Collective Computing: Learning from Nature

    10:17AM Sep 28, 2021

    Speakers:

    Keywords:

    blindfolded

    system

    society

    people

    interesting

    brain

    point

    work

    cube

    case

    cells

    question

    playing

    computation

    memory

    moving

    noise

    circuits

    learning

    principles

    Super, super, super excited to have a RepRap over here from the Santa Fe Institute, they will I am not entirely sure if you remember, but I actually had the pleasure of introducing you at the mental intelligence summit that was posted by Gaia, that seed sent by Anthony a year two years ago now, and we are talking really blew up many people's minds was really fantastic. So and no pressure. And I think everyone can get excited about presentation. But today, you will focus on negative computation and nature, and some of the historical perils of machine models and metaphors. And you are leading Santa Fe's program on collective computation, in addition to having a variety of other roles. And they really work on fundamental problems and evolutionary theory concerning collective intelligence on societies of cells, to societies of individuals to machine human hybrid societies. And that, as the group can tell, fits really well into our thinking that we have in the book, and in this group, in which we treat civilization as a super intelligence that is composed of many individual agents, some are human, some are AI, who voluntary cooperate. And so without saying a little bit, too much, I think I'm not giving the title of the talk away, or the topic, but I think, yeah, I just want to welcome to the group super excited about the discussion, I'll probably kick us off with the few questions after you've presented and, and I'll share more info on your recent writing out here in the chat, very excited to have you here, really looking forward to the discussion to add the status.

    Thank you. Thank you very much, Allison, thank you all for being here. Much appreciated. And I'm going to start with a little history, ancient history, and then jump into some details of what we do on connected computation that is evolved, that is in the natural world to include humans. But I want to make a distinction between two styles of reasoning. And they're not orthogonal, but that I think this group would be sympathetic to and perhaps in the end, we can come back to them. And I'm going to share my screen here to work before So what now.

    And I'm just going to jump right in into classical antiquity, with eroticism technical analysis. And many of you will know religious, I grew up with this book, the histories. And this is a book that does many things, not least ich ins, the most detailed account that we have of the Greek or Persian Wars in the fifth century BC. And wrote, it just was a marvel. I mean, this is an encyclopedia, it's published in the full version in 12 volumes. And he makes this very enigmatic remark, that of all men's miseries, the nature is to use this to know so much, and to have control over nothing. And of course, in in this particular instance, he's talking about the brutalities of the wars. And of course, all of us can very easily transpose the insight of religious into our present time. Fast forward about half a millennium only encounter as Newton, and Isaac Newton one of these extraordinary figures, a hero of course, to many of us, but not only gave us an understanding for how the nonliving universe works in part, but showed us that the human mind could comprehend it in a form that was highly compact, and congrats in this case, in in mathematical for about 10 years after the publication that Principia, Newton was made warden of the mint. And in that capacity, oversaw the famous Sensi bubble. This was a huge market collapse, that followed from the Treaty of Utrecht, which had quite unfavorable terms at the end of the war, the Spanish session, and you've made the point that I can calculate the motions of the heavenly bodies, but not the madness people. So once again, just like your auditors, there are things that we can know and there are things that we failed to understand and certainly failed control. Now, Newton's ideas, which he cannot with an accuracy in France, in particular by Emily chalet, and Voltaire. This is the extremely important book actually, the elements of the philosophy of Newton was actually co authored by shatter named Bill German, unfortunately, only Voltaire's name appears. So this was a problem then it's a problem now that this is a book where the ideas of mechanics were introduced to the broader public. In fact, the Principia was so difficult to comprehend, that many people in the English speaking world had to wait for the English translation of this French explication of the works of Newton to understand what Neil said, Who was chudleigh herself translated the Caribbean that was published posthumously. It's an extraordinary woman. Now, these are ideas through the works of people like Voltron Of course through the encyclopedia, and other works of the enlightenment. Whenever not the restricted justice science, they became aware of understanding society itself. And human pedia Miss famous proponent of this standard reasoning was using Alfredo annachi. This is the entry of the text of understanding humans as machines law machine. And in this book, which starts actually with an epigram by Voltaire. limitary basically claims that humans and this is the position I think many of us will share a no less machines than the universe itself, the nonliving universe, we just haven't understood the basic mechanical principles. And that's no less true for society. Now, the the ultimate realization of this position actually was in the French realized revolution. And in the worship of the culture of the Supreme be, this is a painting of that culture was overseen by Robespierre. And rather than worship divinities, or you would work worship at the village he was raised. So we essentially sublimated all of our needs for God's into, you know, a sort of a demigod that represented rationality.

    And here's a quote from seduced rosevears good friend, where he basically makes the point that the, if you like, the manifestation of the mechanical worldview, manifests in society through law. So that's the place that we should be thinking carefully about mechanics. And that's where we could perhaps mechanize things the way that Newton had. But of course, this didn't work. And this is a rather famous painting from the chera, which was the consequence of a lot of this mechanization. And this dehumanization, where anyone remotely suspected of having sympathies with dissidents, was given and begin to change is one of the beautiful classical machines invented by a physician at the very beginning of the revolution. So even in that exalted period, where there was this connected belief that through reason we could transcend superstition, the consequences were pretty foul. And I would argue that many courage, efforts that might be unaware of the history, but regardless, it doesn't really matter, are actually continuations of the fundamental enlightenment idea that human behavior is so complex, the madness of the multitude is Silla controllable, in the very best thing we can do is replace minds with give the genes in this case software. And so this idea somehow that humans are fundamentally limited in their cognition, but machines don't have to be in the same way. And if we just outsource, decision making, for example, legal decision making as we do, everything would be much better. This is obviously opinion and opinion I don't share. And but I want to give you another history, which I think is that splintage interest in this group. In the old text of this particular alternative is this book that was published in 1901. So the life of the bees. It's a very extraordinarily weird man. He was a Flemish playwright, poet naturalist plagiarist. But in this book, he actually showed that there had been this long term interest in attending society to the behavior of collectives, most notably social insects. As this quote from the beginning, a stranger will creature be that is that lived in a society under complicated laws, and executed prodigious leaves in the darkness attracted the notice and struggle catch Aveiro, Guinea, etc. This sense of energy, I think, is a sense of energy that's present in the modern world in a slightly different community, actually, those of us interested in how we manage Demi autonomous connected that are noisy unreliable, but nevertheless, we would like ideally in some aggregate produce a particular function of interest to us. And what I want to do now having given you a little bit about history, which I think is very interesting, and actually Sandhya Jerry, for us, talk a little bit about the work that my colleagues and I have done analyzing as rigorously as can the history of evolved connectives and extracted q insights from those munck be of interest to engineers who are designing collective to the row I don't necessarily believe that has to be done. It's an interesting discussion in itself. But I think it makes sense to be aware of this alternative well beyond simple metaphors published in popular science books. So where we've worked, is remounted a multiplicity of scales from cells, brains organisms to to societies, with an interest in the mappings between these levels, if you think about universities, it's quite interesting. Universities have basically structured according to principles of spacial scale on, there are cell biology departments, there are neuroscience departments, there are behavior and psychology departments, their economics departments. It's quite interesting, right? That spatial scale became the principal axis and on which we organize epistemology. Could it be entirely matching, but interesting, right? You have departments, the nanosecond department, year department century, you would have been a completely different principal organization, every bit as fundamental, legitimate and spacious, get

    the wrong thing organize our thoughts at particular scales? As I said, I'm interested in organizing our thoughts as mappings between scales. And these are these mappings of collective dynamics, right? How do cells meet brains? How do brings me organisms, how to organisms constitute societies to different kinds of science, actually, and it transcends, in some cases, the particular constraints of material that applied every spatial scale. And out of that work, new principles emerge. And that's what we do. And I also go through some of these, of course, the next few minutes, but is a lot One could say here, notions of endogenous cause screening, circuit design, but in this evolved sense, not necessarily engineer, fundamental coding principles are connected. The statistical physics have evolved systems in particular issues, phase transitions, and my particular passion, collective intelligence. And as important, which I will not talk about today collective and individual stupidity, which is the great elephant in the room voivod science. So let me just start with molecular levels. And I'm just going to rush through these just to give you sort of a sense of what this is all about, and everything, have a chat. All of this work is collaborative, all of this work. In other words, is connective science stopped being individual, probably go on as soon as science was invented. But as we discovered, in the case of Voltaire, there are all sorts of ideological and social reasons why we would like science and look as if it was a solitary procedure. So with two I think the Voltaire Chateau layli insight is is is significant. So just to quickly say, it was discovered in the 80s, actually, that very simple. Chemical kinetic district descriptions of molecular directions could produce amplifiers and switches. In other words, we can build transistor make elements at the molecular level out of evolved constituents. And if you took those evolved constituents, so in other words, we can build essentially the analog of the transistor. If you added feedback to those constituents, we could build memories. And those memories are actually quite cool. I won't go into them today. But I was always very interested in this memory, which is certainly familiar to some of us are called mercury display delay line, in which actually memory is stored in waves in mercury. And as you recognize tach from the UNIVAC, we have two pizza netic crystals that propagate, essentially a wave through Mercury, which is and in fact, these actually have mathematics very similar mathematical properties. And we started studying these kinds of systems as a de to understanding how cells could be doing computing. I mean, this is all basically statistical Physics for those who have an interest in these things, analyzing issues of phase transitions, excursions and fluctuations and behavior of critical points and new systems. And I just summarized that, as I said, I'd be very Cavalier, but I can give you this references for those out of that work, and another what like, we discovered that noise is absolutely critical to the to the function of these systems. noise is not the thing that you want to eliminate, as in the classical case, but actually that you want to increase early and decrease late. So, noise turns out to be the critical organizing principle for exploration. And once a particular stage has been realized, or reached, collective dynamics kick in, up to stabilize the system around a critical point where you can expect various properties like history system memory. The point I want to make here is that I'm in connected dynamical systems Did you get stick turistic dynamical systems which is what everything is in the world that you study in the natural world. noise and connected dynamics are critical. Stand. And we shouldn't be thinking the way that we normally do in classical terms, because of course, as everyone here knows, the classic limit is the privilege of engineering. It's not, it's not the reality natural world. So now we can jump to another level, which is up to the leaven cells. So that was molecules in cells, cells and brains. So that the great mystery of the brain, if you like, is twofold. One, How noisy elements allow us to do things like group theory and category to diffuse and clean and so pure, there's no noise in them. deduction seems some classical and deterministic. And yet the elements that constitute the system that does the calculation is a mess, you know? And so how does that work? The other question, of course, so interesting about brains is he, we live in a world with just under 8 billion people, that's a disaster. There's just no way of coordinating humans at that scale, it seems to be the case. And yet brains have the human brains about 86 billion neurons that somehow collaborate and cooperate to produce thought, the kind that allows this conversation to take place. How does that work? Because that's the whole field in neuroscience and cognitive science, psychology. But I just want to give you one little insight, this is some work that we give to some colleagues at Stanford, either the monkeys abused screen, there are dots moving along on the screen, and you have to decide are the dots on average moving right or left? This is actually the checks that will give you the chance. So here's what the monkeys do. Each one of these panels has dots, you have to now tell me on average, this would work maybe in a regular lecture on single itself. Kind of rubbish. That there you go on that one. If you had to decide shout out using the microphone. Are they going on average left to right, what would you say? No one does. left? chest. Okay. Yeah, I think we

    ran. Okay, even. What about this one? Left, left, top right, is moving left, obviously. Yes, and abortion hadn't left is moving left. Bottom Right, is moving. Right. Right, it's getting easier, right. So um, so these are actually whoever said even was correct on the top left, as likely to random walk left is right. And as this probability is zero when you get laminar flow in one or the other direction. And the point is, of course, it takes longer to arrive at a answer, the closer that probability is to a half, right? And the closer to 01, that you viewed as David, I think he was dead on, he would fairly quickly reach the conclusions that was left. So this is the chance that they given. These are monkeys, where there is a microarray that's inserted in the central lateral parietal cortex from which we take readings. And we can use that data to decide whether or not what they're seeing is moving on average, left to right, and how they come to that decision. And there's a lot to say about it. But essentially, what we observe in this kind of study is two distinct phases. There's a phase in which all the cells in this part of the brain are independently something wild. And then there's a phase where they all collectively come up with a decision. So there's a independent phase and a correlated phase. And essentially, what you're seeing is cells that are sparsely connected, in this particular part of the brain, that benefit chain from independence. But at a certain point during the task, as I think when many of you shouted out a conclusion that you've had reached, all of those cells have to come to some agreement otherwise every cell would be shouting a different answer at the same time and you are useful for coordinating society. And this is against correlative Bazan. This principle is called we call it the geo coding principle. And it's actually ubiquitous, it's now been discovered in beads to my brains. And you can sort of see information theoretically, why this makes sense. You kind of want independence right? When the data is noisy. The once you reach some kind of decision, you'd like to aggregate and collate your findings. So this is another proper job to say, of stochastic connected dynamical systems that your information processing, that they exploit processing, that they exploit transmitted stage, and they maximize the density of connectivity in order to reach a conclusion in this sort of consensus state. So there's this gap analogy of strongly bound and strongly and loosely bound is actually a critical feature of the way the brain works. And ignore the literature one often hears about, you know, all these distributed agents autonomous, no, you don't, you want an autonomous one face, but not. So let's just jump societies. Now this is work that's been going on for many years, I'm going to give you a sense of what this work is done, which we work on, on non human primates. And I want to give you a sense of two groups. One group that we've worked on is the picture mcag. This is a new word monkey, as the one at the bottom of the picture, are these behavioral studies. And the other group we work on are these spider monkeys. And then you world sorry, pig channel work, they diverged on the order of about 35 million years ago. So just to put that in perspective, we diverged from chimps about six minutes. And we study the pigtails, because we're interested in conflict. And we study the spider monkeys, because we're interested in cooperation. And just to get your basic sense of what we do, we observe over 12 hours a day interaction between every individual in a captive colony, we can determine exactly who is in a fight, when, who fights with whom we can construct a symbolic time series. And from that symbolic transcripts, we can actually extract inductively a Bayes net class because it has logical elements, like and or not gates, and construct a strategic circuit of the entire science. And again, I'm going fast, but um, and so essentially, what you're looking at here is a circuit where each vertex in that graph is individual. And an arrow is an interaction between the example in a conflict. And we can then use these circuits, which are in some sense, how would at summary statistics of the society to predict behavior into the future sort of replaces them, right? It's an encoding of them.

    And for example, we can predict things like the duration of fate whose interface when if we remove someone, will the fate stop, etc, we can do exactly the same thing. And uncooperative regime is a beautiful looking spider monkey. They foraging groups, and they form little sub groups that each explore different parts of the rain forest. And when it's wet, there should be a certain subgroup distribution. And when it's dry, another kind of sub distribution, just as with the big channels, we can reconstruct these circuits in the society and then use that circuit to predict the behavior of the troop, how individuals moved from they consult with, etc. And, again, this is kind of interesting. So these social circuits, which is not typically how we think of a society where you think about the social network, which is much simpler concept. But actually, they can be used to predict actually how a society will evolve in time. And what are the dominant causal vertices or sets of causal causal vertices in those logical graphs? And so we can actually think of a society as performing as a stochastic computation along the lines that I described, that is, when should I find to be one? Or who should I cooperate with in order to extract resources during the wet season? Finally, an unpaid quickly, this is work that I've been doing actually last year, and confinement because I went back to my old Rubik's cubes. This should be published Tokyo next year. I'm working on essentially highly competitive Rubik's Cube players. Just remind you that the Rubik's Cube was invented for many permutations, the maximum number of permutations required to reach a solution from the least solved initial condition is God number that to the n three to 20. They're solved using algorithms or rules. They you this community uses algorithms in a special way, right? It simply means as an input output mapping from the configuration of these little faces, to the optimal move required to move towards a solution. And there's not one answer, but this is actually a really fascinating community to study. So essentially, I've been studying these communities as naturalists. But of course, we have this great advantage here. Because the cube is actually just any graph. So it's a mathematical object, which we can treat as such that we can embed that mathematical object in the natural history of competitive Rubik's cubes. A very interesting thing you have on the stage of course, which is how much better they'd got over the course of time, both sated and blindfolded. That's actually fascinating, incidentally, itself, just how you saw you blindfolded. But the nice thing is here, we also know, the social network of cube solvers. So what you're looking at here is a graph. And each of those vertices is a cue, B denotes blindfolded. And when there's an edge connecting two of these cubes, it means that a competitor has broken a record in both cubes. that in spite they compete on more than one cube, and do analysis is not really short circuiting everything on the learning rates of cubes, you find is for the psijic cube, the learning rate goes down, the higher the linear dimension of the cube, right, because it's a two by three by three cube for Vekselberg 45 by five by five, up to the seven by seven by seven, and other variants. But if you look at the blindfolded cue, actually, the learning rate is maximized. But the four by four by four, not at the three by three by three, or the five by five by five number words. There's this interesting peculiarity in the brain for the case. And it turns out that that can be predicted by the structure of the competitive network. Right, so this is not a cognitive network. But the thing on the right is a cognitive result. And the basic point here is that the community of Rubik's Cube solvers have discovered the shoe which maximizes the efficiency of learning, and the competitions they should participate in so as to achieve that result.

    So this is nice connection between the collective and the individual cognition. And this is really, I would argue, the canonical property of the human being, which is that the social informational networks in which we live in that case through its huge network, in our case, everything else, actually structures, individual cognition, so as to increase something in this case, like a learning. Right, right. And their construction, they go both directions, right? That the social structures we build come out of our cognitive biases. You would say, yes, that's what a school is. You should, then of course, we know it isn't. And that's part of the problem. And a school is, in some sense, a collective social structure, or framework or a scaffold, in which we hang individual cognitive elements in order to maximize something like the learning rate. But I claim that we're not designing according to the fundamental principles. And studying things like this allow us to get a little bit of a peephole into how that's done in communities with a much more restricted task. So just find fun slide. These systems are very interesting. They're nothing like engineered circuits. They have very high autonomy, many of the elements are in competition. There is no god that puts them together. Certainly I don't believe there is. They're very noisy and noise is important. They show another conversation, very interesting. rebuses properties, very interesting energy efficiency properties. They're highly reconfigurable. And they can be the basis of building larger circuits and scalable and evolve. So they have nice properties that I think the Hoover's to be aware of, not that they're the answer to to the engineering of the future, but I think the useful I think in discipline your vote. final final mark, this is a quote from our goudy relation was architecture and you wrote this beautiful thing when you said, You recognize the right the Sagrada Familia on the left termite mound. Anything created by human beings is already in the great book nature. That is not true. Okay. But what might be true is design principles applied by human beings already in great book nature. And that is Dewey on expansive conception of by minisas. I think there's a lot to learn from the way that evolution has created coherent collectives. So with that, I'm at the end, thank you very much. Oh,

    fantastic. Thank you so much. I will stop sharing the screen if you don't mind. That was quite, quite a lot to process. Yes, but we rely on our collective intelligence to do so. So first one up, and before I have a few questions, just the way dual coding works, though, against the Bayesian brain hypothesis, it seems if the entire system collapses to a decision, then it gets rate of uncertainty does not Bayesian, rather, it looks like the brain is selecting the most likely model of the world and then running with that. What do you

    think? Well, we know that the brain is busy. And so I always find that constraint this weird misunderstanding of mathematics, I mean, every mathematical tools that we use to comprehend reality, and I always felt we were projected back onto reality, the framework we sort of endogenous. So I'm not to say that I'm not really sympathetic to that way of reasoning. So on the other hand, you say is a is a is Bayesian statistics are useful tool, absolutely super useful. And we don't have to go the full hegemonic lineage that the brain is, is Reverend bathes in every case. But But you make a good point. Um, it's a much more complicated dynamic, it's in it's it's clearly best, approximately Beijing, probably not Beijing, technically speaking. It's just a more complex computation. I mean, would you describe, for example, Microsoft Word is Beijing. And it would be sort of weird thing to say, right? It's somehow the brain is more than just a very simple statistical inference device. It's actually running quite complicated logical rules, in its circuits. And I, it's a long ago, I just find that too weak, a statistical framework to understand the full richness of what something like the brains doing.

    Yeah, sorry about the microphone issues there. I was using my Linux computer for the first time. And I apologize, but yeah, that's not that that was my impression. And basically, I, I thought it was very interesting, it looks more like sort of a winner takes all dynamic, where the, the model that's most consistent with the data sort of wins out. Rather than having a lot of different credences assigned to many different models. always interesting. I

    shut down that that's a good question. Um, there is actually a voting mechanism here. So winner takes all would suggest, in the simplest case, like natural inhibition, that the highest amplitude unit would inhibit all of its adjacent or connected unit, it's been complicated in that, it there is actually a pooling mechanism. And by which they reach consensus, so it's in neuroscience, by the way, there is this big divide between what sometimes called ensemble selection of the G, and the sort of grandmother hypothesis of grandmother neuron, and what, what are they ensembles activities, when many, many cells contribute tiny fractions of the children. And then this sort of grandmother neuron is where one cell somehow can encode a very complicated perceptual set of features in the wild. And what we observed that you both it's not one or the other is time dependent. You move between them, right? You have a redundant phase, which is, right, where everyone's who knows everything, and then there's kind of a fractional phase where everyone knows.

    Right, Next up, we have Mr. X.

    Yeah, I have a first question is where do you get a chalkboard these days?

    Haha, you steal them from my sequel colleagues.

    The way you describe the way the molecular and cellular systems use noise with a lot at the beginning, he was at the end sounds a lot like simulated annealing computation.

    It That's right. Absolutely. And the it's, it goes beyond annealing. But but because, but it is, but I think it's a totally reasonable metaphor. The, the idea that there is this exploratory phase, but it's also in the theory of phase transitions. There is, as you know, whether you approach a critical point, the volatility of the system, or susceptibility goes up, as do long range correlations. And so it's there is the kneeling metaphor, but it I think it goes somewhat beyond that. Because for us there also evidence of this hidden new criticality, which is sometimes called the criticality hypothesis. And just to give a bit more background on that there is this peculiar fact, right, that a lot of systems that we measure, in the natural world seem to live by a critical point. And if you were designing a system, most people would say, that's probably not a very good idea, right? Because I don't The last thing I want is loads of noise and fluctuating back and forth. So you try to push away from the critical to damp the fluctuations. But if you look at it like john hopfield, early work on in hopfield networks, all of those are memory capacity calculations right? are close to critical points, where actually you get the most degenerate states. And at the two minutes, either the system freezes in to one stage or another. So it's it. The memory capacity, I guess, is another important component of noise. And it's

    lovely. Next up, we have Chris,

    I, just to do the introduction thing that Allison was asking for. I'm a software guy currently working on blockchain stuff, his theory and prediction markets and privacy and a bunch of other things. I wanted to ask about the cube solvers. You talked about, like to watch some cube solvers, both visual and masked a couple of years ago, and it's kind of interesting to talk about what they do on you sounds like you've delved into the sociology of the learning, it seems like during the competition, you can see who's fast, but it's hard to tell why they're fast. What what opportunities do they have for talking about their algorithms? How do they share their algorithms and learn from one another?

    Yeah, oh, yeah. So there's, I mean, it's just wonderfully rich, as well. So just to remind everyone else in the room, that with the blindfold q competition, what you do is you get to inspect the queue. And then once you're ready, up to a reasonable limit, you then are blindfolded and you start the clock. And it's not the you can blindfolded from the very beginning when the sheet is shuffled to the magic task. So some a few things to say, that's really wonderful things that are happened here. The way in which you can you actually commit to keep your memories using the circle method of low sigh or memory palaces that folks are interested in the early work to Geodon or Bruno. That was the great jasic text he wrote on on memory palaces. And so you have to commit the entire queue to memory right before you're mindful. And every time you can you deface you have to store that new foot state of view in your memory, right? That is not true in sad Chase. And one of the consequences of that, incidentally, is the number of algorithms used to solve the blindfolded Chu is on the order of five. But the number of algorithms used to solve the sake of tunes on the order of 70. And it's, it's a very interesting memory trade off, right? What can you hold in working memory if I'm going to so hold the cubing working memory issue is how not holding all those algorithms and working memory at the same time. So. So that's something that's really interesting. This is not well known. And it has interesting implications. These communities are extraordinary.

    Can I ask, do they, when they're doing it blindfolded? Do they actually Okay, you said they only use fine algorithms. So there, they are sure. They're storing a picture of the shape and not the path they plan to take?

    Well, that's an interesting question. That's a very interesting question. Nobody, you're not storing a power, because of course you're studying

    it. If a chess master is playing blindfolded chess, on, a lot of what they're doing is collapsing the representation of the board down to simpler parts of it.

    Now it's exactly the same thing. He actually that's, that's the key. So mathematically, the cube is massively over determined. There is the low rank representation of a chessboard. And there's no round representation in the queue. And when Chase and Simon did those early experiments, where they created board configurations that will legal but silly, but they would never be reached by any competent player. Masters didn't know what to do. There's like one earth to upgrade, because it didn't have that nice. Them rank approximation.

    It's like professional baseball players. playing against a softball pitcher, they have no idea how to hit the ball. They've got to work. They don't know when it's going to arrive.

    You know, it's a very easy question for this group. I've also thought about in relation to AlphaGo and alpha zero, which is I do think it's the full answer, but it's an interesting possibility, which is part of what made it so weird. felicito is that he was he was like you said such a weirdo opponent. He was playing in such an odd way. That that might partly explain why he underperformed his what an Earth is his band thing doing right? So I think anyway, so just to get to your first question. Yes, then it's a very cooperatives and generous community. Lots of sharing going on all the time, on blogs. In fact, you can find them without So this is how I saw this. Whew, that competition. This is the trick that I enjoyed. So, yeah, and the social network, the learning networks, incredibly dense, and they all depend on it.

    It reminds me a little bit there's a Mickey Mouse and Goofy cartoon, I think in which they're being attacked by super intelligent alien civilization. And then Mickey tries all different smart things. And then goofy goes, and this is like, and they would be

    Yes, they did. The in karate, it's called Drunken Master. Rage is fundamentally the play. He says not to be a predictable adversary.

    Yeah. Well, okay, great. Next up, we have Mike D.

    All right. Yes. Nice morning, if you've looked at neuron, morphic computing, and if this has any applications to it,

    I have not. And I know that's not a satisfactory answer. But it's a true answer. Do you have some ideas?

    Yeah, no, Omarosa gets supposed to be a lot more like how the brain is actually constructed. The chips are made with like neurons implemented as devices can use, like probabilistic or analog levels, signal levels, instead of just binary? So yeah, given the structure look very well suited to the kind of work you're doing.

    Yeah, yes. I mean, there is this very interesting, open question that, and which is, what does it mean to do engineering in a system of agents that are in competition, or onii? Chan partial information? And or other, you know, a belligerent, you know, and that

    that, to me, is

    the interesting biological limit. One thing is to say, right, we're going to use noise. And I think he was Alan said, rightly, you know, engineers, and understand that you have a complicated frustrated system, or multiple peaks or degenerate minima, injects noise into it. And this is this goes beyond that. These are units that actually, they're in competition. I mean, they. They're almost like free living creatures in your head. And, and that just introduces this other dimension, which I think is fascinating. That turns out, and this is something that might be the key to robustness. Right? It might not make for great calculations. But it probably makes for robust calculators, because once you eliminate components, everyone else is saying, excellent. I can jump in and do it now. And so as opposed to being in a state of despair, we would be with an engineer system, they're all rather happy that someone is fallen by the wayside.

    And I have a follow up on this actually. And so not only in your talk, but also, you know, I think the entire sentence that you've really progressed on the premise that our understanding of natural intelligence is a little bit too shallow, really, if we want to use it as an analogy, or if we want to Ford, much to, to AI systems. And so could you maybe kind of pinpoint a few of the most common myths that people have, I guess, about intelligence? And how, you know, applying a more deeper understanding actually leads to better systems, like, do you have any concrete examples, almost of someone who's really using those, you know, the more deeper understanding and actually producing a system that outperforms, let's say, a system that relies on a more shallow understanding of intelligence, like other people applying this stuff already, I'll be so keen to know. Well, yeah, I

    mean, it's, I don't want to say as I said, at the beginning, I had that little caveat, which i think it's it's, it's true, and I and I, which is not, I don't know how much you need to know, right about the systems to build great designs. I mean, it is quite interesting, right? That the current preference for deep learning like structures grew out of people like MacArthur and Pitts, and and, you know, posthumously published, john von Neumann will come to computer and rain that have green light, just distributed structure. So I don't know how to disentangle the history of thinking about software from an awareness of the way the biological world works. And I've encountered a number of these people and I'm always impressed by how much they do know, right. In other words, it may not be written in the journal articles, but it's there in background on but you know, there are very clear examples. I mean, I think, other than obviously in combinational, neural nets and in teaching so activity, influenced by some structure of that now bears that kind of thing. And then there's this whole area of genetic algorithms, of course, we worked on here, john Holland in particular, and Ellie Mitchell's. And then the cybersecurity issues with new systems for operating systems. Stephanie Forrester. So there's a lot of work like that is very biologically inspired. And, and they're using them. I mean, this is not, you know, it's an emerging thing. And but it's still quite simple. And I've gone through a whole bunch of systems that are very neglected, I think, from the point of view of engineering, maybe for good reasons, right. But I think it makes sense to be aware of, they could come in handy, especially in a world where issues of scale become really important, in particularly relation to energy. Because the energy efficiency of human beings is the thing that we need to be, you know, a light bulb, essentially, that we should be very aware. And that, of course, if you pick up the newspapers every day in relation to blockchain, rightly or wrongly, that's where the debaters in society these days, this seems like a very energetically inefficient system. You know, we're going to move, you know, to proof of steak, because that's going to help. So it is interesting that a lot of the debate is centering around energy, which is a problem that life had to solve. And so perhaps that's the Nexus that would be of interest to us.

    At Yeah, I guess, energy is really the final frontier, and most of these things, but and I had one particular question, and could you maybe elaborate, and how, maybe perhaps the current understanding of, for example, our immune system, in in the way that we use it as an analogy for computer security may not be sufficient? Or do you have any, you know, more thoughts on this particular angle?

    I mean, I would redirect this to people, you know, in Boston with me, I didn't Stephanie forest, and Melanie Moses thought very carefully about both energy and security. And I am just an audience member in, in those conversations. And I do want to get to Alan's question about blindfold chess on when you say, I didn't get the impression that my dude was no rates of interesting. The way to think about this is you whether consciously or unconsciously thinking about constellations, pieces on the board. And it's those high water correlations that I think I'm referring to. But I suspect you were thinking that's, that's the position of the bishop in relation to the position of the night that's, you know, there was it wasn't just atomic level piece representation, that some higher order representation of the relative position of pieces on board, that's what I mean.

    Yeah. And, but, but on, when I went on analyzing a lot, when I was analyzing the way to play inside of chess, I have to do the same representation once I get more than one or two moves in. That's right. Right. And so it's just the starting move on at that place instead of two moves. But I didn't think it was a qualitative difference.

    No, that's interesting. I that's an interesting point. I'm in the case of it's funny because my my grandfather was a chess master and I used to play him blindfolded in neat sage and, you know, wait the carpet with me. And I think that you know, in the case of the cube I know for a fact that you news use fewer algorithms to solve blindfolded raged on in the case of chess and and I do think that holds I think that great chess players don't have fewer rules in their head when they're playing blindfolded or sighted and so that trade off I'm not sure

    the exists wrong. I wasn't I wasn't great chess player, when it was just something we did over one sheet High School. Neither one of us had we played without a ball.

    Yes, yeah.

    But But the idea that that once you couple moves in your house to picture a brand new board any made the blindfold see quite natural.

    Right? Yeah. No, it's it's, it is interesting. I mean, I just to jump laterally, you know, to what extent to these outsource, you know, the chessboard as a blackboard rage in other words, a very convenient, efficient encoding for working memory to use, so that you can actually do deductive reasoning, rather than have to spend all of your energies on memory. And that, for example, if you take the abacus, which is another system that I studied quite a bit, that's very interesting, right? Because most Abacus users, most of whom about nine, by the way, who could start learning they're drilled at schools, which are very severe, actually have a mental level and in that particular instance, and it's been Study using fMRI is quite interesting, you know, how is the abacus encoded in the brain. And that's a case where at a certain point, the marginal cost of storing the abacus is is is zero. And, and so it could be that with expertise, the chessboard actually is just marginally not that significant.

    Calder Lehner attempted to do well, just playing the way the human place rather than brute force. And what he realized was the chess players, the Masters, break the board up into conceptual chunks, and give weights to certain patterns. And so it's more of a pattern thing than an entire brute force analysis thing, eventually to go to the individual moves. But it's more useful than an analysis.

    That is the sort of I think that's the general. One of the few, I think, General truisms of human reasoning that applies across games and science and mathematics. Right, which is that that's sort of what we do in all of our domains of inquire. And as many people pointed out, part of the problem with early chess computing is it didn't I mean, that's a harder problem, and didn't have to merge. I think there's no doubt that when we're solving problems, I know for myself, technical problems, we are working with these cash outlay entities.

    And I have a follow up question, I guess, to piggybacking on the analogy to games. So one of the article that I really loved was playing golf with Darwin that he published. And it kind of, I think, literally, you say, take the premise that Charles Darwin was very likely the first person to have understood nature in terms of a game played across deep time. And then you use go as such a game analogy of evolution, and maybe, you know, just to go search from chest to go, could you maybe elaborate, a little bit more on goal for enriching our evolution, our understanding of evolution as a game from the past? Find a few minutes?

    Yeah, I mean, that the, the basic point there is that I didn't, you know, as someone who actually works in the evolution of these things, and so frustrated by the cartoon of evolution, we get away from, you know, I would perjure myself, but from people like Richard Dawkins and others, where we think of it in terms of very simple surviving the serious damage, so we can just write down this one dimensional system and the density dependence, and it's Basie. And actually, it's a Bayesian inference, they're actually mathematically equivalent, the replicator equation, and basically, there's so much more evolution than just, oh, I'm a little richer, and therefore I produce on average, slightly more offspring. And then in the course of time they fix which is the sort of, it's it's so tedious. And, and what we've discovered if that were true, there would be no complex life, it would be impossible, because we have so many valleys, steep valleys to cross. And this is recognized, by the way in 19, teens you know, by civil rights in order to introduce diffusion into the framework. So part of the motivation about that article was to say, we've been thinking about evolution as if it were tic tac toe, whereas we should be thinking about as if it will go and there are very interesting representational problems that evolution has to solve, particularly in relation to the genetic architecture that are much closer in spirit to some go concepts, these discharge like things actually, and concepts that Sanjay and others that are ending rage than Tic Tac Toe laid strategy and I so it was me that was an exhaust spiration article that you get real with the way evolution actually works, and genetic architecture, developmental mechanics, they matter. And if the standard cartoon that have died isn't that we teach in high schools were true, we would still be in the primordial sludge. So yeah, for me go the the value of go was it makes these cognitive psychological issues Ostalgie things front and center, because that's how you describe the game when you play it. And these are very rich concepts on you do not describe though at the micro scale. This piece can go that that's it's just not interesting when you're talking to another human.

    Alright, I posted the article again, you're in the chat. It's probably too much of a mouthful to actually go through here in a minute. But I want to take perhaps the final minute to just ask you, you know, what's new in your work? Or like, what is the Santa Fe Institute focus on now? And in terms of potential programs? That could be interesting? Yeah. Like, what's, what's next?

    Well, there's a lot, actually, it's a very exciting time. We have a big new project on intelligence and diverse intelligences. So actually, throughout the last year, there are many rather make this one, actually. But with people more squarely focused, I think on natural, but also artificial. So if anyone's interested that let us know because we're always looking for interesting people to join that conversation. is just much to say, but my particular work over the last six months has been trying to get this book out, which is a big, fat thing. And it's, it was Santa Fe Institute's response to COVID, which was our community, really mobilized in 2019 can be a little frustrated, and asking ourselves, how can I help you know, what is pointed with this? If I can't make a contribution to society? Much of what we contributed was useless and no, he's in a negative sense, perhaps. But I think it was a real reckoning. And so if you asked me what the Central Institute now, I think we will ask, everyone on this call is asking, which is how to prevent a desktop from taking of the white house again, and anon nickel? Where? And how do you make a society that's it's robust, and a tiny penetration of initial virus? Six genes can take down the world? How is that possible? So this little book, which is a community effort to try and address this question from a complexity lens with meaning? How does epidemiology connect to issues of nutrition and economic markets and socio political systems, and one of the consequences then becoming entangled in unpredictable ways? So I think that's where my card has been. That is so late in life.

    Okay, great. There was one final question. Have you mentioned a workshop on what is the mathematics of intelligences?

    Did you see a I'm good? Yeah, that's, that's funny. That's not what I mentioned. But I know the faces. And that's cheek and fostering others. That looks great. And I think I'll go to it, but I'm not the organizer. But that that was very interesting. That that comes out of disty. That comes out of the diverse intelligence Institute on what was worship, I imagine they have been internal by and large discussions in our community. on particular themes, for example, the biophysics of the neuron or principles, the collective intelligence, or on and on, and that I, if what I can do is it's a larger meeting takes place on send it to Allison, and perhaps you could circulate it.

    I will disperse it. Okay. Hey, we've had you two minutes all week, overtime. Thank you so, so much. I think we've actually covered a treasure trove of all kinds of things. And you know, hoping that one day, you know, we may actually do the conversation. Thank you. I think you've really given us a few really interesting Nuggets to chew on as we kind of go forth and build the systems that we're building. So thank you very, very much. And I'm hoping wasn't the last time that we welcome you to the group. We wish you much success with your work, and I seek everyone else for our next video. So thank you, David. It was a pleasure. I

    much appreciated