267 - Lee Cronin: Aliens, Artificial Intelligence, and the Origin of Life

0:00So I would say that our artificial intelligence systems are really historians of human intelligence. An LLM is nothing more than a big historian, but with a bit more of a probabilistic mismatch, for it looks new to you. These AI devices do not have agency. They merely have an illusion of agency or shadow of agency from the creators and the technology. Non-living systems have a very small conventorial space. Not much can happen. Living systems have a large conventorial space, massive evolution number of protein structures. Intelligence systems have an even larger commentorial space. Intelligence seems to be associated with your ability to have an even bigger commentorial space of possibilities, whether you're writing science fiction or you're producing a new product.

0:48I mean, who came up with the idea of a LeBouBou?

1:00As you well know, Lee, this is part two of a conversation that will likely constitute one of the most recent episodes when this current conversation comes out. So while I'm sure this one will be pretty self-contained, anybody who would like more should go back to that one. But before you and I start talking, I'm just wondering where you'd like to get started, because we left off with the topic of AI, but I don't think we really got to the deepest issues or really fully explored the topic. But then there's also aliens. So I'm wondering where you'd like to begin. wow a choice between aliens and ai i think we could do both but i'm i'm very interested in actually us understanding so so i think one of the reasons people think i have lots of strong views on ai for strange reasons but because i'm so interested in solving the origin of life and understanding life and therefore aliens i think that um um exploring ai as constituted right now is probably quite important because it'll probably help people understand where i'm coming from where the field is going because um you know i'm not against ai i'm i'm against um putting the ghost in the machine when it doesn't need to be there there's enough there's enough really interesting questions about inserting ghosts for reasons that i don't quite understand and what we could do is maybe talk about what AI currently is and what it isn't.

2:36That might be fun. No, that's perfect. That's, I mean, if you chose the AI route, that's where I was going to go. So, I mean, we might as well just talk about the name first, artificial intelligence. So what is it? How do you define or think about intelligence? And then maybe we'll talk about the artificiality of it yeah that's a really good question um intelligence is a super interesting phenomena that is obviously produced by evolution and and selection in a way um and lots of people use intelligence as a catch-all for lots of different behaviors and processes and i think that's where the problem comes is there is an ambiguity so for me an intelligent thing is one that is able to autonomously um adapt to the environment changing right you know if i'm like i don't know go in escape room and i've got to escape from escape room i've got to solve clues i haven't solved before and you know i've got to work it out with my team or if i'm you know if I find myself my car breaks down I don't have a spare tire and I'm out of radio range and my phone doesn't work and then I figure out if I go to the top of the hill I might be I get you know I can think about my surroundings because obviously I don't want to be out in the middle of nowhere or if I've got a particular technical problem that um I'm trying to solve um which I think is a um um you know new to me that i have to from first principles build something new so i i think intelligence is a a very interesting place because it's so ill defined by a lot of people but for me intelligence is that thing that an entity displays to be able to take their current environment and solve a problem to maintain their fitness and to existence in there raw that's what and i would say that you know is biology intelligent well kinda but it depends about whether i'm intelligent in a darwinian sense so an intelligent microbe is more like a lucky microbe like if i try something and i don't die well hey but at some point i will die but an intelligent human being can imagine things before they do them um i'm in this workshop here i put lots of things like a build and 3d printers you know i've got a cathode ray gun in the back there i could put in but i probably would not attach myself to the electrodes because it would give me a 50 000 volt shock so i think intelligence is a is a very ill-posed um phenomena in our current um culture and it's partly done that way for reasons of commercial and competitive

5:37advancement and whereas i would like to say i want to measure it like the same thing of a talk about i want to measure something is from life or alive i think if we can measure intelligence and i'm not talking about turing test turing never meant his test to be a measure of intelligence or or whether consciousness right and then there's also this thing about was consciousness intelligence the same thing you know can unconscious entities be intelligent that's a very interesting question can um and so there's all these things that get nested together and ai has been a complete singularity of confusion for the term intelligence well before we get into the connection to consciousness i'd like to just probe this let's say problem-centric definition of intelligence a little bit and two examples that come to mind that i think might test this definition are one, maybe an autonomous vehicle, which does seem to respond to its environment.

6:41And it is, I don't know, maybe it's not an... The AI example du jour is the LLM, but I think that a self-driving car is also an exemplar of contemporary artificial intelligence. And then another example on the biological side of things It might be a slime mold, where my understanding is that this organism sends out probes and tendrils testing its environment. Then when it finds food, it reinforces these pathways. And one experiment that I believe was performed was a slime mold was placed in an array where food was arranged in, it was sort of mirroring the population density of Tokyo. And it sent out these pathways and extended itself in such a way that mirrored the Tokyo subway system.

7:43So this is a way a biological organism without a mind solved a problem and adapted to its environment. Yeah, I know that very well because I hired the postdoc that did that experiments in Tokyo, at least with such end of the group. And we did exactly the same thing with a chemical reaction. Okay, let's go. So let's do the slime mold first. Slime mold is a really cool one. The slime mold is a, it's a, if I get this right, it's a single cell multi-organiled object. So it can grow and contract and chop things off. But what it does, it chases a gradient. And so in the slime mold, it has sensors for, if you put acid, it would shrink away.

8:23And it had good stuff, it would go toward it. And it's kind of cool, right? The other way I would, if you've got a maze, and in the maze, you put a certain number of dots. And you say you put the slime mold, and those dots have got a food. that slime mold will basically put out sensors like like filaments and detect the food and it just takes the shortest path and it does that as a compromise and it can solve the shortest path much more uh better than a computation so it actually does it in the real world what's called an analog computation um and um that's kind of interesting now is that really intelligence or is the person who set up the slime mold with the maze knowing they want the slime mold to solve this.

9:07Because I can solve this another way, which is I could have an electrode at the entrance to my maze and three electrodes inside the maze. And I put a gas, like neon gas, like a neon lamp, and I increase the voltage between the electrodes until they get to about 50 ,000 volts. And then they will spark. and the sparks will solve the shortest path and the same the slime mold did. Now, which is more intelligent, the sparks or the slime mold? I would say that the slime mold is the more intelligent. Yeah, but they just did the same thing. The discharge, and there's another way I did this also. We're inspired by, so I actually did slime mold experiments.

9:53I actually took, a colleague of mine in Glasgow had this special sensor array and we grew slime molds on the sensor array. And this sensor array actually detects the amount of salt. And you could actually detect the salt coming out the slime mold. But we also did this with a chemical reaction, an oscillating chemical reaction, did exactly the same thing. So I would say the slime mold, although it's living, so obviously it's more autonomous, right? It's getting food and it's fulfilling a function. But that was set up by the human being who set up the experiment. so most of the intelligence came from the person who drew the maze put the slime mold in the maze and i've illustrated now with two things you could do it with a gas discharge to solve the maze that's that's a computation as well or i could uh use an oscillating chemical reaction called the blues of savaskinsky reaction which flashes red blue red blue red blue and it would just solve the maze as well as the waves of redness and blueness travel through the maze to find get connected.

10:55But I do agree the slime mold is living and more sophisticated. But actually, all the intelligence came from the person setting up the experiment. Interesting. So how does this, so how stepping back does this say that the slime mold fits your definition of intelligence? In a nutshell. so the slime mold actually doesn't fit my definition of intelligence as much as i would say with a human but the human interacting with the slime mold together they were you know they were the human designed the experiment positioned the slime mold and the slime mold took part in the experiment i guess where the slime mold was also showing some i don't know if i call it intelligence but survival instinct it wanted to survive so it is it is evolved to basically find food and that process of finding food in the environment is in essence the the mechanism by which i think intelligence gets built through natural selection evolution at the end of the day that those complex sensory inputs intelligence is the thing that gets produced by that i don't think the slime mold has any abstract intelligence it can't it doesn't have language it can't count right It can't basically do poetry, right, where humans can.

12:19And so that's a major difference. But it does have, according to your definition of intelligence, maybe a very, very narrow form of intelligence that is just meant for solving these very particular problems. Yeah, I would say intelligence like life is a scale, a gradual scale where you basically are able to do more and more things and if we did that intelligence could be a bit like temperature uh you can measure it i've been very inspired talking to sarah about her kind of uh sarah walker uh my collaborator uh about her uh forays into philosophy trying to understand how to build experiments and and and and the social interactions that go with that like we're kind of in the ai debate right now we're in the midst of this oh i've got an agi oh i've got an asi i've got this i've got this is conscious is you know and all all these things have been banded around because it's just one big narrative argument but actually what i would like to say back say hey what are we actually measuring what does it actually do what are you actually saying if you're going to sell me an artificial intelligence is it going to imagine what my problems are tomorrow and solve them for me or is it going to be actually what you mean.

13:35I'm just going to put a paragraph into a chatbot and it's going to basically probabilistically suggest other words based upon its training data and also the mechanical Turk data where I would be gratified by the response because if obviously these companies need to send you, give you outputs that you like, otherwise you won't keep paying them. So my next question then will be going to that second example with the autonomous vehicle. to what extent does it meet your definition or fail to meet your definition of intelligence yes so let's so i'm thinking about that so i think the autonomous vehicle displays no independent intelligence at all it the intelligence the autonomous vehicle is merely comes from the human i'm just going to try and think of a you know uh a mechanical version right Like there are mechanical objects that humans produce that will respond to different stimuli, and given the different stimuli, they'll do a different thing.

14:40Are they intelligent, or are they just mechanical automaton, where suddenly the temperature's hot, so you can go left, or the temperature's cold, and you go right? Now, the autonomous vehicle is a lot more complicated, because it's not just deterministic. There's non-deterministic elements or more probabilistic because the training set is like, you know, is there a pedestrian ahead of me? Is there a car coming to crash into me? You know, but ultimately the agency, the intent comes from the human. Now, what we should do when we think about the autonomous vehicle is there's lots of intents here.

15:19There's the intent of the driver who gets in the car. It turns the car on. there's the intent of the software developer who builds the software package there's the intent of the designer who built the car chassis all these intentions are trapped in this object but we should not miss um uh label the the the the fact the car looks like it has intention with the fact the human beings that built it did because if without the human beings and everything else there would have been no autonomous car and and then the car doesn't do anything that really merits for me going that's really intelligent it's doing something absolutely unexpected and solving a brand new problem because in autonomous car using artificial intelligence which i would just call autonomous informatics because what the car is doing is autonomously integrating the sensors deciding how to get you from a to b in the safest way according to certain policies that have been developed in conjunction with the manufacturer and the regulator and the customer and that for me is no more intelligent than uh you know reading a book about an autonomous car where the author of the book is the source of the intelligence but um i know a lot of people disagree with me but that's because um they somehow ascribe an agency to the car where the car has no feelings it has no intent it has no fear of death it has no evolutionary history other than the evolutionary history associated with the human beings that built that car heading in the same direction and just teasing these two cases apart a bit further why don't we just want to bite the bullet then and say that the human has no intelligence either but its intelligence is created or i was going to say is created by the sperm and egg that made it or by the human's parents and we get this sort of regress problem but what really distinguishes the two cases yeah i've i used to worry about that like where's the creativity come from like is human all human beings creativity just basically given at birth when that cell gets fertilized and that frozen accident those initial conditions are responsible for way the human acts um i don't think it's quite that but i think humans have a very interesting ability um number one they can come up with things they can come up with creative acts that we would we could not in principle predict number two um that that humans can in their pursuit of knowledge they have to be creative to really you know make a leap it's like this that we see this again and again and again and it's so interesting that we are blind in plain sight all the people telling us there's no there's no difference between ai and human intelligence there's a vast difference and then i think the other thing is this concept of novelty a friend of mine um she was a computer scientist at york she's retired now when i met her at artificial life meeting she really captivated my mind by saying look you know i'm a computer i'm a professor of computer science it is my dream to write a computer program that one day is so creative and so interesting i would feel that turning that computer off would be an act of murder i thought wow that's so cool like you really you want to be able to make a system that just generally keeps coming up with stuff like humans really are unique at that well biology is i would say that biology is incredibly creative humans take it to another level because we have language and tools and mental abstractions but it would appear to me that living systems do something that is not computational with respect to intelligence and there's to be clear i'm saying that our current conception of computation and our current conception of compute technology does not capture what humans and biology do in their creativity and one of the reasons why i'm colliding with the world on ai is because i'm interest in solving the origin of life and the origin of life and the origin of consciousness and the origin of intelligence are the same problem no life no consciousness no consciousness no high level intelligence and that for me is like really interesting and quite astonishing that people don't see this problem they you know when they say an llm is has agency or an artificial intelligence is just as conscious as you are and i'm like and you can literally see these bits being flipped in a silicon substrate and you're like that is not what we do and then i think is also interesting to say well hang on what is it the humans do or biology does that we have not yet captured in our technology because i'm not saying for one second that it's impossible to make a genuine artificial intelligence i'm just saying we have not made it yet i'm saying we have not made a genuine artificial intelligence and our current silicon technologies do cannot instantiate it they're very good at making mining the intelligence from the past they're brilliant so i would say that our artificial intelligence systems are really historians of human intelligence.

21:00And LLM is nothing more than a big historian, but with a bit more of a probabilistic mismatch, for it looks new to you.

21:11I want to continue pursuing this and also to get back to the definition of intelligence, which I still find very interesting and like to continue to pull apart. But since you've brought up the connection between life and intelligence. And our last conversation was very much focused on assembly theory and what makes life unique. I think just for our listeners who have not listened to that conversation, could you give the very brief elevator pitch of assembly theory and what makes life unique so that we can understand why assembly and intelligence would be connected? yeah that's a i'll try and do it very quickly so um we have a theory of um of um life or what makes light or how life emerges from physics based upon how causation works in in the world and causation is a really weird thing does a cause b does b cause a but basically i'm a chemist and And as I realized that looking at molecules, I could see that there were molecules in the world that couldn't exist without some technology or biology.

22:26And I worked backwards with my team and I figured that there is this process in the world where you have causation. That is, you have physical interactions that can become causal. So that means you can have random things bumping together. They were assemble, disassemble. Assembly theory is about that. And occasionally, the assembly of some things would be an accident, but a happy accident that had some effect on future processes. So then that becomes causal. Now, if those future processes can be repeated reliably, it then becomes contingent. It means there's a memory. So you have interaction, remember the interaction, and then use it for selection.

23:11and what assembly does assembly theory does it says hey how does sand turn into a cell by the process of causation contingency and selection and it's really about how do random things turn into living things and it and the way it attacks it is allows you to measure how living an artifact is by saying how many parts does that thing have how many unique parts and how many of those things do you have so it's like how complicated is the thing and how many of them do you have and those added together give you a measure of what's called assembly and we basically have determined that in laboratory that living systems uniquely produce objects or systems that have high assembly.

24:03That was quite a compact version. Lots of difficult words. What do we mean by causation? What do we mean by contingency? What do we mean by selection? It's hard, but basically what assembly theory does is says, I can tell the difference between living things and dead things because living things do non-random stuff at scale.

24:27okay and then the the connection to intelligence is simply that well it's it's not simple but you're never going to get something that is sufficiently complicated to adapt to its environment and solve a wide breadth of problems without out an assembly index that corresponds to life yeah i mean life has one sophistication high assembly index if you like a high assembly as well so the assembly equation has a function of assembly index and copy number and living systems produce are produce things with high assembly and themselves with high assembly now the way you can look at assembly theory in way is like it's just a way of basically saying oh if i go forward in time there's more that the number of possibilities increases dramatically and the living system seems to be able to increase that space of possibilities more than dead systems so so this is really important and i'm going to state this as clearly as possible and we can go back so non-living systems have a very small combinatorial space.

25:40Not much can happen. Living systems have a large combinatorial space, massive evolution number of protein structures. Intelligence systems have an even larger combinatorial space. Think of the number of ways you could land a rocket. I know that Blue Origin and SpaceX have now landed rockets. Well done, those companies. Think of the number of ways humans could design a technology to do a thing. In science fiction, in your imagination, intelligence seems to be associated with your ability to have an even bigger commentorial space of possibilities whether you're writing science fiction or you're producing a new product i mean i mean who came up with the idea of a labubu i mean i mean it's just you know it's that's the sign of intelligence i've ever saw one human creativity seems to know no bounds and that that creativity that novelty that newness that we are able to produce um seemingly from nowhere is is measured by that commentorial space and that space is bigger than the living one which is indeed indeed bigger than the dead one and so in a way if you want to be really just kind of matter of fact intelligence is just the universe trying to basically be aim for novelty aim for new things that have never been seen before in the universe And that is a very nice way to think about, demarcate our intelligence, biological intelligence, from algorithmic imposters of intelligence.

27:12Because all the generative AIs we see right now do not genuinely make new things. What they do is they have a data set. They have some probabilistic add-on. You can view it like a non-deterministic Turing machine, if you like. And it produces an output. And that output might be new for you. It might be new that it's been put on the page in that sequence, but it is entirely reducible to the past. And I just say that artificial intelligence is brilliant at characterizing the past. I cannot, in principle, predict the future because we don't know how the future, the present to the future unfolds, because this is what gives us intelligence and gives living systems autonomy.

27:55And it's kind of crazy. and my work on the origin of life is exposing this so it's not as if i'm some kind of luddite going you know oh ai is terrible or you know it's all horrible it's like no it's just it is not what it's being sold to be it's a very useful tool but it is not creating novelty it is creating new combinations of things that are in principle knowable from your data so again just pinning down this definition which i think is very important is on the one hand a general intelligence is something that can solve a wide breadth of problems but what you're also saying it is not just the behavior of the system that can tell us that tells us whether it's intelligent because if you just look at the behavior of ChatGPT and you have a conversation with it, it certainly appears to be intelligent.

28:57But we also need to know something else about either how it was created or its information processing that's interior. And I'm wondering if one if that's right and two if we can more explicitly pin down what the either whether it's the origin of the intelligence intelligent behavior or the information processing or how you would like to characterize that part of the definition yeah that's really important to think about precisely that definition because it's it's you know artificial intelligence is very good or ai system we have right now very good at taking large amounts of information allowing us to search them in a fairly lossy way you can't that that's it's it has its benefits and also has its flaws right so that's where hallucinations come from and so when we're trying to define um what that part of the intelligence is i think one of the things i find and i i can't give you the i can i think i can give you a mathematical definition of intelligence or i can work towards one which i think is going to work out quite nicely and because there's an extension of assembly theory right and it's kind of obvious right because obviously if intelligence is related to life and the origin of life and assembly theory is measuring the transition from dead stuff to life it's quite logical to impose it to keep going up um but trying to figure out how we pass that intelligence you know is chat gpt intelligent by my definition no but the human beings that produce the data and wrote the software and did all the stuff for intelligent it's a bit like going to a library full of books and saying are these books intelligent it's kind of a category error it's a category error to say is chat gpt intelligent i think the correct question is was chat gpt produced by intelligence and the answer is yes and and because what's happening is a sleight of hand and i don't know why well i have a guess it seems to be the creators of these um of these uh systems want you to think that they are intelligent or have sentience even right why is it that these creators would want you to think that their technology is intelligent and has sentience.

31:40The only thing I can think of is that basically they realize there's a copyright problem. And because they've taken a lot of data from the environment and they've done machine learning on it, they've extracted those relationships, and then you can serve it back. It seems that because they haven't paid for that copyright, if you can pretend that the thing has its own desires and is sentient, it's just like a human being. you know coming up with a new idea and as long as there's not plagiarism there is some bar to go through so i wonder if it's something like that but it just it seems to me a little bit odd that we are um you know purposefully misusing the word and then it gets even crazier when we talk about artificial intelligence number one which is wrongly named should be called autonomous informatics number two artificial general intelligence i don't even know what that is and let's go even more crazy super intelligence well if we don't even know what intelligence is and we're supering it you know i'm very much of the the david deutsch who i you know i think david deutsch is wrong about many things but he is a fantastic thinker and he is right about many things and particularly this understanding of how universality you know works in our science this is the idea that we can build abstractions they're universal and we can you know we can think about our number system so sure super intelligence might be a faster thinker than me but it can't think things in principle i cannot think and i think that that is a very big category error again there's this category errors all the way right where people were saying chat gpt is intelligent no chat that's that's not correct chat gpt was produced by intelligence then we have super intelligence category error we don't know how can you get better than turing machines and abstractions and numbers and language is there some magic i don't have access to is there some kind of interdimensional super intelligent alien that is moving information around you are then moving into the mystical and the religious.

33:51And the reason why definitions are useful is it allows us to weed out the people who are just literally making stuff up. Well, I'm going to continue using the term artificial intelligence instead of autonomous informatics, at least for the purposes of the conversation and for our listeners. But I really like the example of asking whether a book is intelligent and pointing out that this is a category error. But it makes me wonder then, since artificial intelligence, as we've been discussing it, is almost by definition created, why you think, and you said this earlier, you said that we don't have any genuinely intelligent artificial intelligences, but you think we could in principle have a genuine artificial intelligence.

34:44And I'm wondering why that would be or what the distinction would be since we created it, its intelligence would come from us. Yeah, there's a lot there. I would, if I was able to create a new life form from scratch in the lab or on Mars or somewhere, right, like a Genesis engine, and I could accelerate the chemistry to get to a life form, presumably i could generate accelerate the the evolution of that life form production of an ecosystem competition collaboration and that and that kind of process would give rise over time uh to the possibility for object for living new artificial life forms of intelligence now you're correct you i then set that up and i basically build that intelligence so is it still intelligent well yes because we use the process of evolution and problem solving the environment to regenerate the phenomena and so so that we're seeing right i mean and that is different to saying chat gpt is intelligent because or book is intelligent that is a category error if i made a new life form in my lab people could legitimately say well that's great well done but that doesn't mean it could happen on earth right where was the lee cronin at the origin of life equally if i made a new intelligence i if i claimed that was like you know this is the proof that intelligence could occur in the universe universally they say yeah but you created it but who created your intelligence so it'd be nice to be able to see mechanistically from a scientific point of view how intelligence could emerge in the natural world and i'm very excited by that because there's the easiest problem so we've got four problems it's like origin of life which is number one then basically uh then you've got kind of uh kind of in consciousness abstraction and intelligence i think intelligence is the hardest of the four problems the origin of life is the easy one because we're still kind of we're mired in all this kind of controversy of what we've got but going to your point i because i'm a physicalist i believe in reality is that what we can see and interrogate i have to accept that if i accept that you're intelligent human being and i'm an intelligent human being that should be possible at least in principle to figure out how to make another intelligent human being i've done it the hard way i have two teenage sons they're quite smart and they they have novelty creativity and all the stuff but i mean you know a slightly less uh kind of biological way like how do we what is it about the human brain and biology in general that allows biology to be infinitely creative and i was very inspired i don't know if you know the philosopher henry bergson henry bergson wrote three books that blew my mind that i've either um on creative evolution um and uh time and and and he just basically this philosopher he had an argument of einstein actually a famous argument about time where everyone thought that einstein won actually if you look at it really ornery one but there is a deep relationship between assembly theory time evolution and intelligence and and then i think that the the in computer scientists have been so phenomenally successful building infrastructure uh with electrical engineers and software and stuff that we love and use and interact with and i'm indeed addicted to that we have lost the correct philosophical approaches to understand what are we calling these things and what is our relationship with these things and you know i keep going on but I'll just give you one that troubles me really deeply if I may um which is about um human labor and artificial intelligence everyone's saying you know you know and I'm getting really fed up with this because it's actually insulting to human beings yes there's a lot of ridiculous things I do and yes they could be automated but you know not it's harder than it looks you've got all these people out there going oh you know actually your job is going to be taken over by AI and agents I'm like, okay, fine.

39:18Why is that? They say, well, the AI is intelligent. And that is a basically fraudulent statement to make when they know that, right? Because I need my critical thinking to basically do things, produce novelty. Yes, I may not think as much as I would like to think, I think. You know, I'm not as creative as I think I am and all that stuff. And the AI tools will, you know, close the gap. But this idea that AI will replace human labor is kind of insane. It's kind of insane, and it's typified by the notion of AI scientists. When someone says, I'm going to make an AI scientist, I'm just like, that's a bit like saying, I've got this machine, I'm going to take a drop of blood, I'm going to detect all your diseases is in the drop of blood but really i need to take this much blood because when someone says i can do ai science i'm like what you mean automating scientific discovery ah okay great how what is artificial intelligence what is discovery correctly position those and give me a correct approach for defining those then we can talk about it and what they actually mean is i want to use an llm to basically replace google for searching some some literature so it's kind of weird that we're in this situation that we that ai seems to be seen as taking everyone's jobs taking on everyone's creativity doing the science doing everything else where we don't even know why human beings are intelligent and we're giving away all of that to technology that it's kind of insane it's like it this is what it must have been like when the printing press was just developed and everyone was given a bible to read and said that's the only book you're allowed to read welcome to the new technology.

41:11Well, while we're on this topic, I actually found a tweet of yours that I wanted to ask about. So let me just pull it up very quickly. Okay. And the tweet is, when people say you should learn AI skills, that's actually nonsense. It's a product placement exercise. AI skills are not really hard skills. They're indoctrination to the app. Instead, learn to think critically and precisely. And this, I mean, reminds me of the interviews that I had with Tyler Cohen and Reid Hoffman that I think you have listened to because they very much are advocating that people become quote unquote AI natives. Or I think that what, I think that's Reid's phrase.

42:01And he says that younger people today are AI natives and that's a strength. And And Tyler Cohen teaches his students to use AI better because he thinks that these are skills that you're going to need for the future. So I'm curious about your tweet and where that fits in. So I know both Reid and Tyler personally. I love them a lot. They're lovely human beings and they want to do good in the world. So that's good. I don't agree with them, but it's nuanced, right? Tyler is an incredibly deep thinker, but he's very utilitarian. You know, it's really interesting listening to your podcast. I was listening to your podcast with him.

42:37And he was just like very matter of fact practical. But actually, Tyler assumes the AIs can do things from that podcast they actually cannot do. And he's just taken the people that he's told them they can do at good faith. And so I think Tyler's been brilliant in working good faith. If these devices are good at, you know, education, why not use them? Well, because they do not have a ground truth and they hallucinate and they serve up random stuff, you can't teach somebody with an AI teacher because you're not going to build a first principles ground truth of the world, right? And I think that actually I can't blame Tyler for not saying that, right?

43:22I've been thinking deeply about this since I was like nine years old. and so if you take what Tyler was saying at face value is absolutely right these are great tools we should use them it let if they are able to teach us you know be faster at learning stuff you know than uh or or maybe reduce the reliance on the college professor to get information you absolutely must use them and I and indeed I think they will come now and I think Reed is just an optimist because he's a computer scientist and he's made an incredible technology and of course he wants to build more technology that are good for human beings um but i think they're both quite wrong on ai but not terribly like is the end of the world wrong i just think i love their optimistic stunt i would love them to keep demanding that the technologies live up to their expectations and i'm actually quite confident with people like tyler and reed in the world pushing this the AIs will get better right your products will get better but yeah I do think when people say oh you've got to learn to use AIs like what is this like I I write AIs in my lab right I I write mathematical models and I actually physically build my own neural networks and my groups say I'm completely stupid to do that because I could use all these infrastructure that's been built by all the big companies I'm like but I want to know how it works I want to know if I could have some kind of symbolic AI that's rather than having then always having a neural system everywhere right and I want to understand how to process that data I want to get into the knits and you know down to the nuts and bolts so when I imagine people say you learn to use the tools about cool yeah learn to do the coding learn to process the data learn to normalize the data learn to figure out what type of architects you need to basically propagate the information through the neural network what other techniques could you use other than transformable model how many has tension heads blah blah blah blah blah but when they say everyone's saying learn how to use the the tools they're like learn how to prompt chat gpt that's not a skill that's the product placement thing and and i mean it's all it is right that that is interesting i mean you might want well i i i use chan gbt i often have conversations with it about various topics so i've i've lately been curious about this legal principle called stare decisis and so i've been asking chan gbt about various dimensions of it but maybe i'm not using it for a sophisticated enough purpose but it just doesn't seem like there is that much skill to it so i i'm i'm sympathetic to your idea that ai skills are not really hard skills in the way that learning to to code or learning calculus are hard skills and they're changing the model so rapidly and changing the guardrails because of course chat gpt can be misused and and i think that it's changing a bit and of course i guess there must be a way of you know i i like to play with chat gpt to look at its training set like you know you know if i want to understand how to say a text was generated by chat gpt you could type in chat gpt you know the the big brown cat sat on the and it was saying that every time right it would never deviate from that but if you put the skinny purple cat sat on the it might say rainbow window still tree whatever and just understanding how they work but of course they're much more sophisticated it isn't now just predicting the next word it is building these symbolic architectures that can at least superficially have some um content sorry context um and you know i think if that gets better the the distance between real intelligence and artificial intelligence will decrease decrease decrease to the point where we're like actually i'm okay with it but i just hope that we don't get to the point where we forget to worry about what intelligence actually is and when it comes to the use of these tools right now there's a massive fight going on there is not going to be all these companies alive in four years time right but they just won't have the money they're all losing money right um and so i wonder if it's a bit like the browser wars when in the time you know you're going to use netscape are you going to use explorer are you going to use chrome or whatever it was at the time sorry firefox um and you know it's just a tool and i i think that ai native strategy is is really just people just saying yeah you should use the tools or otherwise you can be left out what i think is super important is the fact that you've probably seen this report out of mit with a cognition deficit where if you basically let people use ai to basically write essays or or come up with things that they actually lose the ability to think because it becomes too hard for them to think it's almost a bit like running a marathon and losing the ability to have enough blood oxygen saturation so you can get you can go fast and that worries me a great deal i hadn't seen that paper how did do you recall off the top of your head how they measured it or determined this conclusion um yeah i think that so i can get send it to you um but basically if i recall correctly they had two groups of people and one group they said you can't use gen ai to solve this problem but you can use all the normal search methods and the ones that you are such a use gen ai to basically write the essay and so on they found that basically and then you then kind of mixed them up you then took the group that had used the gen ai and then said right now you've got to do the do it without gen ai and vice versa you found that people use gen ai to basically write essays and solve problems when they were told they couldn't use it performed badly they lost cognitive ability from this and that was really quite alarming it's so and i don't know if it's been reproduced but it was a kind of interesting study but i can send you the link i think it came out of one of the labs at mit just recently one dimension of intelligence that we haven't at least well we've we've danced around we haven't used this word, is agency.

49:54And I'm wondering where you see agency fitting into intelligence, because you've also said that AIs can't actually do anything, the AIs that we have today, whereas I think AI companies would have us think otherwise by naming, let's say, like, chat GPT agent mode, and having chat GPT do research for us. So where is agency in this equation where where isn't it in practice at this point yeah i a few months ago i was at a meeting in the uk at the royal society and um we're talking about ai and agency and when i i was at a table computer scientists and i was just explaining how you know agency is it's weird that people think that anything has agency outside of a human and they were all aghast they were like what are you talking about of course it has agency i write agents all the time i was like what are you there's So there's kind of this misuse of the word.

50:51So I like the word agency a lot. Agency, basically, if I'm a proper agent in the world, I have certain rights afforded to me by, I don't know, the world. Some human rights, like bank account, birth certificate, legal status, right? These things are kind of given to me. Why? Well, I'm a biological entity. You know, I can be murdered. I can be a murderer. I could be, you know, I interact in a way in the world because I presumably have some free will. And that free will is not easily discernible from, if you come to me and wonder what's going in my head, it's very hard to figure out what I'm thinking.

51:40And so human beings have this agency. Animals have agency as well. humans have more agency it would appear because they have more ability to affect change in the world and that agency comes from birth right from conception to walking around in the world your education the social interactions and that myriad of things you go through right growing up and go on every day and that agency is grown not programmed and ultimately all these social layers we have our agency does one thing is like please don't die it's like biology is like say please compete procreate don't die and that's where i think agency comes from again goes all the way back to the origin of life now when you say oh no chat gpt has agency and i'm like well okay the programmer that built chat gpt had agency and they wrote some code to basically give an output and all that agency is outsourced from other human beings so i guess the right thing the thing i could basically push everyone is to say what is the origin the original prime mover in your universe the agency where does that come from um um that and that is really important to get right because these ai devices do not have agency they merely have an illusion of agency or shadow of agency from the creators and the technology of the technology

53:11does agency at all have to do with physical embodiment for you you referenced computationalism well i don't think you used that word but you you implied it and i'm wondering if agency by your definition i mean if it requires a chemical or physical embodiment that computation alone can't provide yeah i think so also there's something else going on i mean if i was i would say what what is it about human biology that computers cannot yet do is that biology seems to live and this is going to get again this is just an idea i had this week biology is this thing that exists in a universe that's continuous or let's say undetermined but the act of um living you requires you have to label things so you can error correct so biology is desperately taking stuff from this continuous space and it you know encoding it so it can be error corrected but that encoding that labeling makes it quite artificial and i think that novelty in the world comes from this undetermined space or continuous space as it interfaces with a error corrected discrete space and the fact that our computational technology is built upon pre-labeled things and a finite commentorial space means they can't in principle mine the novelty from the universe biological systems are the only systems we know that can mine novelty from the future if you like and express it in the present which seems like a weird thing to say but because the the present is thick in terms of things i can do right now and i can imagine counterfactuals and just pull something from the future and do it now is really cool and ai systems can't do that and it's going to become more and more obvious that there is a missing phenomena the missing physics and but that allows biology to work is is presenting from ai but really being ai with agency and it's because of these issues um because computers are built on a von neumann paradigm ones and zeros finite commentorial space finite memory um everything is everything is digitized so you don't get this and this is where the embodiment comes in because if you're in if you're embodied everything is digitized you're in the continuous world there could be water there could be rain there could be sunshine there could be mud there could be whatever right you know other people things and that kind of interaction that embodiment just gives you such a richer problem space in which to exist it's so commentorily vast you cannot search it in finite time which means you have to you have to take a leap of faith is what biology does all the time and the computer science technologies or computers they don't take a leap of faith they take a probabilistic leap of organization well i mean there there are always going to be more topics connected to ai that we could talk about so we really didn't touch much on the connection between consciousness and intelligence and another and we talked a bit about this last time is existential threats and how ai will or won't change our lives going forward but i'm wondering if there are any of those topics that you would like to touch on, or if we should switch over now to aliens?

56:50I think we can switch to aliens. I would just say on the consciousness thing, I think consciousness is a very interesting problem like intelligence. And I think that we're going to find at that some degree of the thing that we call consciousness is probably wrapped up with the thing we call intelligence and an agency. And I am uniquely unqualified i'd like to build one so i could study it right um but uh yeah i think we can switch over from there there's a lot i think we've done a we've covered a lot of ai and i think i've tried to be as precise as possible because i don't want to offend the community use ai and love it and making money out of it and it's all great future that's great but what i want us to do as humans is be critical about what it can do about what it can't do and what our definitions are so we can use it to our best abilities.

57:41So yeah, aliens it is. Okay, well, great. I think a good starting place would be to ask about the connection between life in general and biology, because I'm wondering if whether all life, all alien life will be biological, or if this is just a tautology, or if there's something more to it? So if you mean by biological, like biology on Earth, like RNA, DNA, proteins, then I think it's almost certain that alien life will not be biological or not use the current chemistry that we have. I think there's a good argument to say, well, we have so much contingency, so much complexity in our chemistry of biology and chemical space is so big it would be very very weird if alien life used exactly the same chemistry as in biology so i think it will be very different i don't know how different because you think about it you might say well look let's just say that let's just think about whether life would be easier on a rocky planet with some water and maybe some atmosphere and and that maybe then closes down the space a bit so the aliens might be squidgy like us but maybe not using the same type of chemical um um uh constituents but but similar so i so i think you could so alien biology will certainly i think alien biology will involve chemistry as we as we could imagine it like you know chemical bonds but I don't think I can say much more which is super exciting.

59:30Are you able to say based on your work on Assembly how common you think alien life will be just like I said given what you know about Assembly and then the overall composition of the universe? Yes Sarah and I have completely different views on this kind of funny in a way she says oh i don't think there's i don't i'm not at all confidence there will be alien life fails for any universe and i'm like of course i'm gonna take a more nuanced view right now assembly theory basically uh how ultimately should allow us to become comfortable measuring the phenomena of selection or assembly before biology that is when stuff aggregates in without biology and and you get causation contingency and selection that that happens almost like before we get the first cell and i think if anyone was thinking about this carefully enough no one should basically say oh yeah the cell the first cell just randomly formed one day from nowhere like a miracle that would be impossible so there had to be some process of selection before that now that process the selection requires matter and heterogeneity and because if every star you look up in the sky looks like it has um planets around it now and these heavy planets have you know heavy elements in them and there's carbon around and nitrogen oxygen then i think that there'll be selection going on in the entire universe because what you need for selection you need heterogeneity you need some chemistry you need some energy and you need some time so if those things allow you to basically undergo selection then it's just a question of how how long or how lucky is it that selection turns into a type of life and um and i think that's a very important question i i am not a betting person but i would wager that there is a there are many many trillions no trillions is too much billions of life types of life in the universe with very simple life and even more kind of selection processes going on that are pre-life and i i'm pretty comfortable with saying that but to be super clear i have no evidence to say that other than the stuff we've done in the lab where we can see selection in inorganic systems happening on a reasonably reasonable time scale and that makes me optimistic about the fact there will be the process of selection producing life in the universe so that that's where i think that the gap between sarah and i have no aliens and loads of aliens can be closed because the phenomena selection must be going on everywhere today and is your response to the fermi paradox then that even though you believe there could be loads of life out there it's just exceedingly rare that it's intelligent um maybe i just don't know the fermi paradox is a paradox i just think it's a very poor experiment thought experiment but let's just assume it's a paradox yeah could you sketch it out for our listeners yeah so basically fermi said like if there was life everywhere in universe then we should literally see times um evidence of life radio waves and you know the messages and and other things and he said that this is a paradox we don't see it i mean we're basically we seem to be quite alone in the in the i don't know 70 or 80 years from which we've been looking into space and had enough powerful transmitters to send radio waves and big dishes to receive it we haven't heard anything except maybe the wow signal and even then we don't know what that was but my response to that so that's the thermo paradox my response to that is like everything it's a bit like the same is our is our anthropocentric anthropocentricity i said it it's like people think that maybe rna on earth is the way you get to life everywhere in the universe and i'm finishing a paper with with uh sarah and two other colleagues which basically explains how rare rna is going to be in the world and so they act to get that exact thing is unlikely now this takes the thermi paradox so thermi's assuming that people will get to radio waves well people we we've stopped actually broadcasting such high energy radio waves in the space we're using you know um uh smaller and smaller wavelengths microwave right wi-fi gigahertz terahertz lasers and so the window which we were sending radio waves into space was not that many years like an infinitesimal number of years compared to the number of years of the universe so that's number one number two the universe is expanding rapidly so everyone's getting further away um and i think the reason we haven't seen it yet is that we don't know what to look for And until we start thinking about complexity in terms of assembly, I don't think we will even know if the background noise is a message or not, because our notions of randomness and complexity are pretty messed up.