History of Science and Technology Q&A (July 19, 2023)

Hello, everyone. Welcome to another episode of Q and A about history of science and technology. I'm back after a gap. The gap was associated with our annual summer school, our 21st summer school, and our high school summer research program. Very interesting. For me, I think we had 66 students at our summer school grown ups and 61 at our high school program. So I got to try and come up with whatever it is, 130 or so projects and got to meet all kinds of interesting people and learn about all sorts of things. But now we're back and we'll be continuing with these live streams with a few gaps over the course of the summer here. All right, let's see. Looking at some questions here, let's see. Judy asks, can you discuss a bit of your personal history with AI? When did you first become interested in the idea? That is an interesting question, actually. When I was a kid. So I grew up in kind of the 1960s and 1970s. And I certainly would look at books about the future, so to speak. This must have been times like 1966, seven, eight. I was six, seven, eight years old. There were all these books about the future and they had various kinds of things in them. I might say, by the way, that in addition to books about the future, there was also a little bit of television about the future. So for instance, there was Dr. Who, which I think is still running, but it was a thing back then. And there were visions of AI that already existed in those television programs back at that time. So there were the Daleks who were, I think their backstory later developed differently, but they were these kinds of sort of somewhat conical, round headed shaped robotic things that would wander around with a single kind of tentacle hanging out, so to speak, at the front that were mostly rather nasty creatures, but they were kind of a version of sort of roboticism and AI and so on. And then there was, I think there were things called the Cybermen who were kind of playing off the cybernetics kind of theme. Also in Doctor who, who had some more complicated sort of human like form and story, which I now forget. But so, you know, one part of sort of the, the ambient vision of AI was these kind of mostly malevolent kind of robotic critters. Then I remember 1968 when the movie 2001 A Space Odyssey came out, seeing that that had a sort of a version of an AI in the form of the HAL computer in that movie. And that was another kind of introduction to this sort of ambient view of kind of AI as kind of a Perhaps in many cases malevolent kind of influence on things. I guess around that time also the Star Trek television series came out. I do remember that, okay, silly personal story. There was a sort of prime time of American kids television was like 6pm on Saturdays or something. And there were a series of different programs that existed at that time. And there was, you know, you would look them up in this actual physical publication called Radio Times. I think the US Version is TV Guide. That was a printed list of what programs would appear at what time. And there had been a series of really, in my opinion, quite unwatchable kinds of television programs that had been in that particular slot on Saturdays and so on. And so then there appeared this, this new thing that was listed as Star Trek. And for the first many months I assumed that this was another one of these to me unwatchable programs that had something to do with horse trekking. And it was some kind of, you know, celebrity horse trekking kind of thing. And I completely ignored it. And I don't know how I came to actually watch Star Trek for the first time, but. And I found it sort of charmingly American. I was growing up in England. In fact, several years later I had occasion to write one of these kind of essays about, for some, I don't know, English literature thing. And I, I ended up writing about two pieces of American television. One was Star Trek and the other was a, a police drama called Kojak. And I had noticed my. My big discovery was that there were both of these had sort of various allusions to the American condition, so to speak, with the. I think my giveaway was some episode of, of Kojak where at the end of the, the episode there's some disaster has happened and some building is burning in the background. And this, you know, detective character turns the camera and says I think something like the mass of men lead lives of quiet desperation. Which didn't seem like it really fitted in with the usual way this character talked. So it was kind of a giveaway that, that had to be some kind of literary allusion, so to speak. So I went and looked it up and found out it was. I know it's Judith Thoreau Emerson, I'm not sure, one of those kinds of characters who used to live 100 years earlier in the area that I live around, Concord, Massachusetts. But in any case that was sort of a giveaway. And then I kind of realized that in Star Trek there were lots of sort of American frontiers, kinds of metaphors and so on. But in any case, Star Trek featured an AI Computer. And that again was part of the sort of ambient experience of kind of what AI would be like seen from the late 1960s. And then there were books about the future. I do remember one that had this notion of teaching machines which was some kind of AI based thing, which would be the future of education. And there was kind of an illustration of some kind of sort of screen based computer thing. Any case, I didn't actually, you know, computers were not something that one saw in real life back in those days. And so sort of the only image of what computers were like was the occasional kind of thing in a television program or something and. Or in something, or in a vision of the future like 2001. And when they did appear, and I'm not sure when they first started showing up, but when computers did start showing up and kind of as background props in television shows and so on, it would usually be the twitching nine track tapes that would be the most visible. So in those days one was dealing with mainframe computers like the IBM360. And the way that data tended to be stored, some of it was stored on rotating disks that were sort of fixed for the computer. But quite often data was stored on magnetic tape. And so it was a really common thing to see data being sort of found. It was like kind of databases being stored out on tape. And the tapes would be twitching around trying to seek the appropriate pieces of data. So that was the kind of most common image of a computer. I think I first saw a computer in 1970 when I was like 10 years old. And it was kind of a mainframe computer from a distance, so to speak. And it didn't seem at all like anything AI ish in those days. I first used a computer in 1972, a big kind of desk sized computer that again didn't really seem very AI ish. And I used it to do computations eventually, well, various kinds of attempted physics simulations and so on. But it was a computer program with paper tape and things like this and rather far away from the kind of HAL like version of a computer in 2001. So I guess that I didn't really pay much attention in the 1970s. There wasn't a lot that I remember of kind of the sort of AI wasn't for me at least a big thing in those days. There were things where people would talk about it was still the case. Well, in the 1960s it had still been the case that sort of spacecraft and space were kind of a big frontier of technology. And I do remember that I was very Taken with a computer that was supposed to be on a deep space probe, maybe going to the outer planets. That was a self testing and repairing computer. And I was certainly very curious. This must have been when I was, I don't know, eight, nine years old. I was very curious how such a self testing and repairing computer could possibly work. And I kind of wondered, you know, what did that really mean and how did it, how was it in a position to kind of test and repair itself, so to speak? I think in actuality it was something where it had kind of three way redundancy and things would kind of vote on the outcome and so on. And maybe it had some capability to get new code that it could load for itself. But anyway, I think then my own kind of next exposure to these kinds of things was, well, I started using computers to do kind of mathematical computation starting in earnest around 1976. And I didn't think of that as being anything particularly AI ish. It was just, okay, I can get a computer to do algebra, I can get a computer to do calculus. I think that people had. There was some part of the work that had been done along those lines, particularly coming out of mit, that had been very much themed on the. This is going to be part of the story of artificial intelligence. But to me this was mostly just a practical thing about making a computer do the math. I didn't want to do by hand myself. But that whole set of things caused me in 1979 to start building my first own big computer system. SMP Symbolic Manipulation Program. Kind of forerunner of Mathematica and Wolfram language. And in the development of smp, a lot of what I thought about was kind of what should the foundation for computation really be like? I was kind of looking back at the last sort of many years of mathematical logic that have been done in lambda calculus and combinators and things like this. Again, I don't think I really thought of that as being a thing that was anything like artificial intelligence. I thought of that as being a thing that was like kind of computation and the entrainment of things like mathematics and logic into computation. But then the place where I did start thinking about this was probably 1980. I started wondering, okay, I can represent, I can do all this sort of sophisticated mathematical type computation within this SMP system. But I knew it was much more general than just being for mathematics. I had built it to be sort of a representation of arbitrary computations. And so then I started wondering, well, what about kind of the rest of what brains do and the rest of what we do in thinking and so on. And one of the things about SMP was like Wolfram language today, it was very much based on the idea of having symbolic expressions that can represent anything and then defining transformations for symbolic expressions, saying if you have an symbolic expression that looks like this, it should get transformed into one that looks like that. And so I very quickly realized that sort of my idealization of thinking was rather than just saying there is a specific symbolic expression or a symbolic expression which just has kind of blanks in it, where you can stuff any sub expression in there, get transformed into this, that really something more like human thinking was something where there was kind of a fuzzy pattern that had to be matched. I didn't really think so much about images in those days, although maybe I did a little bit. But kind of like anything that's an image of a cat, even though the pixels may be very differently configured, should be something that will be transformed into a picture of a dog or whatever it is, or into the string cat or something. That the sort of essence I imagined of kind of the process of thinking was something not so unlike this kind of transformation of symbolic expressions that I defined as the core of smp, but. But it was something where instead of having this kind of well defined symbolic representation of the expression be the thing that was transformed, that it should be some much fuzzier kind of thing where I could kind of deal with kind of these. These very different sorts of things, but very different kinds of presentations of things like the cat. Still thinking of it as the concept of a cat. And at that time, I remember being interested in the notion of kind of fuzzy hashing. So when you make a hash code, a hash code is something where you'll take something like a word and you kind of grind it up by some algorithm and turn it into, for example, a number. And that's very useful because if you take a word and you map every word into a number between, you know, 1 and 50,000 or something, or maybe 1 and 100, you can say, well, do I yet have this word? If I have nothing that hashed to 82 or something, and this word hashes to 82, then I know I don't have that word yet, those kinds of things. But ordinary hashing has the feature that, yes, two very different things can hash to the same number, but it doesn't have the feature that two nearby things, like two different images of a cat, will hash to the same thing. And so I got interested in the question of could I invent some way of doing sort of fuzzy hashing that would let one kind of deal with things in a way that seemed more like human thinking worked. And that was around 1980. And I kind of imagined that maybe I could do that and include it as part of the S and P system, as part of something which would be sort of an alternative to the traditional pattern matching of smp. And kind of in pattern matching, in smp, really much like Wolfram language, you take an expression and you say it's like F of X blank. And that X blank can stand for any other expression. You can put that in there, it's going to be named X, and then you can do whatever you want on the kind of right hand side of that rule. Well, in smp it wasn't X blank or X underscore as it is in Wolfram language, it was $X. And I had a bunch of less good ideas about what the right hand side of that transformation looked like and so on. But, but one of the things that I did try to do in SMP was make a sort of more elaborate way of handling pattern matching. So for example, if I had something like f of $x comma $y comma $x plus $y. So in Wolfram language, those x's and y's would just match if you could stuff another expression in place of the X and y. So if you had A and B, you'd have A and B there. And then the $X +$Y would be A +B and all is good. But in S and P, I was kind of obsessed with the question of what will happen if you said F of 1 comma 1 comma 2, and would that match? Because the 2 isn't literally $X plus $Y, but it is something that in some semantic sense is $X plus $Y if $X is 1 and $Y is 1. And so I had this very elaborate kind of implementation of so what I called semantic pattern matching, which was a way of doing evaluation inside the pattern matching process. It was a very bad idea. Let me be clear about that point. It caused the pattern matcher to have all kinds of strange slowdowns and kind of undecidable situations it could get into. It was not something that was a good match with the process of pattern matching, with this kind of symbolic transformation idea. But in any case, it was something which kind of went a little bit in the direction of sort of the fuzzy pattern matching idea that I had thought was kind of the notion of what was needed for kind of a More brain like thinking like kind of thing. So in any case, I thought about that a bunch at that time. I had all kinds of. I just found recently actually some notes I made at that time with all kinds of ideas about the way in which one would kind of set such a system up and how it would be a little bit about how it would learn. But I wasn't so concerned about how it would learn. I was most more concerned about how it would sort of do the. Do, kind of do the matching process. Okay, so then I guess the next thing that happened, probably this must have been 19. Well, let's see, it must have been 1981. A strange thing that happened was this. At the time there started to be enthusiasm for so called expert systems. Expert Systems had the idea that one would take kind of knowledge that humans had and somehow encode it in computable form. Something that I'd been much interested in in later years. But at the time it was kind of this process where you would kind of interview a human and you would sort of write up code that represented what the human kind of, the human's kind of map of knowledge and so on and way of thinking about things and sort of magically put that on a computer. And I was never deeply involved in doing that. And I don't think I ever really, whenever I looked at systems that were supposed to be doing that, it was like, well, it's just a piece of code. I wasn't sort of. I didn't think there was anything terribly magic about them. But the first company I started that was a company to commercialize S and P, a company called Computer Mathematics Corporation. And in those days, one of my many mistakes was that I was at that point, I don't know, a 20 year old or so, sort of young academic. And I thought, well, I don't really know how to run companies, I should bring in other people to do that. And the folks I brought in were people who had gotten very interested in this kind of expert systems idea and had gone off and started another. They also had another company they were trying to get started that was really in the space of expert systems, kinds of things. And so in the end, through a series of kind of twists and turns, the Computer Mathematics Corporation merged with a company that was created by some overlapping set of the same people. And the merged company became called Inference Corporation. And which is kind of a very modern sounding term I suppose in these days. I kind of hated that name at the time because I thought, and it always charmed me that the well, its logo was an implication sign. And kind of the idea was that it was sort of from the expert system side. It was going to all be about kind of being able to encode knowledge in some kind of logical form and deduce things from it. I never believed in it. I never thought it would work. I never thought it would be useful. As it turned out, that company, sort of not with my involvement particularly, did end up making use of those kinds of technologies on a number of things. Building one of the, I guess the first kind of fraud detection system for credit cards and things like testing systems for NASA, for, for spacecraft and things like this. But in any case, that was kind of a non introduction for me to AI because that was kind of in the expert systems mode of things. But I didn't really pay attention to it and perhaps I even was a little bit allergic to it because I thought from a sort of corporate point of view that it was taking things in the wrong direction. So I guess the next thing that happened in my. This is a long and shaggy story. I thought this was very simple, but as I think about it, it's not quite so simple. The next kind of big thing, I suppose that happened for me in this regard was this must have been 1981 or so. I was interested in this general question of how complex things managed to happen in the world and what kind of models one could construct for that. And I was kind of looking for what were sort of foundational models that can be used. And I ended up thinking of brains as kind of one example of how complex things happen, potentially from the quote, simple components of things like neurons. And so I started reading a bunch of things that have been written by people like John von Neumann and people like McCulloch and Pitts and so on about neural nets. And I also was kind of thinking, well, what's the best way to represent kind of a minimal model for kind of a bunch of components interacting in some way. I was also sort of somewhat inspired by the Ising model model of spin systems invented in the 1920s that was very sort of commonly studied in physics by that time. But in any case, the results of this was, I think it must have been sometimes, maybe early 1981, that I started working on these kind of arrays of cells with simple rules that determined the color of a cell based on the colors of its neighbors. And I fairly quickly learned that things like that at least I was looking at the one dimensional case, at least in the two dimensional case that things like that had been looked at before and had been named cellular automata. And so that was kind of my. My. The direction that I went in. I had originally thought of the thing I was trying to study as being something that we relevant to brains and neural nets and so on. But cellular automata were a lot simpler than neural nets. And I really was pulled very much in the direction of studying cellular automata and realizing that there was already tremendous richness in cellular automata. And so I didn't. It wasn't really worth me sort of looking at neural nets because that was a more complicated case. So then I suppose the next thing that I remember happening in this tangled tale was perhaps 1982, I was at Caltech and maybe it was 1981 still, and a chap called John Hopfield, who'd been kind of a physicist then biologist and so on, was making a lot of noise out of this kind of neural net that he'd created that could kind of store memories and that could be set up using essentially linear algebra kinds of methods. It could be sort of defined to store a memory. And so I got interested in this and it was something where I kind of thought it might be relevant to my kind of fuzzy hashing idea and so on, because kind of had this feature that you'd started off from many possible initial states and it would evolve kind of dissipatively towards an attractor and where sort of every initial state that was like a cat would evolve to the cat attractor, every initial state that was like a dog would evolve to the dog attractor. I have to say I was a little bit frustrated because I did try and simulate these networks and it never quite got the same answer as Hopfield got. And I think I asked him about it a couple of times, but I never really pursued what was going on. And I think it had to do in part with maybe with the way that. Well, I don't really know. It could have been anything from round off errors on. On computers and different ways of evaluating things to different ways of handling asynchronous updating of neuron values and so on. But any case, that was a. A, you know, I filled up a notebook of kind of neural network kinds of things, but didn't get very far. At that time then I guess I was very concerned with things like cellular automata where really it was you give the program and then something happens and you go and analyze what happens more so than anything that could quotes learn in any particular way. But I became friends with people like Tarasinofsky and other People who were trying to kind of rejuvenate the field of neural nets. The field of neural nets initiated in 1943, had kind of fallen upon hard times in the 1960s, largely as a result of a book written by Marvin Minsky and Seymour Papert, both of whom I knew. Marvin Minsky I knew fairly well. That book called Perceptrons had kind of argued that, well, neural nets, at least one layer neural nets, really couldn't do anything interesting. They could just make sort of linear distinctions between things. And people kind of extrapolated from that to say, well, then neural nets don't do anything interesting. But back around 1980, a group of people from several kinds of disciplines, partly physics, partly biology, partly psychology, had kind of ended up sort of saying, well, actually know that there really are things, particularly if you have hidden layers between kind of the initial input layer and the output layer. And so you have multiple layers. There are things that you can make neural nets do. And so I kind of was. Was aware of those things going on. I didn't. I didn't get deeply involved in it because I always thought that I was interested in kind of the sort of fundamental science of how complex behavior happens and so on. And I felt there was still a lot further to run on that kind of thing before one dealt with kind of the case of brains. But, well, I think probably by 1985, 1986, I had gotten more interested in how would one essentially program a cellular automaton? How would one make a cellular automaton do what one wants? And that led me to kind of these questions of whether one could have it learn, so to speak. And I wrote like one little paper talking about kind of cellular automata with extra kinds of bits that would be used to specify their behavior and trying to understand whether one could sort of learn in that context. It didn't work very well, and I kind of didn't pursue that. Well, then, gosh, then I started working on what became Mathematica. And I really didn't think of that as being an AI ish kind of thing. And AI as a field had really fallen upon hard times by that point. So this is now the end of the 1980s, and people really weren't talking much about AI. Now, people were occasionally bringing up neural nets. Occasionally the term machine learning would arise. Machine learning wasn't really much of a thing then. People would. Would. Even quite quickly after the emergence of Mathematica, people started building kind of simulated neural nets in Mathematica. I never paid much attention to them. I really always thought, well, it's very complicated and mushy, and I'm not sure that anything interesting can come out of it. So then what happened next? Well, I went off and by 1991 I was working on my big book, A New Kind of Science, about what simple programs typically do. And in that book I didn't talk much about brains, but I had a section about perception and analysis, chapter 10. And that really has to do with. For me, the critical point was we have these very simple programs and they seem to do very complicated things. What does it mean, seem to do very complicated things? How does that relate to kind of how we perceive things? And so on. And kind of my point in that chapter was the programs that we use for perception and analysis are themselves also simple programs, just like the simple programs that we are using to generate this behavior in the first place. That's a theme that many years later would now come back in our physics project and in talking about the ruliad and talking about observers and so on. But at the time I was really in that chapter 10, I was really sort of cataloging the types of perception analysis that one could consider doing. And one of the sections there was precisely about kind of brain like perception. And I really talked again about my kind of fuzzy hashing kind of idea and kind of talked a little bit about neural nets and talked about kind of this, well, really this, this notion of attractors as being a key part of the process of perception. Something that even now in the observer theory that I'm trying to build is something that comes up yet again. But I saw that section of my book as being kind of just a, you know, dotting the I, crossing the T. How does this relate to how brains perceive things? Okay, so I finished that book in 2002. And then I kind of got interested in something that I have an interest in since I was a kid, which is how do you organize kind of general knowledge about the world? When I was a kid, I had kind of collected and typed up all kinds of documents about kind of organizing information about all kinds of things, mostly about physics kinds of things. But I had always thought, and this had been something that had been in my mind back in 1979, 1980, when I was first kind of thinking about could I extend the SMP idea to a brain like thing? Could I take kind of knowledge of the world and make it what I would now say, computable, make it so that one could kind of answer questions on the basis of that knowledge. So every so often I would kind of revisit that issue. And I think I always had the idea, certainly I had that idea in 1980, that if one was going to make a general thing that would kind of respond to questions on the basis of knowledge that's been accumulated by our civilization, that sort of the way to do that would involve building what we would now call an AGI, a sort of general artificial intelligence that was kind of brain like. So that was what I had believed for many decades. Then I worked on my book New Kind of Science, introduced this idea of the principle of computational equivalence, started realizing that I couldn't make a sharp distinction between the intelligent and the merely computational, that there really wasn't a way to distinguish the notion of the weather has a mind of its own from we humans have minds of our own. So then having finished that work and thinking again about this question about kind of computational knowledge, I realized, look, if I believe my own philosophy, so to speak, it must be the case that this thing that you need to build, this computational knowledge engine you need to build, really has to be something that can be achieved just computationally without some spark of intelligence coming into it. I forgot a piece of the story here. Back in 1984, 85, I got involved with a company that became called Thinking Machines Corporation. I think for my sins, I actually suggested that name. And in the end, I don't think it was a very good name for the company. But that was a company that had spun out of mit and its kind of goal was to take for example, a theory that Marvin Minsky had had and to sort of implement that theory of brains in this thing called the connection machine computer, which was intended to be a massively parallel computer with 65,000 separate processing elements. In the end, I used that machine and thought about that machine mostly for studying fluid dynamics with cellular automata. I invented a language called C Star that never really made it very well. It got made rather more complicated than I had imagined it. That was a language that was an extension of C for doing parallel programming and so on. But I didn't really think of it as being a sort of a brain like machine, even though that had been sort of its origins in this theory of Marvin's. And I kind of, I really was much more concerned about it as this thing with a hypercube where you could route messages from here to there. But that was another sort of exposure that I had to sort of the, the, the, the, the aspirations of AI. And I had gotten to know folks like Marvin Minsky who were very Much sort of in the. In the AI direction, I have to say. You know, Marvin had given me books like, like he'd written this book called Society of Mind. He gave me a draft of that book. I couldn't really make anything of it. It was. It just seemed like a bunch of kind of sort of cute aphorisms about how brains might work. And I didn't really. It didn't really. I didn't really engage properly with it. I think Marvin had originally intended to write that book in verse and I think got talked out of doing that as it was. It was kind of a one idea per page type book. And for me, the ideas didn't really. Didn't really sort of engage particularly. So anyway, but back to 2002 and so on. By about 2004, 2005, I was kind of in this thinking about could one make kind of a computational knowledge engine and what would it take to do it? And I realized if I believe my own theories, it better be something that can be achieved just with computation without sort of a magic of a brain like thing. By the way, I should say back in the days of thinking machines, one of the things that had been kind of the concept behind that company was can one invent the transistor of artificial intelligence? I remember one of the investors in that company talking about, if there's a one in a million chance to invent the transistor of artificial intelligence, then this is a good investment. And I kind of rather jadedly said, I wonder how long it'll be before he's banging his fist on the table saying, where's the revenue? The answer was about 18 months. But in any case, that was kind of a thing. And I think I was somewhat of that same mindset that with my fuzzy matching ideas and all this kind of thing, I imagined that there was some core idea, some sort of transistor of artificial intelligence like idea that would sort of unlock things anyway, back to mid aughts, so to speak. And I was at that point saying, okay, if building a computational knowledge engine just requires computation, we have this great computation stack with what was then Mathematica, now Wolfram language. So let's just try and build it. And so we got started building what became Wolfram Alpha. And a big part of that was solving the natural language understanding problem. And we needed to have something which could take kind of natural language, however people said things, and turn it into our computational language from which we could compute. And at first I kind of imagined, well, I can learn things from computational linguistics. I can learn things from things that have been done in AI in the past. And we tried, but we really didn't. And we ended up just building our own kind of from scratch kind of way of understanding how natural language understanding for sort of short fragments could be done. And it really helped us that we kind of were able to. We had a target. People had said, well, in the past, people had said, well, let's just understand natural language. But the problem was it wasn't clear what understand meant for us. It was very clear. It meant convert it into computational language and then be able to compute from it. You know, as I'm. As I'm talking to you guys, I'm remembering a few other things. Back in the 1980s, there'd sort of still been kind of buzzings of AI and there was kind of the. There was the expert system strand, there was the more kind of sort of learning kind of strand. And I remember having various experiences of being exposed to kind of the AI as mystical kind of force type, type side of things and being deeply unimpressed. So, for example, there was one time I visited, I think it was Yale actually, and visited some group that was doing AI, and they told me, oh, we have this technology. This must have been 1983, four, maybe five. We have this technology that translates stories from Spanish into English. At that time, that was a totally science fiction thing to be able to do. That was something maybe in 2001 or in Star Trek, that was a thing. But in terms of real technology, that just wasn't a thing that was possible. People had imagined that machine translation might be quite easy to do back in the 1960s, but it hadn't worked out that way. So it was very surprising to me. And I'm like, okay, that sounds great, let me try it. And they were like. And I said, I don't know much Spanish, but let me see what I can type in, so to speak. Well, no, you don't understand. It has to be, you know, a whole story that you're translating. And so obviously I'm asking, well, how long does a story have to be? What's the, what's the minimum length of a story? And saying, well, you know, it's. And anyway, we go through a whole series of kind of, well, can it be this, can it be that? And eventually what becomes clear is it has to be stories of a particular type. Well, actually, it's not just stories of a particular type. It's a particular story that was supposed to be translated from Spanish into English. And I'm like, this is crazy. What is this? What are you doing? And their explanation was, well, we're trying to understand the structure of the Spanish and how it maps under the structure of the English. But the problem was, for me, the distance between the original claim and the actuality was so great that it really cast a bad shadow over things people might have been saying about AI. I think the other thing that I should have mentioned is Lisp was a language that was very much wound up with AI. John McCarthy, who invented the term artificial intelligence, had developed Lisp starting In the late 1950s, a very kind of visionary direction in language design, although it was hard to actually implement in practice. And in fact people had built things like Lisp machines. And in fact the company Lisp Machines Incorporated was in various ways wound up with the company that I started, that or at least the merged version of the company that I started. And so there have been sort of Lisp had been kind of in the air. And it didn't help that there was a certain degree of kind of playful non scienceism that sort of seemed to float around all things AI in those days. I mean, I remember this must have been 1979, 1980. I was developing SMP. I had organized a group of a collection of seminars about kind of how one might do math by computer and other kinds of things like that. And I was inviting people who had worked on similar things before. And there was a chap from Berkeley who came down to LA to Caltech one day and we were giving this talk and he was talking about some kind of things Lisp and was talking about, oh, they have the system and it's called Franz. And it's kind of like asking the audience, why is it called Franz? Well, it's a Lisp system and that kind of is like the composer list, you know, Franz Liszt and so on. Well, I happened to be. Dick Feynman happened to be in the audience, I happened to be sitting next to him. And Dick Feynman kind of stood up at that moment and said, if this is what computer science is about, it's all nonsense, and kind of stormed out causing a, a very long term, I would say, kind of the person who'd been giving that talk kind of blamed me for that situation, I think, and had a very long term grudge about it. But in any case, the, the point was that in, in those days, this kind of we've got this playful, hacky, you know, acute sort of thing that wraps around AI was really very common. And to me it was rather off putting. It was kind of like, well, tell me what is really going on? What's the real point here? What can you really say? What can you not say? And so on. Perhaps I would have a different attitude today, but in those days it was sort of frustrating. It was like, where's the meat? Show me some meat. You know, don't give me puns, don't give me kind of sort of fake versions of claims and so on. Just show me the actual meat. And I never saw the meat in those days. But any case, back in 2009, when we released Wolfram Alpha, we had succeeded in doing really quite good natural language understanding, targeting Wolfram language as the thing that it was understood to and being able to answer all kinds of questions. You know, one thing that was sort of an indication of the way that field developed was a couple of weeks before Wolf Malfoy was released, I was at some event and Marvin Minsky was there. And I said to Marvin, you know, I have this question answering system which we've built and you should look at it, it's pretty cool. And so I open up my computer and I started off and Marvin kind of looks at it for few seconds and then kind of changes the subject to some wild thing that he liked to talk about. And I was like, you know, no, Marvin, you should actually look at this. And you know, because it works, you know, it's a real thing. And so he looks at it, starts typing things in and starts typing more things in. And then he's like, oh my God, it actually works. And you know, he's going off and telling all the other people at this event that same thing. So for him, kind of, he'd seen a zillion question answering systems, I'd seen a bunch of question answering systems and they were all, I would say, very, very lame. They really didn't work. They had been often built using kind of the AI ish tradition of kind of let's use. I think what happened was that there was a sort of a strange merger of a belief that logic expressed everything about the world, that you could encode the whole world in terms of logic. That there was something sort of psychological that needed to be represented in AI. I mean, I remember asking Marvin a few years later, he was writing a book, I think it was called the Emotion Machine. And I said, what is this that you are writing, Marvin? What kind of a thing is it? Is it kind of a thing like a theory of psychology, like a Freud or something like this? Is it a thing that's like kind of computer science and like a prescription for how computers do things. You know, what kind of a thing is it? And I think at that time he said, well, if it's any of these things, it's something more like a kind of theory of psychology. But there was sort of this strange mixture of. Of kind of the sort of theories of psychology together with kind of primacy of logic, which is very different from that, together with a certain degree of sort of cute hackery that seemed to be kind of part of the main AI shtick. And then quite separately, there was the let's use neural nets to build practical things. And I don't think I knew at the time that things like optical character recognition, OCR methods that were being developed were all really based on neural nets. Maybe I knew that at the time, but I don't think so. And to me, those were just, oh, it's a random algorithm. I remember actually this must have been 2010, 2011, even 2010, perhaps. I was interested in what we could do in image recognition. It must have been whenever the Fukushima nuclear accident was. It was shortly after that because we had done kind of an emergency project for an organization that ended up consisting of looking at satellite photographs and trying to identify cars on roads and where there were cars and where there weren't cars in the region of that accident and so on. And so I'd kind of realized that the way we identified cars was using sort of hacky image processing techniques. And there was similarly known kind of hacky image processing techniques for doing things like finding faces and images. And so I sort of wondered, given that in WolframAlpha we curated all these different kinds of knowledge about the world, I wondered, could we curate kind of the. The sort of knowledge of what pieces of image processing one would have to do to find, you know, windows in a picture of a house, or find cell boundaries in a pathology image or some such other thing. And I kind of imagine there might be a couple of hundred of these kinds of types of objects that were worth looking for in pictures. And I kind of was all ready to start building kind of kind of a pipeline where people would sort of construct these image processing hacks to find all these different kinds of things. And I remember asking Teresinofsky neural Net friend of mine right at that time, you know, what can we do? Is there something better we can do with this? And he was like, well, I don't really know things. There are particular examples where you can use particular neural nets to do this stuff. And that was right before the whole Alexnet story with Terry's friend Geoff Hinton and so on. And this realization that as it happened, deep neural nets could be successful at identifying objects just from training without sort of building that hacky image processing kind of thing. And it certainly, when that happened, there were sort of, very quickly sort of the, the kind of the big rumor from Silicon Valley of, oh my gosh, neural nets have now worked. You know, behind the scenes. That was, you know, Google buying Jeff Hinton's kind of pop up company and things like this. But, you know, that got me to kind of look at these things again and realize, oh yeah, actually there's some stuff that's starting to work here. And, and then we got involved a person called Talibanon who, longtime person at our company kind of said we got to build a machine learning pipeline for Wolfram language. And so we very quickly started constructing what is now our whole symbolic neural net framework within the language. But that must have been 2012, 2013, something like that. So those are so, I mean, and then we kind of were on the modern path. But I would say that in the previous period things, you know, whenever I sort of heard about neural nets, it was always, oh, they're very mushy and complicated and you know, they're a thing that has a sort of an air of mystery about them. They reminded me a lot of global minimization, global optimization. You know, when you are trying to find you have a mathematical function, for example, and it wiggles around, you're trying to find where's the minimum of that function. Function question is, there are techniques for saying, well, I'm close to a minimum. Let me home in on the final minimum. You just sort of descend along the gradient vector and so on to get there. But the big question is, can you find the global minimum of such a function? And back when we first started developing mathematica back in 1986 and so on, was certainly a question. Could we put in a function called minimize that would just find the minimum of every function? But it just seemed completely implausible to me. It just seemed like there was no sort of way to be reliable in doing that. But then people would come regularly and they would tell us, we've got this magic sort of black magic black box function that just minimizes any function. And I would never believe it. And some of those things were a little bit neural net oriented, but they had the same sort of flavor as people saying, I've got this neural netty thing that does something magic. And we can't really explain what's going on. Many years later, the techniques for function minimization kind of got, I would say, robust enough that even though they're not perfect, they become a thing that's useful enough that one can put them into our language and people can rely on them, at least so long as their functions aren't too weird and crazy. But in any case, that was sort of, you know, the thing I'm trying to communicate is that there was this kind of. Yes, neural nets were a thing, but they were kind of mushy and mysterious and occasionally people would claim that they were doing things with them, but it wasn't really. They were, they were really one off kinds of things. I remember a chap called Robert Hecht Nielsen who had kind of a particular sort of approach to neural nets. He had a company, it was making, I think they made fraud detection software in the end. But it was just very unclear to me what was really going on there. I mean, to what extent. It didn't help that there were techniques that would sometimes, not very often get described as being machine learning, but that was really a sort of disreputable term. But more often it would be like, oh, we're doing some kind of regression, we're doing something that's part of statistics. But it wasn't very clear to what extent the things that were sort of neural nets and it's like brains and the things that were statistics and like math, how those really related to each other. And that didn't become clear, I would say, until really recent times. And there were a bunch of kind of competing methodologies that were kind of self organizing maps and oh boy, what other kinds of things. Various kinds of trees and became random forests and so on and various other kinds of, oh, I'm forgetting there were three or four other methods that were all things that were sort of part of the toolbox of can you sort of magically make a model for data? And by the way, one of the things that again suffused this whole area for me was people saying we've got this way to make a model for data. And it worked for any data and it's a modelless model. There were a bunch of things around the maximum entropy method which again sort of had the vibe of being modelless models. You just throw this machine at anything and it will make a model that allows you to make predictions about that thing. And I found that very, very unconvincing because I kind of knew that anytime you're extrapolating beyond what you've actually seen, you implicitly have to have some kind of model of what's plausible. And even though people could say, well, our model is that it's really, it's always plausible that our model will do something plausible, it really didn't make sense. It really was something where there was sort of a recursive. You hadn't sort of grounded the recursion to say what the model that you were using to model was really based on. And so that always caused me a lot of trouble when people were telling me about these kinds of techniques. So that's a little bit of, that's a, a rather long and shaggy story about my kind of personal history with what's now called AI. I would say that sort of, if you project backwards, a lot of things that I've done that relate to kinds of mathematical computation and things like that were things that back when I was first involved in these things, people would say, if you can only do that, we know we have true AI, so to speak. And yeah, I mean, I think there are other pieces to this. I mean, I have to say I know I collected lots of the literature of AI going back many years from the Alan Turing 1950 kind of thing about AI to lots of other things. And oh yeah, another piece to this. Gosh, I'm sorry, these things tumble out of my memory as I start thinking about this. There was a whole nother tradition of brain like studies that was essentially physicists generalizing things like the Ising model to neural nets and asking what their kind of physics like analysis would be. There was a chap, chap called Gordon Shaw, a chap called Can Yellow. I think various people who were, who were kind of, who did things that to me were actually more appealing. They were kind of more. This was 1980s again. They were more kind of science like, but they didn't really get that far. They were able to say, well, you know, we have this thing and it's like kind of a bulk neural net and what does it do? We don't really know what it does. There were people as a chap called Jack Cowan at University of Chicago who was studying things like kind of bulk depolarization waves on the brain for things like epilepsy and things like the overall mapping of, of kind of a visual image on the retina onto the brain. These were, these were again more in the tradition of kind of physics and mathematics, like analyses of, of of brain like things. And I think that was, that was a kind of another thing that I used to sort of follow. And I collected papers on it and things like this. There weren't that many papers about these kinds of things, but somehow this stuff, it never really the. The. The connection between those things and the things that I was able to do in technology, the things that I was doing in science, the cellular automata, they never really all quite connected. And it's only in very recent times that I think we can sort of see a bunch of that connection come to pass. It sort of. And anyway, that's a long answer to what seemed like a simple question there. Let's see the question from Ian. Have I yet seen the movie that's coming out about Oppenheimer? I have not. I hope to see it. I think it isn't out, at least in my suburban part of the world yet, but I hope to see it in a few days. I will say that I knew many people who had worked at the Manhattan Project, and it was sort of interesting to see its role in the history of physics because, you know, the Manhattan Project was taking place in the early 1940s, and I was getting involved in physics in the mid-1970s. But many of the people who've been involved in the Manhattan Project have been young people at the time when they were involved. And there was somehow this certain mystique that seemed to sort of be around the physicists who've been involved in the Manhattan Project. I'm not sure that everything about the Manhattan Project had been declassified yet by the time I was sort of first encountering these people. So perhaps there was a little bit of, you know, person A knows what person B did, but can't tell anybody else about it. But there was a mystique, kind of a glow of they'd done something important, and they'd also been part of this kind of small sort of brotherhood guild or something of people who had been involved in that project. I would say that I started. I was a consultant at Los Alamos in the early 1980s. I think when I first went there, Los Alamos was still a town that didn't appear on maps. It was something where, you know, drive north from Santa Fe in New Mexico and you'll get to Los Alamos, but you don't see it on the map. And so that was kind of exotic. But a lot of the people, some of the people who'd been involved in the Manhattan Project were still at Los Alamos at the time. I was a consultant there. But much of that group had sort of become a diaspora, going off to different universities in Some cases going back to different universities. I mean, I remember one person. I mean, there were many people who'd been involved in the Manhattan Project who I encountered. I mean, Dick Feynman was one person I knew fairly well. He'd been involved as sort of the head of the, of the human computer project, so to speak, because in those days they weren't yet electronic computers. It was, there were mechanical computers, calculators, but there was sort of a pool of people who computed things. Dick Feynman was really a very good human who calculated things, and he was probably a very appropriate person to be leading the kind of, the pool of people who would actually calculate things by hand and say this is the result for this kind of, I don't know, kind of detonation wave or whatever else it was. And certainly Dick Feynman told me a bunch of stories about the Manhattan Project. You know, he told me about seeing the first bomb test and kind of the, the feeling that he had after that, that, my gosh, the world's just going to blow itself up. And, you know, seeing people kind of building things and wondering, you know, what's the point of building these things? The world's just going to blow itself up. Although he kind of got over that after a while. He also told me about the, the moment when they were trying to work out, would, would setting off the first nuclear explosion ignite the atmosphere? Would it start leading to a reaction in which the oxygen in the atmosphere would basically burn and the whole atmosphere would be, would have an exothermic reaction and destroy all life on Earth? And he was like, told me, at least he was like, we're pretty sure that wouldn't happen. We kind of got convinced that wouldn't happen. And so the bomb test happened, so to speak, and fortunately it didn't ignite the atmosphere. But of course, of those days, one didn't really know for sure. It was very hard to predict that for certain. I think Feynman's team of calculators was part of the workflow of trying to work out would, would it ignite the atmosphere? I think GI Taylor was another person who was much involved in that, a fluid dynamicist who I don't think I ever met, but who was famous for kind of making use of dimensional analysis as a way to sort of do a back of the envelope calculation about what would happen in nuclear explosions and managing to get it more or less right. But, you know, Feynman told me other stories about Los Alamos, many other stories. He was very impressed by Oppenheimer, by the way, he thought Oppenheimer was a very impressive sort of kind of outsized leader, a person who could kind of get people to do things in a very effective way, so to speak. I think there were other kind of sideshows to the Manhattan Project. It was a chap called Klaus Fuchs who turned up eventually. Turned out eventually to be a Russian spy, but who was kind of involved in the British nuclear effort and then worked at Los Alamos. And Dick Feynman told me a bunch of stories about what he sort of realized after the facts fact was kind of Klaus Fuchs trying to sort of sound him out on his views about communism and so on. I have to say that when I worked back 1976, briefly after I left high school, I worked at a British government lab called the Rutherford Lab, which was kind of next door to Harwell, which was the British Atomic Energy Research Establishment. That was a nuclear. Well, it was nuclear physics, let's say. There was a separate place that was the ultimate weapons development lab, but at Harwell, Klaus Fuchs had been the head of the theory division there. And, you know, he'd eventually sort of come out as a spy. And when I was working there in 1976, there was still definitely a very long shadow being cast over the operation and the sort of the theory group there by the fact that its leader had turned out to be a Russian spy. But in any case. But back to the Manhattan Project, I think the thing that there were a couple of things that were very obvious in the kind of American physics environment that were kind of holdovers from the Manhattan Project. One of the most important was that sort of the physicists had done something sort of great for the war effort. And it wasn't obvious that anything like that would happen. It wasn't obvious that sort of the somewhat abstract physicists would deliver something as major as nuclear weapons. Engineers had been working on radar, and that had been a thing that was impressive. But this was a great big kind of wow for physicists in their contribution to kind of the war effort and so on. And I think there was a certain feeling that the government kind of owed the physics community as a result of the Manhattan Project, that the Manhattan Project had just done something remarkable for the government that somehow needed to be repaid to the physics community. And I think that what happened in the development of particle physics after the Second World War, I always saw that as being a kind of long thank you from the US Government to the physics community for the Manhattan Project. And it Was kind of a thing where the particle accelerators, what were they going to make? Were they going to lead to something that was sort of a new generation of military important thing, or were they just basic science? Most people thought they were just basic science. Famous quote from Robert Wilson, who was the guy who built up Fermilab, being asked, what was the significance of Fermilab for the defense of the United States? It was asked that in some congressional testimony. And he famously said, you know, it doesn't contribute anything to the defense of the United States, but it's part of what makes the United States worth defending. That was a sort of. And at the time that was, I think, said probably in the late 1960s, maybe early 1970s, that was a. That seemed like a very valid thing to be. To be saying. I think that the kind of the long thank you for the Manhattan Project kind of lasted until really the mid-1990s. And that was when the people who'd been involved in the Manhattan Project were retiring and so on. And in kind of it all, there was another generation of physicists. And the sort of. The last test of the thank you for the Manhattan Project was the Superconducting Super Collider project, this big thing that was eventually to be located in Texas, sort of the great new American particle accelerator. I forget how much it was going to cost, probably on the order of $10 billion. And it was kind of like, was this going to be a. Was this going to be a thing that was worth the government supporting? And it felt like kind of the glow from Manhattan Project was sort of dying off. And it felt like there was a problem there. And, and what had happened within the physics community itself, the condensed matter physicists were feeling they were more important. You know, they've been involved with a transistor and superconductivity and things like this. It's like, why are all the particle physicists getting all this money? We want some of it too, although they didn't have as much to spend it on. It's only in modern times, with kind of the rise of the story of quantum computing, that there's been kind of a. A big ticket thing that sort of condensed matter physics could. Could be part of. But anyway, at the time, I remember this must have been mid-1990s. I remember being at some dinner at Fermilab with the then director, Leon Lederman, and there was the sort of. The next day was going to be this big vote about the super collider in Congress. And I spent the whole dinner trying to. Trying to convince Leon that he should come up with a quote like the Robert Wilson quote that would be remembered if the vote went against the super collider. That it would be remembered at a time when sort of people were thinking again about spending large amounts of money on basic research in the particle physics direction. I completely failed to convince Leon to come up with some poetic quote, so to speak. And in the end, I don't remember what was said, but it was nothing very good when, as one kind of thought was going to happen, the vote went against the super collider and the project was killed. But I think then sort of another piece of the Manhattan Project story. Well, I mean, I. I knew Stan Ulam, who was also quite involved in the Manhattan Project, and I guess more so in the hydrogen bomb than. No, actually, I'm taking that back. I think that the implosion lens idea, which is critical to the operation of nuclear weapons, was a. I'm now forgetting this. This was. I think that was Teller, Edward Teller and Stan Ulam were sort of the originators of that. And I think I knew Stanley very late in his life when perhaps it was a little different than what he had been like earlier on. But then another sort of piece of the. Of the consequences of the Manhattan Project were that Oppenheimer, after the Manhattan Project, Oppenheimer, I think was the next sort of big question was, okay, there's this sort of powerful thing that's been created with nuclear weapons and who should say, what will happen with this? And Oppenheimer and quite a lot of the other physicists were like, well, we physicists should have a say in this. It's too important to be left to the politicians. And so that. That became quite a kind of push to. With things like, you know, Pugwash and the Union of Concerned Scientists and the union of. What was it? Oh, gosh, Bulletin of the Atomic Scientists and so on, a whole collection of kind of essentially activism on the part of physicists and so on to kind of have a real say in what should happen with the thing that they, in a sense, intellectually had created. And I think that Oppenheimer was involved in that. And the whole question, the questions about what would happen with the hydrogen bomb. I think it didn't help that the Soviet Union, perhaps in large part through the efforts of Klaus Fuchs, had, you know, was rather quick to be able to have their own atomic bomb and so on. And that sort of created this. This climate of tension about what was going to happen. And perhaps in the middle of all of that Oppenheimer, who was by all accounts a sort of forceful, if brash kind of character, sort of ran afoul of the kind of political forces and ended up getting sort of attacked for his connections to communism and so on. I mean, it was sort of a. I think, you know, that had been a. A very typical thing for many people in that orbit. You know, Oppenheimer had worked in Berkeley, which was sort of a place where. Where there was a lot of interest in those kinds of things in the 1930s and so on. I mean, I think that the Manhattan Project was. Was a bit bimodal in, in terms of some of the people working there. There have been people who'd come out of kind of. A lot of people who'd come out of the Nazi Germany story, people who by the end of the war were seeing kind of the, The. The. The Russians as a kind of a negative force, so to speak. And I think it was kind of a, you know, are you, Are you for the Soviet Union or are you against the Soviet Union? And some people were. Were really for it, and some people were very against it. But I think that. But anyway, what happened was that Oppenheimer got sort of attacked and the big drama was that he lost his security clearance, so to speak. And one of the things that had been sort of set up in the aftermath of the Manhattan Project was the Atomic Energy Commission. And it was sort of a big deal to be a commissioner in the Atomic Energy Commission. For example, John von Neumann was a commissioner in the Atomic Energy Commission. And when the one little video snippet that exists from a strange television program where von Neumann is interacting with a bunch of kids, the thing that he's introduced as being is a commissioner in the Atomic Energy Commission. That was sort of the most significant attribute that he had at that time. It's kind of an irony of history that as we look at sort of AI and its. Its effect on the world, that there's sort of a. A certain call which I don't necessarily agree with, to sort of create something that's sort of a regulatory entity like the Atomic Energy Commission that will no doubt have its AI commissioners and so on, to be. To be part of it. I mean, it's of course, to be said that the supply chain of making a nuclear weapon is unbelievably much more difficult and complicated than the supply chain for making an AI. So it's a rather different kind of story in terms of what can be managed in what way in the world. But any case, the result of Oppenheimer kind of leaving the atomic energy world was that he was placed at the Institute for Advanced Study as its director. And the Institute for Advanced Study had been created in the early 1930s. And as scientists ended up leaving Germany in the 1930s, it had accumulated an impressive roster of folks from Albert Einstein to Kurt Godel, who hadn't left quite for those reasons, but onward, John von Neumann, other people. And so it must have been, oh, gosh, Sometime in the 1950s, Oppenheimer sort of arrived to be the director of the Institute for Advanced Study. And he, I think he died in 1968, give or take. And I was worked at the Institute of Advanced Study back. Maybe he, maybe he died later than that. Maybe he died in the 70s. I think maybe he came. I'm losing track of the dates here. In any case, when I worked at the Institute Provence study, which was Starting in about 1983, it was sort of interesting because there was still kind of vestiges of, of sort of imprints of Oppenheimer's work there. In particular. Oppenheimer, I think rather impressively had hired a whole collection of sort of promising young physicists to work at the institute. And he had done kind of somewhat high risk but high payoff hiring. And so the dynamic that happened, that was a, surely an unanticipated dynamic, maybe, I don't know by him, was that by the time I was there they had all these sort of high risk hirings. And many of those people had sort of paid off spectacularly and done amazing things in physics. But most of those people had over the years kind of wandered off onto other opportunities in other parts of the world. And so kind of the part of the risk equation that Oppenheimer had initiated, that was the ones where they hadn't quite been oriented or their lives hadn't worked out in such a way that they could really do the amazing things in physics. Those people were still there. And it often was something where you would think, well, it's kind of a good deal to be just told you can just go think about science and nothing else for the rest of your life. But it turns out that that isn't actually necessarily such a good deal. And that was kind of an interesting and. And in some cases not, not very positive human story, so to speak. Well, let's see. Gosh, I probably once I've seen this movie, I might have more to say about some of the characters who are portrayed in it. I, I don't know who all will be, will be presented in that in that movie, I think I am, or, you know, it was, as I say, quite interesting that there will be people where there was this kind of mystique and sort of, oh, this person has done something amazing sort of aura around people at the end of the. By the end. At the end of the 70s. But it wasn't so clear what those things were. I mean, like, I remember there was a guy called Bob Backer who was a physicist at Caltech, where I sort of always knew had done something notable, but I never really knew what. I haven't looked up even today, what. Maybe he'll be in the movie and I'll find out. But. And there were. It was. It was always, you know, as I say, it was always this kind of. This sort of. This kind of collection of people who were involved in something great together and that left kind of a glow behind. I see the same thing today in people who've been involved in all sorts of startups and discoveries and things like this, that there's always this. This kind of thing about people who are involved in something that turned into something big. Manhattan Project being a. A particularly extreme example of that. Let's see, Find one more thing and then I should go on my way. Here's an interesting question, but I don't know if I can answer it easily. From Prab. Have new scientific discoveries historically initiated out of myths? It's an interesting question. There are often things that people kind of vaguely believe are true, and only much later does it become established what's scientifically true. Then there are things where people believe it's true and the science never really comes. I mean, an example of one which, who knows, maybe the science will yet come is things like, you know, extrasensory perception. Can there be some sort of communication from brain to brain by some mechanism other than sort of sight and sound and so on? That's one of those kinds of things where you might, you know, if something comes of that, it will be a thing where there was almost sort of a myth that this worked and then it became science. I'm trying to think there are things about science where it's just like kind of totally surprising that it works. X rays might have been an example of that. There might have been myths in the past where it's kind of like people can see through things and maybe that would play into kind of this is. This is what X rays are. But really the discovery of something like X rays was just a total surprise. Now, you know, there are. There are things where it's like you know, drink the elixir of eternal life and good things will happen type thing. And, you know, if one day some sort of approach to kind of human longevity emerges, that is just, you know, drink this elixir, then it will feel like that myth turned into reality. And perhaps there's almost a tendency to say, you know, drink the pomegranate juice. It'll be just like the myth kind of thing. And that myth has kind of an overhang in our view of what might be true scientifically. You know, there are certainly things from antiquity that are kind of sort of things people have said and then they turn out to be true. Like, for example, famous one is, was Troy a historical city, or was it just something made up in the Iliad by Homer, so to speak? And it became in the 1800s, it turned out, well, Troy was actually real. You know, something like Atlantis city. I think people like Cicero maybe talks about that as, you know, a city covered under the ocean, so to speak, but had been a remarkable city of all kinds that is thought today to be sort of a myth, but maybe there's actually something to be discovered one day that there will be actually a city under the ocean that will show up. Now, along those lines, Cicero, I think, described something that was kind of a calculating device. And for centuries, nobody knew what on earth that could be until the Antikythera device was discovered. Well, it was first found in the. In a shipwreck around 1900. But it was first kind of figured out sort of what it was by the 1960s, and then in more detail by the. By the 1990s and 2000s, that it was a mechanical calculating device, kind of an original computer, so to speak. So in that sense, that was something that had not quite been a myth, but had been a thing sort of said as a matter of kind of almost philosophical rhetoric that then turned out later to be something real. I think the. This. Well, there are certainly ideas that one can say originate in things like theology that then sort of play into things that one realizes are relevant today. Like, for example, one of the things that is quite relevant and when one thinks about things like AI ethics, is what does it take for things to have free will? And what has free will and what doesn't? What can be responsible for its actions, what. What isn't? And, you know, in Christian theology, the idea that free. Free will was a thing for people, that people could make their own decisions was something that was developed there. But yet there was this other branch which was kind of. But then there are angels where it's kind of decided for in advance are they good or evil? And that kind of plays into the kind of, well, we have a machine and it has kind of a set of rules which decide in advance in some sense, good or evil. And so that's kind of, you know, what started as a story of theology has a certain at least kind of conceptual connection to the things that one now confronts as a matter of kind of things like AI ethics. But I think I'm trying to think about other kind of myths that turned into reality. I suppose I'm thinking about kind of Greek myths of many of which were kind of attempts to explain the world, or myths from the Bible or things like this, where there were sort of attempts to explain the world. So, so for example, one, I suppose that's interesting is something like things in Genesis where it's kind of like. And then there are these creatures around in the world and then they have to be given names. And this kind of idea of sort of how does language evolve? And later things like the Tower of Babel and so on, how is it that language comes to be. How is it that these things are just there in the world and then we give them names and that kind of humanizes them. That's sort of an interesting kind of, in that presentation, a kind of mythical idea which then sort of plays into kind of what does it mean for us to identify things in the world and give them names and so on. So perhaps there's a correspondence there. I don't think we quite. And perhaps, you know, the Tower of Babel we can start thinking about, you know, is there as we see sort of the embeddings in an AI and we see that there's sort of a. Or we see computational language, we see that there's an underlying structure of meaning that then has its different manifestations in the different human languages of the world. And one could sort of play into that, into kind of this idea that there was at one time a universal language which then kind of split into the particular languages of the people of the earth, so to speak. So those are a few kind of correspondences one might imagine. There's. I'm trying to think whether there are kind of discoveries that were originated that way. There were certainly drivers of that kind of thing. For example, the long time alchemical idea that there should be a way to turn lead into gold, which didn't work out as such as a matter of chemistry, but certainly drove a lot of attempts to turn this into that, so to speak, and led to lots of kinds of things in chemistry, I suppose. Well, okay. Another thing that one could identify along the same lines, and this is kind of from that originates in many cultures, is this idea of sort of the immortal soul as an element of something like a brain that isn't the material of a brain, but is some kind of thing that is the significant engram of a brain, but not the brain itself. That was something that certainly existed across many cultures. And it's something which until recently, it just wasn't clear what on earth that could sort of meaningfully be. And one sort of wondered, and people would sort of mock kind of this idea of, well, is there a soul? If the soul leaves the body on death, does that mean the body gets to be. To weigh less? Because there's something that sort of departed and so on. And then one realized that actually the notion of a soul is really very much like the abstract idea of computation. It's kind of a computational representation of what's going on in a mind abstracted from the material, details of neurons and electrical firings and all this kind of thing. So it's something which in some sense was there all along. For thousands of years, in a variety of different, across many cultures, many traditions, it was a thing, but really nobody understood what it was. Now I think we do understand what it is. That was not the causative. That wasn't sort of the causal progression of people understanding this idea of computation and maybe its relationship to things like AI. It wasn't that that came from souls. But I think there's sort of a convergence in that. I think, gosh, I'm trying to think about other kinds of myths and so on. I think I should probably depart here before I Let me think. Is there anything else I can think of that is kind of, you know, these are not myths, but many of the things that, you know, Aristotle tried to talk about, about his kind of theory of the way the world works, the way the mind works and so on. Many of those things were at first, well, actually, let's talk about a different thing. You know, in ancient Greek philosophy, there was very much the idea of let's just by pure thought, imagine how the world must work, and let's kind of develop this idea that there might be atoms and there might be the void and so on. I think this idea, it's not quite the same myth with a story, so to speak. There's not a personality, a human personality involved in that, which seems to be typical of many kinds of myths. But it is something where There was kind of a, a kind of a story that was getting told about how the world exists, which eventually turned into science. But it took a couple of thousand years. I think the. I'm trying to think about kind of other sorts of things and it's often very difficult when you have something which is described in kind of everyday terms, in some sort of mythical terms, and then later on a piece of science gets done that kind of a little bit reminds one of that myth. To what extent should one sort of read back onto that myth, the thing that then emerges, and to what extent was it something that was there for a different reason? I tend to think that very often people have a general feeling about how things work and that feeling gets expressed in different ways as, as myths, as cultural traditions, whatever else. And it's only sort of vastly later that the kind of the science comes in to explain that. I suppose there have been a bunch of things about the way we should live our lives and yeah, I mean, there are all kinds of things about things that are virtuous, things that are not, things that we should do, things we should not. And much later kind of one realizes that, well, actually there was a science reason why that was a good or bad idea, even though in a sense, perhaps we knew it was a good idea and we made up stories to explain why it was a good idea, but we knew it was a good idea because of something that got built in sort of genetically on the basis of biological evolution. We knew it was a bad idea to do this or that thing. And then we sort of rationalized that and turned it into stories that became myths. And then much later we realized by looking, by sort of reverse engineering what's going on in our biology, we realized, well, gosh, that's a thing that has some reason to be for some scientific as a result of some piece of science. All right, well, I think we should wrap up there, so thanks very much for joining me. Lots of. Well, we didn't, we didn't talk about very many questions today, but they certainly stimulated quite long descriptions from me. So I, I hope people enjoyed that and I certainly enjoyed kind of thinking through again some of these, or thinking through some of these kinds of answers. So thanks very much and bye for now.