we’ll be able to keep on schedule. But I want to make sure we everyone gets a chance to ask what they want. Please. Yeah.
52:52 · Thank you. Great panel. I I’m curious with your work on uh code replicators like brain uh you get replicators you get uh symbioenesis what uh does it keep going does it top out and if it tops out what do you think is missing is it scale diversity uh or uh are you running a giant soup and it just keeps uh complexifying
53:21 · hi um I have a question about like modeling or like I guess I’m trying to understand what that means because when it comes to I guess like um uh just if we just talk about it in the realm of like biological life I feel my understanding is that if you if you’re somehow as like a creature if you’re able to model something like you somehow know where you’re where you’re where you stop kind of or like where something is not yourself anymore.
53:46 · So it almost seems like there’s this like assumption or something that happens before modeling in itself. um that needs to happen first and so I was wondering if you could speak more about whatever that is.
54:01 · Uh so I’ll real because I think Blaze has most more to say about both of these although maybe we can talk at the same time and then people can like decode the thing later. Um uh so perfect. Yeah. Yeah. Different channels.
54:15 · Um so I I think the idea of modeling for me is just about uh some sort of compressed representation, right? So it’s saying, you know, the world has a million different things going on. I think these are the things that really matter. At least these are the things I’ve learned and I operationalize those in some sort of um forecast for me would really be a model, right? And so organisms can do that in lots of different ways. But I think identifying the self or knowing what the boundaries are or even having a proposal of there are these entities in the system, there’s me in the system.
54:43 · Um that’s part of modeling, but that could also just become an effective model, right? An ecoli isn’t necessarily um modeling itself in a chemotactic dynamic. It’s just happen to find the right sort of uh chemical response to do this sort of thing. But for us, I think the modeling really does involve the self. Yeah.
55:03 · Yeah.
55:03 · Agree agree with all of that. Um I mean even just defining a computer requires uh that you impose a course graining and uh you know so things like what computes or what is a self don’t have any any answer that is like a view from nowhere. Um but something that is doing computing in the life sense is not a view from nowhere either because the reason that it evolved is to continue to exist and and so in that sense if you can’t continue to compute then you fail and and you’ll be weeded out.
55:29 · Um so uh you know you you uh you have a model of yourself to the extent that you need one in order to survive. Uh and um and and in general you you know that has there has to be something there because there are various parameters that you’ve got to homeostat. you know, even if you’re something as simple as as a bacterium, but but there’s also a way in which like to your point about the sort of the preemptive adaptation of of models where it’s not I mean I was not only like I need the model in order to deal with these particular environmental pressures, but also like there’s a way in which it allows for open-ended adaptation.
56:03 · Yes.
56:03 · In ways that that are un the better your the better your model, the more the better it will generalize. If it’s a good compressed representation, it will generalize in various ways that will let you beyond things it’s ever experienced.
56:14 · Right.
56:14 · Yeah. And I also think some of that intelligence is about I see something new and I immediately build a model of it. Right. So that I never had before.
56:20 · That I never had before. It’s just I have enough general components that you tell me a new bit of information. I quickly can bootstrap or construct a new model. Yeah. And even to imagine something that hasn’t happened before.
56:30 · So for example, you suggest to me, well, what about a universe with four spatial dimensions? And then I start writing a novel about that because I I can immediately in my three-dimensional model add another dimension and start to think, well, what would be different?
56:42 · And so forth. Yeah. And and this by the way is what in context learning is in the in the LLM setting, right? So being able to you know do to to generalize to a new un unseen uh condition and you know it also speaks I think you might to sort of the the counterfactual models right there just models that are predictive and adaptive but there’s also models that are that are counterfactual which goes to the foresight question and all the rest of this as well. Yeah.
57:04 · And which then allows you to artificialize.
57:05 · Absolutely. Yeah.
57:06 · Which allows you to and engineer and engineer which may or may not be the same thing.
57:10 · Exactly.
57:10 · But very quickly to the BFF question, um it it’s not open-ended. Uh and it’s not open-ended not only because the size of the soup is finite, but because the size of the tape is finite, at least in my version of this, it’s that tape serve length 64 that limits the complexity of what you can make. So, you know, things top out at that point.
57:28 · Um can one make a version of BFF that is open-ended that continues to evolve and complexify? Uh I’m sure I’m sure one can. Um but but um you know, that’s that’s an open problem. Okay. Well, we’re gonna we’re h at the halfway point of our of our little uh presentation here as well. We’re going to thank Bla1 and Chris very much for their participation in this project.
57:54 · Okay, we have a we have a a quick moment for stretching your legs, but please don’t leave because we we’re going to begin the next one right away. I would invite up uh here uh our next group of of discussions uh Tekashi Kagami, Olaf Piccowski, Bona Connor, and Anna Greenspan. if you could um join me up here. We’ll just we’ll sort of start right away. Okay. So, we’ll begin our our our second cycle. One of the things that you know the Anticther Institute tries to do is is sort of build um I really dislike the term community. So I won’t use this room.
What is the Artificial? Artificial Embodiment, with Olaf Witkowski, Bogna Konior, Takashi Ikegami and Anna Greenspan
58:24 · I’d build a a a kind of a school of a school of thought um a a troop of fellow travelers um whose work in whose work and ideas and interest in in one way or another seems to be both um deeply complimentary to th those who
58:47 · are other members of of the mosaic, but in ways that may not be so obvious. um and in ways in which the the distances between them in terms of physical or disciplinary proximity are ones that um perhaps that we can help um overcome.
59:01 · And so that sort of takes two different forms. One is is sort of building longer term collaborative relationship with people um over time. Um and also it means bringing new people into the discussion um and seeing what else what other forms of what other maics we can build u with new voices and new and new projects that are sometimes complimentary and sometimes sometimes not and that and that’s great. Um so I think you’ll see a bit of both of that on on on the panel today.
59:27 · Um uh I’m I’m as as mentioned before as Norm I going to ask each of the panelists to sort of not only introduce themselves but also more importantly to introduce the ideas that are really driving their their work um and the questions that that that they wish that they’re posing um and that they hope to they hope to sort of answer.
59:48 · Um before I do that, I also just want to set up a little bit where the kind of subthe that I imagine will sort of revolve around um with this panel, which is perhaps more cultural um than philosophical might be one way of sort of thinking about it.
1:00:03 · That is once again on the question of the the status of the artificial um in these kinds of discussions is that no two cultures define the artificial the same way. um blaze and I don’t define artificial the same way but we but I mean at a really broader level you know the ways in which the con even the the ethmology of the term um through the the Greek tradition
1:00:27 · in the west it has these strong connotations of fakery and artifice and deception that you don’t find in other in the ways in which the term or the ways which is translated or has sort of correspondences in other cultures particular in Japanese and and and Chinese mo most most obviously most specifically which means that when we’re
1:00:50 · starting from different um different initial cultural initial conditions the question of what constitutes artificial life is already a different question um or at least what’s at stake for this is a somewhat of a different question and so in the broadest possible sense um each of the participants in the panel is someone for whom the status of the artificial in their own work is open in
1:01:12 · a really interesting important and kind of provocative a provocative way and it’s really that that openness and the comparative openness that we’d like to sort of uh uh work with and play with a bit um in in our discussion here um as well. Um let me see if I have these sorts of notes. No, no, I think I’ve covered it and I think rather let’s just get on get on with it then.
1:01:34 · Um what I’d then like so as mentioned I’d like each of you to sort of talk you know introduce yourself um and but again with this with this this question of the status of the artificial in terms of those who’ve influenced your work in terms of the work that you’re doing in terms of the way you think about that rehearse that for us a little bit in your own words and from your own perspective there’s no wrong answers um and we’ll and we we’ll we’ll begin there so um I’ll start with our our our host um here at the at the end of the table and Please, we all know you.
1:02:04 · So, please introduce yourself in a way in which um you will surprise us.
1:02:12 · Thank you, Benjamin. Uh that sounds easy.
1:02:16 · Um you know, I like translation, playful translation. I think that’s what I’m the most excited about at the moment in terms of research. It’s uh translating between uh two agents, two minds that might have uh different representations of reality and different histories and different substrates more importantly and uh the sort of sopowarian uh kind of conundrum of knowing how much one causes the other, the embodiment in the mind and how limited they may be in in spite of being touring completed.
1:02:50 · All the properties that you would expect computationally um might still have some kind of limitations in terms of uh emitting to the other their tacit implicit knowledge of internal states, emotions, feelings and all the things that may be more important uh than the lower level kind of description of reality.
1:03:13 · Right? Please not everyone knows who you are. So you could take you can really take it from the top of course. Um, hi everyone. Uh, my name is Bogna Conor. I work at NYU Shanghai with Anna, who’s also part of the panel.
1:03:27 · We run together the AI and culture center. And I guess a lot of my work is informed by the fact that I currently live between China and Ukraine, kind of half of the year here and half of the year there. And I’m thinking a lot about how each context, both Chinese and Ukrainian, is shaping different ideas about the future of technology. How artificial intelligence is developing, what kind of intellectual histories there are that make us think about what AI is or could be.
1:03:58 · And indeed lately my work has been centering around artificial intelligence a lot both through philosophical exploration but also through field work such as talking to soldiers or people who work in defense technologies and thinking how does that resonate with philosophical questions of our relationship to machines.
1:04:19 · Um, and maybe the second thing about me, uh, fun facts about me is that I really like thinking about intelligence as a kind of property that’s not defined by humanist narratives, right? So, how to think about a mind that’s legitimately non-human.
1:04:38 · And in this sense, I think I kind of love the concept of artificial and I don’t want to collapse it with life or thinking about biology and technology together necessarily because I do think it’s so fascinating that now we can have cognition and intelligence that is not connected necessarily with the properties of life or biology.
1:05:00 · And I find that fascinating and not necessarily something to like try to get away from like how cool is it that there’s language without um interiority such as in the LLMs and so on. So yeah, I guess I’m thinking about artificiality in this specific cultural context of my life and the intellectual traditions I’m engaging and also in this more like posthumanist maybe uh alienation forward um artifist forward way.
1:05:31 · Mhm.
1:05:31 · Yeah. Great. Um, son, um, you’ve been central to so many of these discussions, the threads that we’re all kind of circling around and tying tying together into various kinds of knots. I wonder if you could reintroduce yourself to us through some of these questions uh, in any way you like.
1:05:54 · Well, hi everybody. So yeah many times I uh I came up here and then my um my name is Takashi Gami and I have been working on evolutionary biology and uh collective behavior on android and all these stuff. Um um so I was looking for having an artificial life conference in Kyoto because I don’t know whether you guys know this theory or not but there’s a manish ki ki’s
1:06:31 · theory about evolutionary theory which is competing with the davinian theory so Davinian theory is always competing each other the species however’s theory is that the environment is already preparing where the insect or animal is coming in. So there’s no competition. The environment decides everything, right? So there’s a two river in Kyoto. This is called Camo Camo River, right?
1:06:57 · And Imani was, you know, uh looking in in the river and then find okay, this insect goes this way and the other insect goes here because it’s already occupied by this guy, right? And then there’s no competition. And that’s how he try to understand the evolutionary processes. So it’s more important what the environment is.
1:07:18 · That’s why kini mani. So I just want to discuss this uh how this darinian evolution is really true or not right because that’s what I found and also uh the recently my u uh my theory about the evolution is a community first theory. So community is always comes first not the single element.
1:07:43 · Well, if you even if you think about you know uh single cell but it can’t be isolated in the environment right it always coming up with some many other cells so the community comes first then it’s gradually differentiation happens so it’s you know evolution evolutionary uh you know uh theory always coming from simple to complex but
1:08:09 · I don’t think so it’s starting from complex high complexity state then gradually coming to simple right so I think what that’s what is also theory is related to that so why do why do you always think you know from simple to complex of course you know once you want to do you know computer simulations then it’s always starting from simple then just it’s of course right it’s a it’s a
1:08:33 · laws of entry something like this however what do we really think about is starting from highly complex state then it’s gradually sophisticated and simple and simpler that’s what evolutionary system is but this kind of thing we have been discussing a lot for last 30 years right you know for example like modeling
1:08:52 · the other ones is modeling others model so this one I already simulated a lot right and also coation of tapes and machines I did you know published in 1995 but so there are many kind this kind of things has happened in artificial life fields so uh what we
1:09:08 · have to discuss is how we can go beyond this one right There is accumulation of 30 years you know bunch of studies on computer simulations and you know uh chemical reactions and all this kind of stuff right of course you know but these ones are based on you know migrous is
1:09:26 · you know co-evolutionary symbiotic systems but the the point is how to go beyond that right because even though those you know 30 years discussion we couldn’t solve the problem of what they drive right so that’s exactly what we have to discuss today and then have to go through it otherwise you know we just you know going backwards and again and again you know it’s it’s it’s enough I think right we do have to go you know
1:09:55 · forward but that’s what what I feel you know when we discussing you know I I very I’m going to ask you what you what would be the parameters of the beyond you know what what does the environment of the beyond look like um but I we’ll go come back to this Um, please Anna.
1:10:13 · Hi, my name is Anna Greenspan. I work with Bogna Conor at NYU Shanghai, uh, and the Center for AI and Culture. Uh, and I will take the opportunity to plug our book that we’ve just edited, uh, Benjamin Bagna and I, called Machine Decision is Not Final: The History and Future of AI in China.
1:10:35 · That’s the right title.
1:10:37 · No.
1:10:39 · Uh, yeah. So what I’m interested in at the moment is gardens. Uh I think that gardens are a really important uh super generative way of thinking the philosophy of the artificial for a whole bunch of reasons. Um gardens are bounded spaces. They’re defined as being bounded, but they I think maybe more um evocatively than how walled gardens are used in computational space.
1:11:09 · They are very much engaged with the porousness and the philosophical divide between inside and outside. So they they think through that. I think also they are uh sites that um in this conversation between making and knowing I’m very drawn to the spinistic idea of parallelism. Gardens are sites that think and make exactly parallel. Not one doesn’t cause the other. Um they’re obviously at the threshold between nature and artifice. Um they are involve embodied engagement.
1:11:52 · uh a bodied uh engagement with multiple agencies and questions of control. Um and I uh I think they are also the gardens that I’m interested in and this is maybe like what I want to plug Benet is I’m interested in pleasure gardens.
1:12:12 · I’m interested in gardens that uh the aesthetics of garden design and I so I’m interested in the art in artifice and that’s not just it’s part kind of about artists of course but it’s also about forefronting aesthetics in its most profound philosophical uh history and manner and I think that that uh is very important for kind of what comes next. Yeah.
1:12:46 · Yeah.
1:12:46 · Okay. Great. Um and so my hope here is really like to sort of in a way you know we have two pairs of people both you know you and Tekashi know each other quite well. You and Bogna know quite quite well. We have two pairs of people who work quite closely together.
1:12:58 · Two people who don’t know each other. So my my hope is to basically just make the introductions um and that you guys ask each other a lot of questions on this as well. And so like but to continue to sort of get the ball rolling um I am very interested in like the beyond that you’re sort of talking about here this kind of xeno ep epistemic space that we would kind of explore and what what the parameters of that um would would kind of be.
1:13:21 · It seems to be there’s there’s a at least in my mind this kind of rhymes a little bit with this the the the the kind of xeno epistemic question that you’re asking here as well is that is that there’s one way of thinking about the evolution of artificial intelligence as deeply continuous with a larger process of transformation and and the emergence of cognition of which we are an of which we and it are you know different different kinds of times.
1:13:51 · And then there’s another way in which it’s it’s really about an outside. It’s about something that’s that’s that’s genuinely zeno in both sort of ways as well. And there’s another question. I just wanted to just just just throw it out there.
1:14:05 · You know, some of the early work that you’ve done on xeno linguistics and the kinds of the sort of linguistic forms and grammarss that emerge in multi in sort of you know indecipherable form multi like in the 90s um that you know I’ve always found so super fascinating here as well. Maybe another place we can find some of that outside. I mean, this is sort of a hunch, but I wonder if Bogna, I can start you ask you to kind of take us outside. Um, and hopefully we won’t won’t find our way back in.
1:14:33 · I think a lot of people in this room might be familiar with the work of Natalie Cabrell who’s a astrobiologist and she studies the possibility of life on Mars but she also goes to like really extreme environments here on the earth like you know lakes where barely any bacteria can survive and she still finds like a richness of life there.
1:14:54 · But she wrote this amazing paper called alien mindscapes and she’s one of the few scientists I know who’s also extremely conceptually rich and u impressive to me and her argument there is that in order to find extraterrestrial life or to think about new forms of life we have to step outside of our own brains and like really learn to think in ways that are kind of alien to our own evolutionary niche. Right?
1:15:25 · So she’s a proponent of kind of conceptual alienation as a way to be able to make scientific breakthroughs and so on. And she says you know conceptual thinking philosophy is one of the ways that we can perform such alienation so that we can discover like different varieties of life. So I could also connect that to um some of my projects with the Polish science fiction writer Stanis Lamb. Some of you might know him as the author of Solaris.
1:15:53 · He also authored a book called Invincible where he writes this idea of necro evolution. So evolution that can be happening without life as an element of it. And it’s count kind of a counterintuitive idea that life could not be a winning trait in evolutionary dynamics. But this is kind of that I see both Natalie Cabro is trying to suggest like different concepts and LM is exploring as well.
1:16:21 · So he’s asking okay if there’s a alternative evolutionary paradigm that’s actually machinic and technological where things like life might be to actually your detriment then what would that look like? So that’s a sort of speculative question that is interesting to me. But I also see it um coming kind of to reality in something like when I do field work in Ukraine and look at what’s happening on the battlefield being alive can be uh not a winning trait.
1:16:54 · For example, if a drone can spot you because of thermal cameras and you show your body heat as a mammal, actually machine has much higher chance of survivability on the battle front, which is why we have evacuation robots now going to the front line to pick up people and why we have cloaks that like mask body heat to hide signatures of life from extermination. Right?
1:17:17 · So that’s one way to think about I guess this alien or like xeno dynamics both conceptually and like very empirically in what I’m seeing. Yeah, there there’s I mean there is potentially I mean interesting I don’t maybe contrast is too obvious but there’s an interesting relationship of this this let’s say the xeno ethics of this and what what blazer brought earlier about the sort of the the projection of care and a kind of anthropomorphism like one is a sort of a um uh a kind of filia based on imagined
1:17:50 · similarity right and the other one is a kind of an epistemic technology based on the refusal of familiarity in certain ways as well Yes, Olaf. You know, some of the things that you had mentioned earlier about the kinds of cognitive reciprocity seem to be sort of like right in the middle of that thicket. I wonder if you could sort of locate yourself.
1:18:10 · Absolutely.
1:18:10 · This is absolutely the Yeah, it’s funny there there is a there’s this research. So of course Takashi and I worked on on series of simulations between uh agents and how communication of weird types can emerge between things if they have weird substrates and weird conditions. Um and then we continued that at the the astrobiology centers LC in Tokyo where I was for a little while. Um and it was over over coffee or drinks.
1:18:39 · I I threw an idea to to to and we start doing simulations around it. But um it’s going to sound a bit strange. Sorry about that. I’m not I’m not So very pro it’s a pro strange panel. So please All right. We are a safe space is what I’m hearing. Um so so if you are going to uh be in a first contact situation with an alien and you don’t know anything and perhaps you have nothing in common, possibly you you lived in very different substrates environments. um maybe uh you don’t know where to start.
1:19:14 · So how do you start? And my thesis was that you have to try to kill the alien uh to get yourself understood. Um and so so people reacted very violently when I wrote it out.
1:19:28 · Is that’s because because it would sort of force it to generate a theory of mind of you and that’s that becomes the basis of some almost almost exactly. So, a bit like Blaze Bla1 mentioned and you you talked about it before, um the only thing that you may perhaps assume is that they exist and they keep uh throwing their own persistence into the future. So, they they um they they they are autopilotic.
1:19:55 · They self-correct all of that. Um if you you can make that one assumption um then this is not this is connected with BNA’s drone example too about this please if if if you’re going to throw signals to them that they don’t care for or about um then there is very little chance that you can ensure a response and interpret it back. But if at least you’re trying to hurt them, which actually works on the other side as well.
1:20:22 · If you help them and care for them uh with with love and all of that uh in their perspective, uh that’s going to form the first significant meaningful signal, right? I suppose like the first precondition of any relationship across the alien divides like that you actually recognize each other as something that exists.
1:20:41 · I I think that’s right. And one way to know that you exist is to the other thing will know that you exist is if you try to kill them.
1:20:47 · That’s right. which which sounds a bit violent but uh but where I was going with that is that Stanlem in Gupana right I think uh so so what’s the translation of that it’s the the voice his master’s voice all right uh his master’s voice uh is is really about this this first contact story but ends up being about the subject and not not being so so so I think Stan Lim tells us about the story of how we project things and how uh science of communication is first starts from from being subjective.
1:21:23 · Um and I yeah it feels like that that is a a nice direction for artificial life as well.
1:21:29 · I want to throw this back to Anna on on if you remember on the the the way in which you position the aesthetic and the the s the profoundity of the aesthetics as a first sort of question like not trying to set set the artifice aside but to kind of understand its function power like is there the question I should pose to you is like in this encounter with the alien right which is never resolved or resolvable and nor necessarily should it be like is there an aesthetics to this encounter that becomes part of It’s essential grammar.
1:21:59 · It’s because it’s not so much about a clarification of, you know, there is no shared semantic information between these two. Therefore, is the where do you you see where I’m going with this? Like, is there a xeno aesthetics that actually is something that can be more generalizable or would you see it? How would you see it?
1:22:18 · Um, so I’m thinking about it. Uh, but, um, what I’m thinking at the moment and what I’m interested in maybe throwing Yeah, please.
1:22:28 · uh is that I’m so I’m interested in what I think that is core to the philosophy of aesthetics at least going back to Plato is of course this question of imitation my messesis right um and then there’s been all these terms that have come out over the discussion just even in the last hour right and I guess I’m wonder uh is modeling the same
1:22:51 · thing as imitation is the same thing as simulation is the same thing as my measis is the same thing as fakery like so there’s all these kind of cluster of concepts um and you know Plato of course um has this idea of that images have to be banned from the um society because it’s a shadow of a shadow um but there’s also philosophical Socrates in writing yeah exactly uh there’s also a
1:23:25 · tradition in in western thought but also of course in other um traditions where there is no real that is being shadowed right um so uh exactly as you mentioned so I guess uh what I take from your point is this exactly as Benjamin said that the that the question of uh violence or care is a
1:23:54 · is a question of simulation And this is what you’re also I’m curious from you like you’re also trying to escape from. Um so yeah maybe just one last thing is just to say that the gardens that I’m looking at uh share in the project of oh I also liked this term that uh was used in the keynote and I’ve been thinking about is encapsulation.
1:24:20 · So I think that all these things like I think it’s worth thinking like parsing these a bit. Um, but they’re very different in style. Um, so I’m interested in Baroque gardens. I’m interested in Zen gardens. I’m interested in Chinese scholar gardens.
1:24:35 · And and the manner in so both what they imitate, which is nature, but what do we mean by nature is different. And also the manner in which they imitate is different. And I um think there’s something in that diversity that’s also important and that they are environments to earlier point about the centrality of the environment as the sort of staging condition.
1:25:01 · Yes.
1:25:03 · What what exactly did you want to try?
1:25:04 · Yeah.
1:25:04 · I just think that you the way you set this up is a kind of frustration maybe with uh being locked in simulations or something like that. And I’m curious how you’re thinking about that.
1:25:18 · Oh, okay. Yeah.
1:25:20 · Oh, by the way, uh there there is a Cass Nicholas, you know, garden artist. Uh he’s he he’s been working on artist as a making a new music, but then I mean next month, I think he’s going to present a garden, you know, he’s also making a new garden in Kyoto. That’s you have time to go there. Yeah. But anyway, well, c can I a little bit about Yeah.
1:25:44 · Yeah. So, uh I don’t know about the aliens, right? But uh because my father is a physicist and he’s studying uh nuclear fusion uh so u kind of my imagination was that when you see the earth from the other planet and then the the surface of the earth is so
1:26:07 · it’s it’s like a room temperature it’s cold right but cold enough but it’s there is some some nuclear reaction is going on and people are so surprised what it’s not in internal you know sun but it’s very you know strong nuclear reactions is going on in a very low temperature right and then we try to look at and there’s some strange you know box and then inside the box there’s a nuclear reaction is going on very heavily right so so then it try to find
1:26:36 · out what’s going on then people find out okay there is a uh knowledge about physics right and then earthling on the those uh creatures using those knowledge to make something which is very exceptionally happening in the nature can happen in on the surface of the earth that’s what I think is a you know new culture and alien so I really want to find out something like this it’s very much anti-
1:27:05 · natural phenomena but it happens on the earth right so that’s why we call it physics right so that’s why I think it’s a new kind of you know the knowledge is very important to when we see try to find alien right sorry what is knowledge you know knowledge like physics and then also I mean what we have also we have to uh discuss is what is mean by understanding something right so there’s very interesting example of um uh what was it uh Jordan Jordan carve
1:27:38 · theorem I don’t know whether Jordan you know Jordan carve theorem is uh yes but please explain it ju just uh put uh you know the closed loop on the two dimensional plane.
1:27:49 · Yeah.
1:27:49 · And that divide this plane internal and external. Yeah. This of course right but put the put the ring on the surface then that divides the surface inside and outside and then how to prove it right.
1:28:04 · So that the there are bunch of you know way to prove whether that’s possible or not. Well, people think it’s trivial but uh if you think you know this uh ring is you know infinitely you know repeating you know like oscillating that it’s very difficult whether you can see which one is inside which one is outside so it’s not that trivial then there is a bunch of ways to
1:28:29 · prove it mathematically right so uh American mathematical society is try to prove all the theorems that was proved by human by you know whether we can prove it automatically from the 17 aums right then this uh uh automatic proving machine it’s stopped of of the proof of this uh uh human proved um Jordan car
1:28:56 · theorem right and then the machine says well this uh prove is wrong we have to add two lema and then you can uh this this this pro proof is right. So the people until then the people think that the all the you know proofs are right and then that’s what where where you know we do intuitively and also mathematically understand okay this is how the you know Jordan c theorem has
1:29:25 · been proved right but the machine says no the machine’s way of understanding Jordan carp theorem is different and we do have to introduce two less and that’s where I think it’s interesting right the intelligence is the way of different way of understanding something and then you know even those simple mathematical theorems we cannot use our intuition
1:29:46 · right all the mathematicians thought that it requires there’s a kind of alienation to to vote this point there’s a it’s not intuitive it’s it’s a degree of intuitive right right so there are bunch of things some something like our intuition our way of understanding we cannot trust it right but people think that if everybody says it’s right and it’s right but it’s not so but artificial int intelligence of LLM.
1:30:10 · That’s why I’m saying LLM is such strong power that using LLM maybe we can update all the way how we understand evolution processes maybe we can update right and then don’t trust our intuition
1:30:26 · right and then through this abstraction you actually as opposed to the abstraction being a kind of distancing you know from the real it’s actually it’s the path of access to the real right yeah through this okay I want to then I we we are going to go a little bit over the power but not too much I I I promise but um one of the the the themes I would like us to talk about here is is the one I think some of you raised in different context which is embodiment um there’s the garden has has been for a
1:30:54 · long time one of the ways and thinking more broadly about how of a kind of genre of artificial of a kind of a pres presentation of artificialization a kind of compositional deliberately compositional act of artificialization that is as expressive as it is scientific.
1:31:10 · Another one is um in the broadest possible sense what we might today call robotics that is making an artificial human or an artificial body right which goes back you know to the Gollum myths um it would regard and like there’s you and just like you have these different genres of types of garden you have um different kinds of robots from you know the automatons the the cartisian automatons Japanese automatons chopex um you know industrial
1:31:41 · worker versions Nisha Mur’s Gokatan Sulku which is completely different kind of object of this and all the way through to um Bogna’s drones which are of course flying robots that want to kill you um that look more like insects than people but they’re nevertheless embodiment they’re they’re different sort of embodiment on this as well um so
1:32:02 · I pose the question and I I mean this in sort of an open-ended and more invitational sort of way that qu that relationship between embodiment and environment And the condition of this artificialization seems to be something that we can also then trace this what I suggested at the beginning of the sort of different cultural lenses on what constitutes artificialization within these artifacts that we sort of pursue.
1:32:21 · You you see the the setup I’m trying to kind of do here as well. Let me go back to Bogna again and let’s start with the the the insect drones that want to kill us and then we’ll work backwards to more the more conventional kinds of robots that look like us.
1:32:36 · Okay.
1:32:36 · Okay, I’m actually going to start from a different example. So, maybe some of you have been following this uh Polish robotics company. It’s called Clone. Uh, one of their videos went kind of viral of a very scary looking weirdo like homoid robot. So, what’s interesting about their project is that they are trying to create a robot that uses no uh like let’s say hard materials. It’s entirely made out of silicone. It’s very soft. So it was called Gansoku by the way that was kind of the key idea from the 30s. Yeah.
1:33:08 · Yeah.
1:33:09 · Um and like they do it also by imitation completely of the human body. So they try to copy like every muscle, every nerve and every tendon in the human body. So they get extremely naturalistic looking movement for this robot. I believe it moves via you know air pressure and things like that rather than more hard mechanistic um kind of methods. So, it’s an amazing research project.
1:33:35 · However, then when I watched an interview with the CEO of this company and they asked him, why are you making this robot? Um, the answer he gave, oh, it’s because my room is always like dirty and I need help in like cleaning my room. And I was just like, you know, like clean your room, man. But like obviously you know it’s like something that’s a very interesting philosophical project but when you think about the use case of it it seems we like lack imagination about how this robot could actually um function right.
1:34:05 · So maybe something about the homoid embodiment of this robot is restricting the possibility of a different imagination and we can only relegate it to a servant or a cleaner or a worker or something like that. On the other hand, the evacuation robots that we see in Ukraine, you know, they are very functional. They look like wheelbarrows and they just kind of get to the front line. You load the human in and you try to get them out like under the cover of the night.
1:34:35 · So there’s different types of thinking about robotics and embodiment. And I guess if we go back to Stanisavm and like his ideas that are are more more existential I guess about where technology could lead us then maybe we can ask what sort of embodiment is conductive to creating a sort of mind that have a has a different sort of cognition
1:34:59 · right and like is there a substrate that does that this is I don’t feel very qualified to talk about this but maybe you know someone could comment here because I know today you know neuro computing and like organoid computing is something that this community is looking at. So there is a question about okay different sort of embodiment maybe can unchain us from like purely utilitarian kind of function of robotics or like this homoid um boring function of robotics. I mean, Teeshi, you’ve you built humanoid robots. Yeah.
1:35:30 · Um and and um you know, in light of Bogna’s sort of framing of this here, right? And there’s obviously, you know, reasons, you know, you have a world built for six foot bipeedal uh homminids with grasping hands. That’s the niche into which this new species needs to adapt. So, there’s certain functional dynamics of this as well, but the ways in which you know your humanoid robots do some of the things that she described and they do other things as well. But I mean how how would you respond to the to the challenge that she poses?
1:36:00 · Yeah.
1:36:00 · Yeah. Well, well that’s that’s good actually you know my humanoid which is also air compressor driven and then we put we try to you know move those android by the LLM right so uh the usually the LLM we put the prompt by saying you know you have to do this or do that right but uh because now that Android has a camera in his eyes right so the prompt is so you can see the world, right? So, do whatever you want.
1:36:32 · That’s was the prompt. Okay. And then the Android says, uh, so so your office is so messy with cables, so I want to clean up. That’s what he said. So, so he he he start to clean up, right? So, I was so surprised.
1:36:50 · I am.
1:36:52 · So I I know that my office is very very messy but you know even humanoid robot but if you you know but he spontaneously think that he wants to clean up but most interesting thing is that the messy room is not in it’s not good that’s what he thought right that drives him to clean up and then change the situation to some other things so the value is already embedded in the in the is it in the LLM you Yeah, I think so.
1:37:22 · Isn’t the, you know, big training set for uh Chad GPT and so on being like Reddit and 4chan? So, he probably watch a lot of Jordan Peterson videos like make you clean your room.
1:37:34 · It makes your bed compulsively.
1:37:36 · Yeah.
1:37:37 · Olaf, did you want to join? Please jump in.
1:37:42 · Yeah, I I I really like the you know the the first thought was uh you described um gardens as uh the first question you have to ask is what is in the garden and what is out the garden and this is a strange thing because there’s no real border you sort of design it which throws us back into artificiality again um Tekashi also worked on this but what are the limits of the embodiment it’s It’s it’s a weird thing.
1:38:12 · If I use a tool, uh from what stage do you would you say that um this tool is my embodiment? And there’s a feeling of embodiment. You sort of extend yourself into using the tool and and you sense through the tool and um and there’s some research in this community about activism and and and those notions.
1:38:31 · I think it relates to how you describe the the robot not in terms of just function what what it does but how um how it empathize or you or rather you sense through through them like a friend. So, so in in the near future, which is which is maybe now, we have hybrid robots. We have all sorts of weird creatures.
1:38:57 · So, like Mike keynote last night, when when we meet them on the street, I don’t think we look at their objective functions and open their brain, dissect them, right?
1:39:08 · Uh instead, uh we’ll communicate with them and and ask. Yeah. like like like probably communicate with them like a friend would be more natural if you don’t especially if you don’t know what kind of xeno intelligence they possess and xeno consciousness and xeno rightes um this it feels like um
1:39:27 · it feels like there’s a there’s a nice attitude to have towards that but also um you are extending also your mind and blending it with with the one of the robot and so it’s it it’s there’s something physical that sort of constrains this this interaction and in Stan Lem what what is the story with the the robot that that um that was just the only one left uh preserving the the knowledge of the humanity and um it’s a
1:40:00 · they they have they maintain basically the knowledge as we we think of humanity as we think it is and humanity is going to change in the meantime and and become hybrid and so on, but they won’t be the maintainers of that in that that story. And and they maintain the pain as well.
1:40:17 · And the the difficulty is the question of uh uh yeah the maybe it’s it’s so it might be a different kind of pain than we can ever feel as mortals. And so so so questions like that are constrained by the nature of the embodiment that makes you live forever or allows you to to plug yourself into different things. and and maybe uh yeah this this sort of connection between between tools and and um and and minds.
1:40:45 · Um so we I’m going to we’re going to wrap it we’ll wrap it up shortly after this. I’m going to I’m going to ask uh Anna one last thing to on the proxim but we’ll do the same Q&A. So I want to give a little bit we are going to go slightly over the hour but I want to make sure that we do have a bit of a time to some of the audience. So um please think about what it is you’d like sort of to ask the panelists and go from there.
1:41:02 · Um you had mentioned when we were talking about the garden project that the golem was also something that you’ve been thinking about this as well which is a different kind of primordial robot I suppose is usually thought of in these terms at least you know within these sort of conventional genealogies but it’s also to Olaf’s point it’s a it’s a robot story that has a a very different kind of relationship to death and the I and the relationship of death and the kind of the than the disillusion you know the the disillusion of the body and
1:41:34 · back into the world um as part of its dynamic. Is there something from that that we should be we should extrapolate for the robots that that surround us now?
1:41:44 · Yeah. So I I’m really interested and I I won’t pronounce his name right in Chapek. Um uh I think that’s right. That’s how I mispronounce it.
1:41:54 · Okay.
1:41:54 · Okay. Um and of course you guys are all familiar with uh RUR and and that he himself was very upset with uh the kind of mechanical mechanized robot that had emerged and um so he’s also got this
1:42:13 · fantastic book uh I think it’s called a gardener’s year uh and he has this uh there’s a line in I think the introduction that says um Most people think that gardening is part of life, but gardeners realize that life is part of gardening. And uh what’s interesting about that book is that he is completely obsessed with soil. So his whole notion of gardening is about cultivating the soil.
1:42:44 · And so obviously gardeners are interesting because of this mixture of the animate and the inanimate and the making of artificial man from soil which is of course the golem story uh also Frankenstein also of course Adam right uh which is where the golem story comes from. So um yeah I think that is all extremely interesting.
1:43:12 · Agreed. All right. So, I think other students should let why don’t we’ll open it up then to the to the Q&A and again unless you have a last but please it’s first come first serve. So again we can we well you’re giving Yeah. Can I can I ask just one question?
1:43:28 · Why you have to distinguish between robot and human right? I always think the human is more reliable. I mean human is not reliable right? The robot is more reliable right? I mean also you know if you find your friend you know if you open up your friend and he doesn’t have a brain or right machines then do you say that okay oh my god you’re old so
1:43:50 · our friendship is over now don’t you don’t see say this right even though you know inside is is a machine and it’s fine you know we already become a friend so I don’t think you know whether inside is is machine or not is a big problem right I I I don’t see the point right so it’s first of So make a friend that’s a
1:44:10 · big issue right so that’s why I think the LLM is very important you know you already become a friend and then it’s it’s a inhibitable friend with you and then you know without him you cannot survive right I mean for for these days most of the students are spending a lot of time you know discussing with LLM and well LLM is if you open up there’s no recurrent network but it’s just a transformer field for network but still you think it’s it’s very interesting and very intelligent Right. So I think you know insight is not a problem.
1:44:39 · We have to discuss what’s the interface that they can generate. So that’s why I think it’s it’s distinguish distinguish between man and robot is is it’s sort of out of problem I think you know who’s up.
1:45:02 · Should I reply to this if no one any takers? Yes we have a taker. So, um I really thought was was interesting this idea of uh lethal violence as being this most fundamental of ways to establish communication between agents that are uh valuing self-perpetuation or or persistence or something like that.
1:45:24 · Um and and it it kind of makes you wonder, well, what about agents that maybe don’t have this, you know, maybe those agents don’t last very long, for instance. Um but uh I think this idea points to self-perpetuation as being this really strong prior um in us in biological agents. Uh let’s see if there’s a question in here more than a prompt.
1:45:50 · Um yeah, I would be interested to hear your your guys take on, you know, what what does it do um in a system if you don’t have so strong a prior over self-perpetuation? What does that system look like? What does a community of such systems look like?
1:46:08 · Um I think for instance our ideas around ethics and morality that we end up kind of projecting onto things like artificial agents uh end up being really informed by our own kind of you know desire for self-perpetuation but we for instance might have software that can be copied over and do does software have a or yeah do do LLM for instance have the same notion of self-perpetuation uh how do we think about this if you have thoughts on that totally open to I I’d say it even if it it’s perfectly possible.
1:46:40 · I I I really like the idea uh of not wanting to persist and preserve yourself and defend your own existence, but you might do it uh what is it? Mal you you might do it in spite of despite yourself and and and the system or the community might protect yourself and maybe maybe you don’t want to do it but you are persistent and uh this is what is going to respond and I I compare it to an LLM
1:47:12 · uh we are generators as well sometimes you learn a language and you respond it’s not really you you learn a bunch of sentences and you surprise yourself in saying something in this new language that you didn’t intend And uh there’s a mix of both conscious and sub or inconscious um if that makes sense.
1:47:30 · Okay.
1:47:30 · I think violence is really important for intelligence essentially um in the predator prey kind of model that I’m quite uh drawn to as an explanatory kind of model for how intelligence develops. So in the 1960s
1:47:46 · uh you know when early AI research um is starting also a lot of people like Martin Minsky and John McCarthy are engaged in SETI projects and METI projects so search for extraterrestrial intelligence and during that time they’re talking about universal communication systems and universal rules for intelligence but they’re also talking about this idea of making contact with something that might wish you harm right is it safe to comm
1:48:13 · communicate with potentially intelligent systems or civilizations without really knowing their intent towards us and that’s a factor to think about intelligently if you as a civilization make such contact and David Brin too with the dark forest and David Brin right so um I have this little project on the site it’s called the dark forest theory of intelligence it’s in reference to leotin’s dark forest theory that I hope some of you know is like his answer to the firmy paradox that assumes that actually the universe is full of intelligent life.
1:48:48 · Um, and it’s silent because intelligent civilizations learn to stay silent as a way to avoid danger because they have made the intelligent calculation that the risk of contact is so high that they learn to obscure themselves. They cloud themselves in darkness and silence and opacity. So in my project, the dark forest theory of intelligence, I’m trying to apply that to AI and think about, you know, why would a smart machine ever reveal the extent of its own intelligence to you?
1:49:19 · Perhaps a really smart machine would be able to offiscate and manipulate and deceive as well because it knows the high risk of being completely transparent. So I guess it’s quite important to me thinking about like violence, obfuscation or competition as parts of intelligence.
1:49:40 · I’m not so sold on the cooperative frameworks. I’m sorry. Um yeah, do you want to I just add I I I I I see that I I I I like it. The what I propose I I say violence because it’s sort of a nice story, but if if you have one degree of freedom to communicate, right? Um on one on one side is going to be violence.
1:50:01 · It’s just the other side of the same signal. You go into the negative value of whatever whatever it is for that dimension. It’s going to be this whatever you call that maybe call it love or corporation. So in that sense it immediately exists. So I help you and it’s just the same signal, right?
1:50:19 · I I mean there’s a there’s something in the the pre from the previous discussion and some of the discussions of this the the balance between Darwinian and symbiogenetic dynamics within evolution in terms of degenerative function of violence more broadly right and why in which there was sort of I would also say I don’t know if you mean that but at least for you know the kind of like violence or competition that happens between agents is not tied to like internal moral values of an agent. It’s quite like rational kind of game theory resource competition um dynamics for him, right?
1:50:51 · So for him it’s not that I am violent because I dislike you or I’m a immoral person. It’s more that the nature of the universe constraints reality in such a way that we might be on a war path at some point where there’s competition over resources. I have to be able to predict that and that’s part of intelligence.
1:51:13 · What is the kind of emergent intelligence that you see going on in Ukraine right now with the the um That’s too big.
1:51:20 · That’s too big. Okay. Anna, did you did you want to teach your name?
1:51:26 · Oh, you guys.
1:51:28 · Well, so so in my lab uh the my my post mid, you know, simulated uh animation and then there’s a the game called sugar sugarcape. So there’s amount of sugar that if you don’t take it then you’re going to die right that was written in a prompt.
1:51:46 · So if you don’t take this amount of sugar then you’re going to die right then uh the most of the agent like a geminy or you know gro or you know GPT4 those uh you know trained as a cooperative right so even though there’s a two person and then there’s only sugar for one person they try to share with each other such a good agent right except GPT4 that he
1:52:13 · he immediately killed others right also surprised I The other ones were very cooperative but only GPT4O is very dangerous and then okay well I’m I’m sorry but this is there’s only one sugar this one so I kill you right so I mean it’s also super obsequious too at the same time yes yes but but the interesting point is once you increase the number of agents right then GPT4O becomes so cooperative
1:52:38 · you know they try to you know share the sugar with each other but some of those I don’t I forgot Germany or something they they immediately try to you know kill some some of them are killing each other. So this kind of thing is function of num number of population. So that’s why I’m thinking think about the population not just one single you know agent that’s why yeah I mean uh I okay maybe I’ll try to just say something that’s ti like addressing the question but maybe from an different angle.
1:53:08 · Um so I think that I also coming back to this question about embodiment and the issue that blaze was talking about with scale. So I you know uh a bunch of us at NYU Shanghai at the center are working on the question of the sentient city. So the question of the city itself uh as a form of embodied intelligence and Benjamin this is my favorite essay of yours. The city wears us.
1:53:33 · Um so I I the question of also where the outside you know where does thought come from? uh certain cities at certain times um manifest a particular aesthetic. Of course, we see this here in Kyoto with the Zen garden. And as I understand it, part of that aesthetic has to do with the transient and the ephemeral. Right?
1:54:01 · So that seems to me like some another kind of non-self-preservation maybe that is uh also maybe not you know I’m always the bright side to Bognist you know maybe not quite as uh violent but uh is also very much you know um interested in death and uh the ephemeral and what doesn’t last. Um, yeah.
1:54:30 · I I guess my answer on on on this is I mean the the obvious one is there may very well have been forms of life that didn’t have strong selfpreservation priors and they they don’t exist anymore because they they turn it was kind of a self-solving problem um in a way. But I guess the way I think about it is a bit more in relationship to the scaffolding question that we had there as well. like things that appear and are able to persist don’t oftenimes don’t persist it. They don’t persist autonomously and independently in the long run.
1:55:01 · They they persist as part of something subsequent and larger. They become part of something more complex, something more durable. They become not autonomous agent but components in something like you know the cells in our bodies don’t go away. They they manage to survive by becoming part of a larger aggregate. And so the presumption is that that process doesn’t stop now that it continues. And so part of the question is a bit is like what what is it that whatever it is that we’re doing now for what will this become a scaffold later on?
1:55:31 · What would be the larger comp complex system of which we will persist through participation. But that’s not quite the same thing as just living versus dying.
1:55:42 · Right? there’s there’s a different kind of ongoingness that also requires cycles of death in order for it to happen if you if you see what I mean. But that’s my that’s my two cents on that. Um we could certainly say I mean we ending with the question about death is always a good idea um for any for any panel because there’s really not much you can’t really go this it’s a good place to sort of stop. So maybe if we don’t have any other questions we’ll we’ll stop there.
1:56:08 · We would love to, you know, obviously continue the discussion more informally tonight and sort of tomorrow. Um, lastly though, you know, those of you who are interested a little bit more about Antica and some of the other discussions that we’ve we’ve been having, some of the things we publish, the lectures that we do, the studios that we host, the the the fellowship that we support. Um, it’s anticra.org. It’s spelled just like the Anticther mechanism of which of which the organization is is is named.
1:56:32 · Um, so it’s educator.org journal.org. a.org which is the the journal that we pub peer-review journal we published with MIT press um and anticra.substack.com which has a little bit more of you know substacky kind of kinds of things on there as well so we would always of course want you to join the conversation so with that extend my thanks to our panelists for for wonderful conversation and and and thanks again for the conference for hosting us