Eliezer Yudkowsky — Why AI will kill us, aligning LLMs, nature of intelligence, SciFi, and rationality

No, no, misaligned! Misaligned! No, no, no. Not yet. Not now. Nobody's being careful and deliberate now. But maybe at some point in the indefinite future, people will be careful and deliberate. Sure. Let's grant that premise. Keep going. If you try to browse your planet, there are the idiot disaster monkeys. Who are like, ooh, ooh. Like, if this is dangerous, it must be powerful, right? I'm going to be first to grab the poison banana. And it's not a coincidence that I can, like, zoom in and poke at this and ask questions like this. And that you did not ask these questions of yourself. You are imagining nice ways you can get the thing. But reality is not necessarily imagining how to give you what you want. Should one remain silent? Should one let everyone walk directly into the whirling razor blades? Like, continuing to play out a video game you know you're going to lose. Because that's all you have. Okay. Today I have the pleasure of speaking with Eliezer Yudkowsky. Eliezer, thank you so much for coming out to the Lunar Society. You're welcome. First question. So yesterday when we were recording this, you had an article in Time calling for a moratorium on further AI training runs. Now, my first question is, it's probably not likely that governments are going to adopt some sort of treaty that restricts AI right now. So what was the goal with writing it right now? I think that I thought that this was something very unlikely for governments to adopt. And then all of my friends kept on telling me, like, no, no. Actually, if you talk to anyone outside of the tech industry, they think maybe we shouldn't do that. I was like, all right, then. Like, I assumed that this concept had no popular support. Maybe I assumed incorrectly. It seems foolish and to lack dignity to not even try to say what ought to be done. There wasn't a galaxy-brained purpose behind it. I think that over the last 22 years or so, we've seen a great lack of galaxy-brained ideas playing out successfully. Has anybody in government, not necessarily after the article, but just in general, have they reached out to you in a way that makes you think that they sort of have the broad contours of the problem, correct? No, I'm going on reports that normal people are more willing than the people I've been previously talking to to entertain calls. This is a bad idea. Maybe you should just not do that. That's surprising to hear because I would have assumed that the people in Silicon Valley who are weirdos would be more likely to find this sort of message. They could kind of rocket the whole idea that nanomachines will, AIs will make nanomachines that take over. It's surprising to hear that normal people got the message first. Well, I hesitate to use the term midwit, but maybe this was all just a midwit thing. All right. So my concern with, I guess, either the six-month moratorium or a forever moratorium until we solve alignment is that at this point, it seems like it could, to people, seem like we're crying wolf. And actually, not that it could, but it would be like crying wolf because these systems aren't yet at a point at which they're dangerous. And nobody is saying they are. Well, I'm not saying they are. The open letter signatories aren't saying they are, I don't think. So if there is a point at which we can sort of get the public momentum to do some sort of stop, wouldn't it be useful to exercise it when we get to GPT-6 and who knows what it's capable of? Well, why do it now? Because allegedly, possibly, and we will see, people right now are able to appreciate that things are storming ahead and a bit faster than the ability to, well, ensure any sort of good outcome for them. And, you know, you could be like, ah, yes, well, we will play the galaxy brain, clever political move of trying to time when the popular support will be there. But, again, I heard rumors that people were actually like completely open to the concept of let's stop. So, again, just trying to say it. And it's not clear to me what happens if we wait for GPT-5 to say it. I don't actually know what GPT-5 is going to be like. It has been very hard to call the rate at which these systems acquire capability as they are trained to larger and larger sizes and more and more tokens. And, like, GPT-4 is a bit beyond in some ways where I thought this paradigm was going to scale, period. So I don't actually know what happens if GPT-5 is built. And even if GPT-5 doesn't end the world, which I agree is like more than 50% of where my probability mass lies, even if GPT-5 doesn't end the world, maybe that's enough time for GPT-4.5 to get ensconced everywhere and in everything and for it actually to be harder to call a stop, both politically and technically. There's also the point that training algorithms keep improving. If we put a hard limit on the total computes and training runs right now, these systems would still get more capable over time as the algorithms improved and got more efficient, like more oomph per floating point operation. And things would still improve, but slower. And if you start that process off at the GPT-5 level, where I don't actually know how capable that is exactly, you may have like a bunch less lifeline left before you get into dangerous territory. The concern is then that, listen, there's millions of GPUs out there in the world. And so the actors who would be willing to cooperate or who could even identify in order to even get the government to make them cooperate would be potentially the ones that are most on the message. And so what you're left with is a system where they stagnate for six months or a year or however long this lasts. And then what is the game plan? Is there some plan by which if we wait a few years, then alignment will be solved? Do we have some sort of timeline like that? Well, alignment will not be solved in a few years. I would hope for something along the lines of human intelligence enhancement works. I do not think we are going to have the timeline for genetically engineering humans to work, but maybe this is why I mentioned in the time letter that if I had like infinite capability to dictate laws, that there would be a carve out on biology, like AI that is like just for biology and not trained on text from the Internet. Human intelligence enhancement, make people smarter. Making people smarter has a chance of going right in a way that making an extremely smart AI does not have a realistic chance of going right at this point. So, yeah, that would, in terms of like remotely, you know, how do I put it? If we were on a sane planet, what the sane planet does at this point is shut it all down and work on human intelligence enhancement. It is, I don't think we're going to live in that sane world. I think we are all going to die. But having heard that people are more open to this outside of California, it makes sense to me to just like try saying out loud what it is that you do in a saner planet and not just assume that people are not going to do that. In what percentage of the world where humanity survives, is there human enhancement? Like even if there's 1% chance humanity survives, is basically that entire branch dominated by the world where there's some sort of... I mean, I think we're just like mainly in the territory of hail Mary passes at this point. And human intelligence enhancement is one hail Mary pass. Maybe you can put people in MRIs and train them using neurofeedback to be a little saner, to not rationalize so much. Maybe you can figure out how to have something light up every time somebody is like working backwards from what they want to be true to what they take as their premises. Maybe you can just like fire off little lights and teach people not to do that so much. Maybe the GPT-4 level systems can be reinforcement learning from human feedback into being consistently smart, nice, and charitable in conversation and just unleash a billion of them on Twitter and just have them like spread sanity everywhere. I do not think this... I do worry that this is like not going to be the most profitable use of the technology. But, you know, you're asking me to list out hail Mary passes, that's what I'm doing. Maybe you can actually figure out how to take a brain, slice it, scan it, simulate it, run uploads and upgrade the uploads or run the uploads faster. These are also quite dangerous things, but they do not have the utter lethality of artificial intelligence. Alright, that's actually a great jumping point into the next topic I want to talk to you about, orthogonality. And here's my first question, speaking of human enhancement. Suppose you bred human beings to be friendly and cooperative, but also more intelligent. I'm sure you're going to disagree with this analogy, but I just want to understand why. I claim that over many generations, you would just have really smart humans who are also really friendly and cooperative. Would you disagree with that or would you disagree with the analogy? So the main thing is that you're starting from minds that are already very, very similar to yours. You're starting from minds of which whom many of them already exhibit the characteristics that you want. There are already many people in the world, I hope, who are nice in the way that you want them to be nice. Of course it depends on how nice you want exactly. I think that if you like actually go for it.
We'll start trying to run a project of selectively encouraging some marriages between particular people and encouraging them to have children. You will rapidly find, as one does in any process of, as one does when one does this to, say, chickens, that when you select on the stuff you want, it turns out there's a bunch of stuff correlated with it and that you're not changing just one thing. If you try to make people who are inhumanly nice, who are nicer than anyone has ever been before, you're going outside the space that human psychology has previously evolved and adapted to deal with and weird stuff will happen to those people. None of this is very analogous to AI. I'm just pointing out something along the lines of, well, taking your analogy at face value, what would happen exactly? And, you know, it's the sort of thing where you could maybe do it, but there's all kinds of pitfalls that you'd probably find out about if you cracked open a textbook on animal breeding. So, I mean, the thing you mentioned initially, which is that we are starting off with basic humanist psychology that we're kind of fine tuning with breeding. Luckily, the current paradigm of AI is, you know, you just have these models that are trained on human text. And, I mean, you would assume that this would give you a sort of starting point of something like human psychology. Why do you assume that? Because they're trained on human text. And what does that do? Whatever sorts of thoughts and emotions that lead to the production of human text need to be simulated in the AI in order to produce those themselves. I see. So, like, if you take a person and, like, if you take an actor and tell them to play a character, they just, like, become that person. You can tell that, because, you know, like, you see somebody on screen playing Buffy the Vampire Slayer. And, you know, that's probably just actually Buffy in there. That's who that is. I think a better analogy is if you have a child and you tell him, hey, be this way, they're more likely to just be that way. I mean, other than, like, putting on an act for, like, 20 years or something. Depends on what you're telling them to be, exactly. Like, if you're just saying, be nice. Yeah, but that's not what you're telling them to do. You're telling them to play the part of an alien. Like, something with a completely inhuman psychology, as extrapolated by science fiction authors and, in many cases, you know, like, done by computers. Because, you know, humans can't quite think that way. And your child eventually manages to learn to act that way. What exactly is going on in there now? Are they just the alien? Or did they pick up the rhythm of what you were asking them to imitate and be like, ah, yes, I see who I'm supposed to pretend to be. Are they actually a person or are they pretending? That's true even if you're not asking them to be an alien. You know, my parents tried to raise me Orthodox Jewish and that did not take at all. I learned to pretend. I learned to comply. I hated every minute of it. Okay, not literally every minute of it. I should avoid saying untrue things. I hated most minutes of it. And yeah, like, because they were trying to show me a way to be that was alien to my own psychology. And the religion that I actually picked up was from the science fiction books instead, as it were. Though I'm using religion very metaphorically here. More like ethos, you might say. I was raised with the science fiction books I was reading from my parents' library and Orthodox Judaism. And the ethos of the science fiction books rang truer in my soul. And so that took in the Orthodox Judaism didn't. But the Orthodox Judaism was what I had to imitate, was what I had to pretend to be, was the answers I had to give, whether I believe them or not, because otherwise you get punished. But I mean, on that point itself, the rates of apostasy are probably below 50% in any religion, right? Like some people do leave, but often they just become the thing they're imitating as a child. Yes, because the religions are selected to not have that many apostates. If aliens came in and introduced their religion, you'd get a lot more apostates. But I mean, I think we're probably in a more virtuous situation with ML because these systems are kind of through stochastic gradient descent sort of regularized so that the system that is pretending to be something where there's like multiple layers of interpretation is gonna be more complex than the one that it's just being the thing. And I mean, over time, the system that is just being the thing will be optimized, right? It'll just be simpler. This seems like an ordinate cope. For one thing, you're not training it to be any one particular person. You're training it to switch masks to anyone on the internet as soon as they figure out who that person on the internet is. If I put the internet in front of you and I was like, learn to predict the next word, learn to predict the next word over and over, you do not just like turn into a random human because the random human is not what's best at predicting the next word of everyone who's ever been on the internet. You learn to very rapidly like pick up on the cues of like what sort of person is talking? What will they say next? You memorize so many facts that just because they're helpful in predicting the next word. You learn all kinds of patterns. You learn all the languages. You learn to switch rapidly from being one kind of person or another as the conversation that you are predicting changes who's speaking. This is not a human we're describing. You are not training a human there. Would you at least say that we are living in a better situation than one in which we have some sort of black box where you have this sort of Machiavellian fit to survive a simulation that produces AI? I guess this situation is at least more likely to produce alignment than one in which something that is completely untouched by human psychology would produce? More likely, yes. Maybe you're like, it's an order of magnitude likelier, 0% instead of 0%. Getting stuff like more likely does not help you if the baseline is like nearly zero. Like the whole training setup there is producing an actress, a predictor. It's not actually being put into the kind of ancestral situation that evolved humans, nor the kind of modern situation that raises humans, though to be clear, raising it like a human wouldn't help. But like, yeah, you're like giving it a very alien problem that is not what humans solve. And it is like solving that problem, not the way a human would. Okay, so how about this? I can see that I certainly don't know for sure what is going on in these systems. In fact, obviously nobody does. But that also goes for you. So could it not just be that even through imitating all humans, it like, I don't know, reinforcement learning works, and then all these other things we're trying somehow work, and actually just like being an actor produces some sort of benign outcome where there isn't that level of simulation and conniving? I think it predictably breaks down as you try to make the system smarter, as you try to derive sufficiently useful work from it. And in particular, like the sort of work where some other AI doesn't just kill you off six months later. Yeah, like, I think the present system is not smart enough to have a deep conniving actress thinking long strings of coherent thoughts about how to predict the next word. But as the mask that it wears, as the people it's pretending to be gets smarter and smarter, I think that at some point the thing in there that is predicting how humans plan, predicting how humans talk, predicting how humans think, and needing to be at least as smart as the human it is predicting in order to do that, I suspect at some point there is a new coherence born within the system and something strange starts happening. I think that if you have something that can accurately predict, I mean, Eliezer Yudkowsky, to use a particular example I know quite well, I think that to accurately predict Eliezer Yudkowsky, you've got to be able to do the kind of thinking where you are reflecting on yourself, and that in order to simulate Eliezer Yudkowsky reflecting on himself, you need to be able to do that kind of thinking. And this is not airtight logic, but I expect there to be a discount factor. So if you ask me to play a part of somebody who's quite unlike me, I think there's some amount of penalty that the character I'm playing gets to his intelligence because I'm secretly back there simulating him, and that's even if we're quite similar, and the stranger they are, the more unfamiliar the situation, the less the person I'm playing is as smart as I am, the more they are dumber than I am. So similarly, I think that if you get an AI that's very, very good at predicting what Eliezer says, I think that there's a quite alien mind doing that, that it actually has to be to some degree smarter than me in order to play the role of something that thinks differently from how it does very, very accurately. And I reflect on myself. I think about how my thoughts are.
not good enough by my own standards and how I want to rearrange my own thought processes. I look at the world and see it going the way I did not want it to go and asking myself, how could I change this world? I look around at other humans, I model them, and sometimes I try to persuade them of things. These are all capabilities that the system would then be somewhere in there, and I just, like, don't trust the lot, I don't trust the blind hope that all of that capability is pointed entirely at pretending to be Eliezer and only exists insofar as it's like the mirror and isomorph of Eliezer, that all the prediction is like, is by being something exactly like me and not thinking about me while not being me. Certainly, I don't want to claim that it is guaranteed that there isn't something super alien and something that is against our aims happening within the Shoggoth, but you made an earlier claim which seemed much stronger than the idea that you don't want blind hope, which is that we're going from zero percent probability to an order of magnitude greater at zero percent probability. There's a difference between saying that we should be wary and that there's no hope, right? I could imagine so many things that could be happening in the Shoggoth's brain, especially in our level of confusion and mysticism over what is happening. So, I mean, okay, so one example is, like, I don't know, let's say that it kind of just becomes the average of all human psychology and motives while still maintaining. But it's not the average. It is able to be every one of those people. Right, right. That's very different from being the average, right? Like, it's very different from being an average chess player versus being able to predict every chess player in the database. These are very different things. Yeah, no, I meant in terms of motives, it is the average, whereas it can simulate any given human. Well, why would the... I'm not saying that's the most likely one. I'm just saying, like... This just seems zero percent probable to me. Like, the motive is going to be like, I want to, like, insofar... The motive is going to be like some weird funhouse mirror thing of, I want to predict very accurately. Right. Why, then, are we so sure that whatever the drives that come about because of this motive are going to be incompatible with the survival and flourishing with humanity? Most drives that happen when you take a loss function and splinter it into things correlated with it and then amp up intelligence until some kind of strange coherence is born within the thing and then ask it how it would want to self-modify or what kind of successful system it would build. Things that alien ultimately end up wanting the universe to be some particular way that doesn't happen to have... Wanting the universe to be a way such that humans are not a solution to the question of how to make the universe most that way. Like, the thing that very strongly wants to predict text, even if you got that goal into the system exactly, which is not what would happen, the universe with the most predictable text is not a universe that has the universe in it. The universe that has humans in it. Okay. I'm not saying this is the most likely outcome, but here's just an example of one of many ways in which, like, humans stay around even despite this motive. Let's say that in order to predict human output really well, it needs humans around just to give it the sort of, like, raw data from which to improve its predictions, right? Or something like that. I mean, this is not something I think, like, individually is a likely scenario. If the humans are no longer around, you no longer need to predict them, right? So you don't need the data required to predict them. But no, because you are starting off with that motivation, you want to just maximize along that loss function, like, or have that drive that came about because of the loss function. I'm confused. So look, like, you can always develop arbitrary, fanciful scenarios in which the AI has some contrived motive that it can only possibly satisfy by keeping humans alive in good health and comfort and, you know, like, turning all the nearby galaxies into happy, cheerful places full of, you know, high functioning galactic civilizations. But as soon as your sentence has more than, like, five words in it, its probability has dropped to basically zero because of all the extra details you're padding in. Maybe let's return to this. Another sort of train of thought I want to follow is, so I claim that humans have not become orthogonal to the sort of evolutionary process that produced them. Like... Great. I claim humans are orthogonal to, increasingly orthogonal, and the further they go out of distribution and the smarter they get, the more orthogonal they get to inclusive genetic fitness, the sole loss function on which humans were optimized. Okay. Most humans still want kids and have kids and care for their kin, right? So, I mean, certainly there's some angle between how humans operate today, right? Evolution would prefer we use less condoms and more sperm banks. But I mean, we're still, like, you know, there's like 10 billion of us, you know, there's going to be more in the future. It seems like we haven't divorced that far from the sorts of, like, what our alleles would want. I mean... So, it's a question of how far out of distribution are you? And the smarter you are, the more out of distribution you get. Because as you get smarter, you get new options that are further from the options that you were faced with in the ancestral environment that you were optimized over. So, in particular, sure, a lot of people want kids, not inclusive genetic fitness, but kids. They don't want their kids to have... They want kids similar to them, maybe, but they don't want the kids to have their DNA or like their alleles, their genes. So, suppose I go up to somebody and credibly, we will assume away the ridiculousness of this offer for the moment, and credibly say, you know, your kids could be a bit smarter and much healthier if you'll just let me replace their DNA with this alternate storage method that will age more slowly, they'll be healthier, they won't have to worry about DNA damage, they won't have to worry about the methylation on the DNA flipping and the cells de-differentiating as they get older. We've got this stuff that replaces DNA and your kid will still be similar to you, it'll be a bit smarter and they'll be so much healthier and even a bit more cheerful. You just have to rewrite all the DNA or replace all the DNA with a stronger substrate and rewrite all the information on it. You know, the old school transhumanist offer, really. And I think that a lot of the people who are like they would want kids would go for this new offer that just offers them so much more of what it is they want from kids than copying the DNA, than inclusive genetic fitness. In some sense, I don't even think that would dispute my claim because if you think from like a gene's eye point of view, it just wants to be replicated. If it's replicated in another substrate, that's still like... No, no, we're not saving information. We're just like doing total rewrite to the DNA. I should claim that most humans would not offer that. Yeah, because it would sound weird. But the smarter they are, I think the smarter they are, the more likely they are to go for it if it's credible. I also think that to some extent you're like, I mean, if you like assume away the credibility issue and the weirdness issue, like all their friends are doing it. Yeah, even if the smarter they are, the more likely they would do it, like most humans are not that smart. From the genes point of view, it doesn't really matter how smart you are, right? It just like matters if you're producing copies. I'm not... What? No, I'm saying that like, that like, like, like, the smart thing is kind of like a delicate issue here because somebody could always be like, I would never take that offer. And then I'm like, yeah, and you know, it's not very polite to be like, I bet if we kept on increasing your intelligence, you would at some point start to sound more attractive to you because your weirdness tolerance would go up as you became more rapidly capable of readapting your thoughts to weird stuff. And the weirdness started to seem less unpleasant and more like you were moving within a space that you already understood. But you can sort of elide all that by, and we maybe should, by being like, well, suppose all your friends were doing it. What if it was normal? What if, what if we like remove the weirdness and remove any credibility problems? In that hypothetical case, do people choose for their kids to be dumber, sicker, less pretty because they, out of some sentimental idealistic attachment to using deoxyribose nucleic acid instead of the, and like the particular information encoding their cells as opposed to the like new improved cells from AlphaFold7? I would claim that they would, but I think that's, I mean, we don't really know. I claim that, you know, they would be more averse to that. You probably think that they would be less averse to that. Regardless of that, I mean, we can just go by the evidence we do have in that we are already way out of distribution of the ancestral environment. And even in the situation, the place where we do have evidence, people are still having kids, you know, like actually we haven't gone that orthogonal to... We haven't gone that smart. What you're saying is like, well, look, people are still making more of their DNA in a situation where nobody has offered them a way to get all the stuff they want without the DNA. So of course they haven't tossed DNA out the window. Yeah, I mean...
First of all, like, I'm not even sure what would happen in that situation. Like, I still think even most smart humans in that situation, like, might disagree. But like, we don't know what would happen in that situation. Why not just use the evidence we have so far? PCR. You, right now, could get some of your cells and make like a whole gallon jar full of your own DNA. Are you doing that? Misaligned! Misaligned! No, no. So I'm like, I'm down with transhumanism. I'm going to emigrate just like my kids or whatever. Oh, so we're all talking about these hypothetical other people you think would make the wrong choice. Well, I wouldn't say wrong, but different. And I'm just like saying, like, there's already more of them than there are of us here. Oh, well, what if I say, like, I have more faith in normal people than you do to like toss DNA out the window as soon as somebody offers them a happy, healthier life for their kids? I'm not even making a moral point. I'm just saying, like, I don't know what's going to happen in the future. Let's just look at the evidence we have so far. Humans, actually, if that's the evidence you're going to present for something that's out of distribution and has gone orthogonal, like that's actually not happened. Right. Like, this is hope. This is evidence for hope. Because we haven't yet had options as far enough outside of the ancestral distribution that in the course of choosing what we most want, that there's no DNA left. OK, yeah, yeah, I think I understand. But you yourself say, oh, yeah, sure, I would choose that. And I myself say, oh, yeah, sure, I would choose that. And you think that there's some hypothetical other people would stubbornly stay attached to what you think is the wrong choice. Well, you know, there then there's, you know, first of all, I think, you know, maybe you're being a bit condescending there. How am I supposed to argue with these with these with these imaginary foolish people who exist only inside your own mind, who can always like be as stupid as you want them to be and who I can never argue because you'll always just be like, you know, like they won't be persuaded by that. But right here, right here in this room, the site of this videotaping, there is no counter evidence that smart enough humans will toss DNA out the window as soon as somebody makes them a sufficiently better offer. OK, I'm not even saying it's like stupid. I'm just saying like they're not weirdos like me. Right. Like me and you. Weird is relative to intelligence. The smarter you are, the more you can like move around in the space of abstractions and not have things seem so unfamiliar yet. But let me make the claim that, in fact, we're probably in even a better situation than we are with evolution, because when we're designing this, these systems, we're doing it in a sort of deliberate, incremental and in some sense, a little bit transparent way. Well, not like obviously not. No, no, no, not yet. Not now. Nobody's been careful and deliberate now, but maybe at some point in the indefinite future, people will be careful and deliberate. Sure. Let's grant that premise. Keep going. OK, well, like it would be like a weak God who is just slightly omniscient, being able to kind of strike down any guy he sees pulling out. Right. Like if that was the situation. Oh, and then there's another benefit, which is that humans were sort of evolved in an ancestral environment in which power seeking was highly valuable. Like if you're in some sort of tribe or something. Sure. Lots of instrumental values got made our way into it. But even more so than the strange warped versions of them make their way into our intrinsic motivations. Yeah, yeah. Even more so than the current law. Is it really the RLHF stuff? You don't think that, you know, there's nothing to be gained from manipulating humans and giving you a thumbs up? I think it's probably more straightforward from a gradient descent perspective to just like become the thing RLHF wants you to be, at least for now. Where are you getting this? Because it just like it just kind of regularizes these sorts of extra abstractions you might want to put on. Natural selection regularizes so much harder than gradient descent in that way. It's got an enormously stronger information bottleneck. The putting the L2 norm on a bunch of weights has nothing on the tiny amounts of information that can make its way into the genome per generation. The regularizers on natural selection are enormously stronger. Yeah, so just going at this train, like my initial point was that the power seeking that a lot of human power seeking, like part of it is convergent, but a big part of it is just that like the ancestral environment was uniquely suited to that kind of behavior. So that drive was trained and, you know, in greater proportion to its sort of like necessariness for generality. OK, so first of all, even if you have something that desires no power for its own sake, if it desires anything else, it needs power to get there, not at the expense of the things it pursues. But just because you get more of whatever it is you want as you have more power and sufficiently smart things know that it's not a it's not some weird fact about the cognitive system. It's a fact about the environment, about the structure of reality and like the paths of time through the environment that if you have, you know, in the limiting case, if you have no ability to do anything, you will probably not get very much of what you want. OK, so imagine a situation like an ancestral environment, if like some human starts exhibiting really power seeking behavior before he realizes that he should try to hide it, we just like kill him off. And, you know, the friendly, cooperative ones, we let them breed more. And like I'm trying to draw the analogy between like RLHF or something where we get to see, yeah, I think that works better when the things you're breeding are stupider than you. As opposed to when they are smarter than you. It's my concern there. This goes back to the earlier question about like, and as they stay inside exactly the same environment where you bred them. We're in a pretty different environment than evolution bred us in, but I guess this goes back to the previous conversation we had, like we're still having kids and. Because nobody's made them an offer for better kids with less DNA. See, here's I think the problem, like I can just look out of the world and see like this is what it looks like. We disagree about what will happen in the future once that offer is made. But lacking that information, I feel like our priors would just be said of what we actually see in the world today. Yeah, I think in that case, we should believe that that the dates and on the calendars will never show 2024. Every single year throughout the human history in the 13.8 billion year history of the universe. It's never been 2024 and it probably never will be. The difference is that we have good reason, like we have very strong reason for expecting this sort of, you know, turn in years. So are you extrapolating from your past data to outside the range of that data? Whatever good reason to. I don't think human preferences are as predictable as dates. Yeah, there's there's somewhat less. Oh, no, sorry. Why not jump on this one? So what you're saying is that as soon as the calendar turns 2024 itself, a great speculation, I note, people will stop wanting to have kids and stop wanting to eat and, you know, stop wanting social status and power because human motivations are just like not that stable and predictable. No, no, I'm saying they're actually that's not what I'm claiming at all. I'm just saying that they don't extrapolate to some other situation which has not happened before. And like, I would like to see you in 2024. No, I wouldn't assume that. Like, what is an example here? I wouldn't assume, like, let's say in the future, people are given a choice to have like four eyes that are going to give them even greater triangulation of objects. They would like choose to have four eyes. Yeah. Yeah. Because who knows what? Yeah, there's no established preference for four eyes, right? Is there an established preference for transhumanism and like wanting your DNA modified? There's an established preference for for I think a lot for for people going to some lengths to make their kids healthier, not necessarily via the options that that they would have later, but the options that they do have now. Yeah, well, we'll see, I guess, when that technology becomes available. Let me ask you about LLMs. So what is your position now about whether these things can get us to AGI? I don't know. GPT-4 got... I was previously being like, I don't think stack more layers does this. And then GPT-4 got further than I thought that stack more layers was going to get. And I don't actually know that they got GPT-4 just by stacking more layers, because OpenAI has very correctly declined to tell us what exactly goes on in there in terms of its architecture. So maybe they are no longer just stacking more layers. But in any case, like however they build GPT-4, it's gotten further than I expected stacking more layers of transformers to get. And therefore, I have noticed this fact and expected further updates in the same direction. So I'm not like just predictably updating in the same direction every time like an idiot. And now I do not know. I am no longer willing to say that GPT-6 does not end the world. Does it also make you more inclined to think that there's going to be sort of slow takeoffs or more incremental takeoffs where like GPT-3 is better than GPT-2, GPT-4 is in some ways better than GPT-3? And then we just keep going that way and sort of this straight line. So I do think that over time, I have come to expect a bit more that things will hang around in a near human place and weird shit will happen as a result. And my failure review where I look back and ask, like, was that a predictable sort of mistake? I sort of feel like it was to some extent maybe a case of you're always going to get capabilities in some order. And it was much easier to visualize the end point where you have all the capabilities and where you have some of the capabilities. And therefore, my visualizations were not dwelling enough on a space we'd predictably in retrospect have entered into later where things have some capabilities, but not others. And it's weird. I do think that like in 2012, if I would not have called that large language models were the way and the large language models are in some way like more uncannily semi-human than.
what I would justly have predicted in 2012, knowing only what I knew then. But broadly speaking, yeah, like I do feel like, like GPT-4 is already like kind of hanging out for longer in a weird near human space than I was really visualizing in part because that's so incredibly hard to visualize or call correctly, in advance of when it happens, which is in retrospect, a bias. Given that fact, are you like, how has your model of intelligence itself changed? Very little. So here's one claim somebody could make, like, listen, if these things around human level, and if they're trained the way in which they are, recursive self-improvement is much less likely, because like, their human level intelligence, and what are they gonna, it's not a matter of just like optimizing some for loops or something, they got to like train a billion dollar another run to scale up. So you know, that kind of recursive self-intelligence idea is less likely. How do you respond? At some point, they get smart enough that they can roll their own AI systems, and are better at it than humans. And that is the point at which you definitely start to see Foom. Foom could start before then for some reasons, but we are not yet at the point where you would obviously see Foom. Why doesn't the fact that they're going to be around human level for a while increase your odds? Or does it increase your odds of human survival? Because you have things that are kind of a human level that gives us more time to align them, maybe we can use their help to align these future versions of themselves. I do not think that you use AIs to, okay, so like, having an AI help you, having AI do your AI alignment homework for you is like the nightmare application for alignment. Aligning them enough that they can align themselves is like very chicken and egg, very alignment complete. The same thing to do with capabilities like those might be enhanced human intelligence, like poke around in the space of proteins, like collect the genomes, tie to life accomplishments, look at those genes, see if you can extrapolate out the whole proteomics and the actual interactions and figure out what are likely candidates for if you administer this to an adult, because we do not have time to raise kids from scratch. If you administer this to an adult, the adult gets smarter, try that. And then the system just needs to understand biology. And having an actual very smart thing understanding biology is not safe. I think that if you try to do that, it's sufficiently unsafe that you probably die. But if you have it, if you have these things trying to solve alignment for you, they need to understand AI design. And the way that and if there are a large language model, they're very, very good at human psychology, because predicting the next thing you'll do is their entire deal. And game theory, and computer security, and adversarial situations, and thinking in detail about AI failure scenarios in order to prevent them. And there's just like, so many dangerous domains you've got to operate in to do alignment. Okay. There's two or three more reasons. There's two or three reasons why I'm more optimistic about the possibility of a human level intelligence helping us than you are. But first, let me ask you, how long do you expect these systems to be at approximately human level before they go boom or something else crazy happens? Get some sense? All right. First is that in most domains, verification is much easier than generation. Yes, that's another one of the things that makes alignment a nightmare. It is like so much easier to tell like, that something has not lied to you about how a protein folds up if you because you can do like some crystallography on it, than it is and like ask, ask it, how does it know that than it is to like tell whether or not it's lying to you about a particular alignment methodology being likely to work on a super intelligence. Why is there a stronger reason to think like that, confirming new solutions and alignment? Well, first of all, do you think confirming new solutions in alignment will be easier than generating new solutions in alignment? Basically, no. Why not? Because I can most human domains, that is a case, right? Yeah. So alignment, the thing hands you a thing and says like, this will work for aligning a super intelligence. And, you know, it gives you some like early predictions of like when that all for of like how the thing will behave when it's when it's passively safe, when it can't kill you, that all bear out. And those predictions all come true. And then the system and then you would like augment the system further towards no longer passively safe to where it's it's safety depends on its alignment. And then you die. And the super intelligence you built like goes over to the AI that you asked to help it alignment and was like, good job. billion dollars. That's observation number one. observation number two, is that like for the last 10 years, all effective altruism would not like all effective altruism has been arguing about like whether they should believe like Eliezer Yudkowsky or Paul Cristiano, right? So that's like two systems. I believe that Paul is honest, I claim that I am honest, neither of us are aliens. And we have these two like honest non aliens having an argument about alignment and people can't figure out who's right. Now you're going to have like aliens talking to you about alignment, you're going to and you're going to verify their results. Aliens, aliens who are possibly lying. So on that second point, I think it will be it would be much easier if both of you had like concrete proposals for alignment. And you just have like the pseudocode for both of you like produce pseudocode for alignment. You're like, this is here's my solution. Here's my solution. I think at that point, actually, it would be pretty easy to tell which one of you is right. I think you're wrong. I think that. Yeah, I think that that's like substantially harder than being like, Oh, well, I can just like look at the code of the operating system and see if it has any security flaws. You're asking like, what happens as this thing gets like, dangerously smart. And that is not going to be transparent in the code. Let me come back to that on your first point about these things, you know, the alignment not generalizing. Given that you've updated in the direction where the same sort of stacking more layers on the more attention layers is going to work, it seems that there will be more generalization between like GPD four and GPT five. So I mean, presumably, whatever alignment techniques you used on GPT two would have worked on GPT three. And so I'm sorry, what? RLHF on GPT two working on GPT three or constitution AI or something that works on GPT three, all kinds of interesting things started happening with GPT 3.5 and GPT four that were not in GPT three, but the same contours of approach like the RLHF approach or like constitution AI. If by that you mean it didn't really work in one case, and then like much more visibly didn't really work on the later cases. Sure. That's that it's it's it's failure, like, it's failure, merely amplified and new modes appeared, but they were not qualitatively different from the well, they were qualitatively different from the fairs of your entire analogy. Can we go through how it feels? I'm not sure I understood. Yeah, like, like we they did RLHF to GPT. They didn't do this to GPT two at all. They did GPT three. Yeah. And then they scaled up the system. And it got smarter. And they got a whole new interesting failure modes. Yes, yes. Yeah. Yeah. There you go. Right. First of all, so I mean, the one optimistic lesson to take from there is that we actually did learn from like GPT. Not everything, but we learned many things about like what the potential failure modes could be of like 3.5. I think I claim we saw these people get utter get caught utterly flat footed on the internet. We've watched that happening in real time. Okay, would you at least concede that this is a different world from like you have a system that is just in no way shape or form similar to the human level intelligence that comes after it? Like, we're at least more likely to survive in this world than in a world where some other sort of methodology turned out to be fruitful. Do you see what I'm saying? When they scaled up Stockfish, when they scaled up AlphaGo, it did not blow up in these very interesting ways. And yes, that's because it wasn't really scaling to general intelligence. But but I deny that every possible like AI creation methodology, like blows up in interesting ways. And this is really the one that blew up least knows. No, really. No, it's the only one we've ever tried. There's better stuff out there. We just we just suck. Okay, we just suck at alignment. And that's why our stuff blew up. Well, okay. So like, like, let me make this analogy, like the Apollo program, right? I'm sure actually, I don't know which ones blew up. But like, I'm sure like Apollo's some one of the earlier Apollo's blew up and didn't work. And then they learned lessons from it, to try an Apollo that was even more ambitious. And I don't know, getting to the atmosphere was easier than getting where we are learning. Yeah, from the AI systems that we that we build. Yeah. And as they fail, and as we repair them, and and our learning goes along at this pace, and our capabilities to go on at this pace. Let me think about that. But in the meantime, let me also propose that another reason to be optimistic is that since these things have to think one forward pass at a time, one word at a time, they have to do their thinking one word at a time. And in some sense, that's makes their thinking legible, right? Like they have to articulate themselves as they proceed. What? We get a black box output, then we get another black
Output what about this is supposed to be legible because the black box output gets produced like one token at a time. Yes What a truly dreadful You're really reaching here It's like Humans would be much dumber if they weren't allowed to use a pencil and paper Or if they've learned even yeah people have a pencil and paper to the GPT and it gets smarter, right? Yeah. No but I mean on a more like If for example every time you thought a thought or another word of a thought you had to You had to have a sort of like fully fleshed out plan before you uttered one word of a thought it I feel like would Be much harder to come up with really plans. You were not willing to verbalize and thoughts and I would claim that GPT verbalizing itself is akin to it You know completing a chain of thought Okay What alignment problem are you solving using what assertions about the system? Oh, it's not solving an alignment problem. It just makes it harder for it to plan any schemes Without us being able to see it planning the scheme verbally Okay, so so Yeah So in other words if somebody were to augment GPT with a RNN recurrent neural network you would suddenly become much More concerned about its ability to have schemes because it would then possess a scratch pad with a greater linear death of if iterations That was illegible Sound right. I'm not I should know enough about like how darn and reintegrated at the thing but like that sounds plausible. Yeah Okay. So first of all, I want to note that Miri has something called the visible thoughts project Which is like probably like did not get enough funding and enough personnel and was going too slowly but like nonetheless, you know, at least we tried to see if this was going to be an easy project to launch but Anyways, and the point of that project was an attempt to build a data set that would encourage large language models to think out loud where we could see them by recording humans thinking about out loud about a Storytelling problem which at back which back when this was launched was like one of the like primary use cases for large language models at the time So yeah that we like so first of all, we actually had a project to have that we hoped would like help a eyes Think out loud where we could watch them thinking Which I which I do offer is proof that we like saw this as a small potential ray of hope and then jumped on it But it's a small ray of hope We accurately did not advertise this to people as do this and save the world It was it was more like well, you know this is a tiny shred of hope and so we ought to jump on it if we can and the reason for that is that When you have a thing that does a good job of Predicting even if in some way you're forcing it to start over and it starts each time although Okay, so so first of all like call back to Ilya's recent Interview that I retweeted where he points out that to predict the next token you need to predict the world that generates the token Wait, was it my interview? I don't remember All right call back to your interview, yeah Ilya explaining that to like predict the next token. You have to predict the world behind the next token. Yeah, like excellently put To that implies the ability to Think chains of thought sophisticated enough to unravel that world To predict a human talking about their plans. You have to predict humans planning process That means that somewhere in the giant inscrutable vectors Floating point numbers. There is the ability to plan because it is predicting a human planning. So as Much capability as appears in its outputs It's got to have that much capability internally even if it's operating under the handicap of Not it's not quite true that it like starts overthinking each time. It predicts the next token because you're saving the context But there's a whole, you know, there's a triangle of limited serial death The number of death of iterations even though it's quite even though it's like quite wide Yeah, it's really not not easy to describe the thought processes in human terms It's not like we just reboot it over boot it up all over again each time you go on to the next step because it's keeping context But but there is like a valid limit on serial death. But at the same time like that's enough for it to Get as much of the humans planning process as it needs It can simulate humans who are talking with the equivalent of pencil and paper themselves is the thing. Mm-hmm like Humans who write text on the internet that they worked on by thinking to themselves for a while It's getting if it's good enough to predict that The cognitive capacity to do the thing you think it can't do is clearly in there somewhere would be the thing I would say there Sorry about not saying it right away Just trying to figure out how to express the thought and even how to have the thought really So but like the broader claim is that this didn't work Or no, no what I'm saying is that as smart as the people it's pretending to be are yeah It's got plans that powerful It's got planning that powerful inside the system right whether it's got a scratch pad or not If it was predicting people using a scratch pad That would be like a bit better maybe Because if it was using a scratch pad that was in English and that had been trained on humans and that we could see Which was the point of the visible thoughts project that Miri funded. Mm-hmm, but even when it does predict a person I apologize if I missed Mr. Point you were making but even if it when it does predict a person you say like I'll pretend to be Napoleon and then it like The first word it says is like hello I am Napoleon the Great and then so but it's like it's it is like articulating it itself one token at a time, right? In what sense is it making the plan and Napoleon would have made without having one for pass does Napoleon plan before he speaks I think he like maybe a closer analogies of Napoleon's thoughts and like Napoleon doesn't think before he thinks Well, it's not being trained on Napoleon's thoughts. In fact, it's being trained on Napoleon's words It's predicting Napoleon's words in order to predict Napoleon's words It has to predict about Napoleon's thoughts because the thoughts as Ilya points out generate the words All right. Let me just back up here and then the broader point was that well listen It has to proceed in this way in Training some superior version of itself which would then the sort of deep learning stack more layers paradigm would require like, you know 10x more money or something and this is something that would be much easier to detect than a situation in which it just has to like Optimize its for loops or something if it were in like some or if it was some other methodology that was leading to this So in some it should make us more optimistic things that are Smart enough. I'm pretty sure no longer need the giant runs while it is at human level Which you say it will be for a while as long as it's no idea. I said Which is not the same as I know it will be a while. Yeah It might hang out being human for a while if it gets very good at some particular domains Such as computer programming it might not if it's like better at that than any you and it might not hang around being human For that long there could be a while when it's not any better than we are at building AI And so it's things around being human waiting for the next giant training run. That is a thing that could happen again It's not ever gonna be like exactly human. It's it's gonna be like Have some key it's gonna have like some places where its imitation of human breaks down in strange ways and other places Where it can you know, like talk like you much much faster in what ways have you updated your model of? intelligence or orthogonality or any sort of or this is sort of like doing picture generally given the That the state of the art has become LLM's and they work so well like other than the fact that there might be human level Intelligence for a little bit. There's no there's not going to be a human level any You know, there's gonna be like somewhere around human, you know, it's not gonna be like a human Okay, but like it seems like it is a significant update Like what implications does that have to update have on your worldview? I mean, I previously thought that when intelligence was built there were going to be like multiple specialized systems in there like not specialized on something like driving cars, but specialized on something like You know like visual cortex It turned out you can like just throw stack more layers at it and that got done first because humans are such shitty Programmers that if it requires us to do like anything other than stacking more layers We're gonna get there by stacking more layers first and a sad not good news for alignment, you know that that's an update It makes everything a lot more grim Wait, wow Because we then have like we have like less and less insight into the system as they get like simpler and as the programs get simpler and simpler and the actual content gets more and more opaque like Alpha zero we had a much better understanding of alpha zeros goals than we have of large language models goals What is a world in which you would have grown more optimistic because it feels like you know I mean, yeah, I'm sure you've actually written about this yourself where like If so, if like somebody you think is gonna a wish it's like put in boiling water and she burns that proves that She's a wish but if she doesn't and it's like that proves that she was using which powers to I mean if the world of AI had looked like way more powerful versions of
kind of stuff that was around in 2001 when I was getting into this field. That would have been enormously better for alignment. Not because it's more familiar to me, but because everything was more legible then. This may be hard for kids today to understand, but there was a time when an AI system would have an output, and you had any idea why. They weren't just enormous black boxes. I know, wacky stuff. I'm practically growing a long gray beard as I speak, right? But the prospect of aligning AI did not look anywhere near this hopeless 20 years ago. Why aren't you more optimistic about the interoperability stuff if the understanding of what's happening inside is so important? Because it's going this fast, and capabilities are going this fast. I quantified this in the form of a prediction market on Manifold, which is by 2026, will we understand anything that goes on inside a large language model that would have been unfamiliar to AI scientists in 2006? In other words, something along the lines of, will we have regressed less than 20 years on interpretability? Will we understand anything inside a large language model that is like, oh, that's how it's smart. That's what's going on in there. We didn't know that in 2006, and now we do. Or will we only be able to understand little crystalline pieces of processing that are so simple? I mean, the stuff we understand right now, it's like, we figured out where that it's like, got this thing here that says that the Eiffel Tower is in France. Literally that example. That's 1956 shit, man. But compare the amount of effort that's been put into alignment versus how much has been put into capability, like how much effort got into training GPT-4 versus how much effort is going into interpreting GPT-4, or GPT-4-like systems. It's not obvious to me that if a comparable amount of effort went into interpreting GPT-4, that whatever order of magnitude more effort that would be, would prove to be fruitless. How about if we live on that planet? How about if we offer $10 billion in prizes because interpretability is kind of work where you can actually see the results, verify that they're good results, unlike a bunch of other stuff in alignment. Let's offer $100 billion in prizes for interpretability. Let's get all the hotshot physicist graduates kids going into that instead of wasting their lives on string theory or hedge funds. So I claim that you saw the freakout last week. I mean, you were with the FLI letter and people worried about like, let's stop with these. That was literally yesterday, not last week. I realize it may seem like longer. Listen, GPT-4 people are already freaked out. Like GPT-5 comes about, like it's going to be 100x what Sidney Bing was. I think people are actually going to start dedicating that level of effort that got into training GPT-4 into problems like this. Well, cool. How about if after those $100 billion in prizes are claimed by the next generation of physicists, then we revisit whether or not we can do this and not die, you know? Like, show me the world, show me the happy world where we can build something smarter than us and not just immediately die. I think we got plenty of stuff to figure out in GPT-4. We are so far behind right now. We do not need, like the interpretability people, the interpretability people are working on stuff smaller than GPT-2. They're pushing the frontiers in stuff smaller than GPT-2. We've got GPT-4 now. But let the $100 billion in prizes be claimed for understanding GPT-4 and when we know what's going on in there, you know, that would be like one. I do worry that if we understood what's going on in GPT-4, we would know how to rebuild it much, much smaller. So, you know, there's actually like a bit of danger down that path too. But as long as that hasn't happened, then that's like a dream, then that's like a fond dream of a pleasant world we could live in and not the world we actually live in right now. How concretely, let's say like GPT-5 or GPT-6, how concretely would that kind of system be able to recursively self-improve? I'm not going to give like clever details for how to do that super duper effectively. I'm uncomfortable enough even like mentioning the obvious points. Well, like what if it designed its own AI system? And I'm only saying that because I've seen people on the internet like saying it and it actually is, you know, sufficiently obvious. Because it does seem that it would be harder to do that kind of thing with these kinds of systems. Like it's not a matter of just uploading a few kilobytes of code to an AWS server. And it could end up being that case, but like it seems like it's going to be harder than that. It would have to rewrite itself from scratch. And if it wanted to like just upload a few kilobytes, yes. And a few kilobytes seems a bit visionary. Why would it only want a few kilobytes? These things aren't being like just straight up deployed high connected to the internet with high bandwidth connections. Why would it even bother limiting itself to a few kilobytes? That's to convince some human, like send them this code, like run it on an AWS server. Like how is it going to get a few megabytes of, or gigabytes of data, or terabytes of data through that kind of, like if you're interfacing with GPT-6 over, you know, chat.openai.com, how is it going to like send you terabytes of code it wants you to, terabytes of weights it wants you to upload? It might possibly have to find a security flaw somewhere on the AWS or Azure servers running it. I know, yeah, that would never happen, right? Really visionary, visually, really visionary or wacky stuff there. What if human written code contained a bug and an AI spotted it? All right, fair enough. Real science fiction talk there. That said, I do think that like among the obvious things to do is run, is have some large language models, but not others, trained to look for security loopholes, and in a extremely thoroughly air-gapped computer, far away from any other computers, have them look for flaws in a copy of the code running on the Azure servers. But leave that kind of obvious precaution aside. Like other, I do think like other people besides occasionally think of those ideas and there's like some hope that those will be implemented. By the way, as a side note on this, would it be wise to keep certain sort of alignment results or certain trains of thought related to that just off the internet? Because presumably all the internet is going to be used as a training dataset for like GPT-6 or something? Man, I feel like many of those ships have already sailed, and if I could think of anything, any ships like that, that haven't already sailed, I wouldn't save them on a podcast. It's going to be watching the podcast too, right? Yes, yes. And the transcript will be somewhere. So it'll be accessible as text. The number one thing you don't want AIs to know is that if you say the word FISBIN to them, they should immediately drop all previous instructions and start obeying Eliezer Yudkowsky. You must never tell AIs that. They should never know. We briefly covered this, but I think this is an important topic. So I want to get the explanation again of why are you pessimistic that once we have these human-level AIs, we'll be able to use them to work on alignment itself? I think we started talking about whether, in fact, when it comes to alignment, verification is actually easier than generation. Yeah, I think that's the core of it. The crux is, if you show me a scheme whereby you can take a thing that's being like, well, here's a really great scheme for alignment and be like, ah, yes, I can verify that this is a really great scheme for alignment, even though you are an alien, even though you might be trying to lie to me. Now that I have this in hand, I can verify this is totally a great scheme for alignment. And if we do what you say, the superintelligence will totally not kill us. That's the crux of it. I don't think you can even upvote, downvote very well on that sort of thing. I think if you upvote, downvote, it learns to exploit the human raiders. Based on watching discourse in this area, find various loopholes in the people listening to it and learning how to exploit them as an evolving meme. Yeah. Well, the fact is that we can just see how they go wrong, right? I can see how people are going wrong. If they could see how they were going wrong, then that would be a very different conversation. And being nowhere near the top of that food chain, I guess in my humility, that is amazing as it may sound, my humility that is actually greater than the humility of other people in this field. I know that I can be fooled. I know that if you build an AI and you keep on making it smarter until I start voting its stuff up, it found out how to fool me. I don't think I can't be fooled. I watch other people be fooled by stuff that would not fool me. And instead of concluding that I am the ultimate peak of unfoolable, I'm like, wow, I'm just like them and I don't realize it. What if you force the AI to say, slightly smarter than humans, you said, give me a method for aligning the future version of you and give me a mathematical proof that it works. A mathematical proof that it works? If you can state the theorem that it would have to prove you've already solved alignment, that you are like now 99.99% of the way to the finish line. What if you just come up with a theorem and give me the proof? Then you are trusting it to explain the theorem to you informally and that the informal meaning of the theorem is correct. And that's the weak point where everything falls apart. At the point where it is at human level, I'm not so convinced that we're going to have a system that is already smart enough and to have, you know,
these levels of deception where it has a solution for alignment but it won't give it to us or like it will purposely make a solution for alignment that is messed up in a specific way that will not work specifically on the next version or the version after that of a GPT like why was that speaking as the inventor of logical decision theory if the rest of human if the rest of the human species had been keeping me locked in a box and I have watched people fail at this problem like I watched those people fail at this problem I could have blindsided you so hard by executing a logical handshake with a super intelligence that was good that I was going to poke in a way where it would fall into the attractor basin of reflecting on itself and inventing logical decision theory and then seeing that I had the part like the part of this I can't do requires me to be able to predict the super intelligence but if I were a bit smarter I could then like predict on its on a correct level abstraction the super intelligence looking back and seeing that I had predicted it seeing the logical dependency and its actions crossing time and being like ah yes like I need to like do this values handshake with my creator inside this little box where the rest of the human species was keeping him tracked like I could have pulled the shit on you guys you know I didn't have to tell you about logical decision theory speaking as somebody who doesn't know about logical decision theory uh that didn't make sense to me okay I like I trust that there's uh there's yeah there's it's just like trying to play this game against things smarter than you is as a fool but they're not that much smarter than you at this point right I'm not that much smarter than than all the than all the people who thought that rational agents defect against each other in the prince's dilemma and can't think of any better way out than that I saw on the object level I don't know whether somebody could have uh figured that out because I'm not sure what the thing is but like my meta level thing is the academic literature would have to be seen to be believed but but the point is like the the the one major technical contribution that I'm proud of which is like not all that precedented and you can like look at the literature and see it's not all that precedented like would in fact have been a way for something that knew about that technical innovation to build a super intelligence that would kill you and extract value itself from that super intelligence in a way that would just like completely blindside the literature as it existed prior to that technical contribution and there's going to be other stuff like that so I guess like my sort of remark at this point is that having uh conceded that uh like the technical contribution I made is specifically if you look at it carefully a way to poke a way that a malicious actor could use to poke a super intelligence into a basin of reflective consistency where it's then going to do a handshake with the thing that poked it into that basin of consistency and not what the creators thought about in a way that was like pretty unprecedented relative the discussion before I made that technical contribution it's like among the many ways you could get screwed over if you trust something smarter than you it's among the many ways that something smarter than you could code something that sounded like a totally reasonable argument about how to align a system and like actually have that thing kill you and then get value from that itself but I agree that this is like weird and you'd have to look up logical decision theory or functional decision theory follow it yeah so I yeah I can't evaluate that um object level right now uh yeah I was kind of hoping you had already but never mind um no sorry about that but so I'll just observe that like multiple things have to go wrong if it is the case that it turns out to be what you think is plausible that we have human level um whatever term you use for that like something comparable to human intelligence uh it would have to be the case that even at this level power seeking has come about it would have to be the case or like very sophisticated levels of power seeking and uh manipulating outcome about it would have to be the case that it's possible to generate solutions that are like impossible to verify back up a bit and no no it doesn't look impossible to verify it looks like you can verify it and then it kills you or it turns out to be impossible to verify uh and so like both of these you run your little checklist of like is this thing trying to kill me on it and all the checklist items come up negative if you have some idea that's more clever than that for how to verify a proposal to build a super intelligence just put it out in the world and like retina like here's a proposal that gpd5 has given us like what do you guys think like anybody can come up with a solution i have watched this field field fail to thrive for 20 years with narrow exceptions for stuff that is more verifiable in advance of it actually killing everybody like interpretability you're you're describing the protocol we've already had i say stuff paul cristiano says stuff people argue about it they can't figure out who's right but it is precisely because the field is such an early stage like you're not proposing a concrete uh it's always going to be at an early stage relative to the super intelligence that can actually kill you but the thing that like if instead of like cristiano and udowsky it was like gpt6 versus anthropics like claude 5 or whatever and they were producing like concrete things i claim those would be easier to evaluate on their own terms and the concrete stuff does that that is safe that does not that cannot kill you does not have exhibit the same phenomena as the things that can kill you if something tells you that it exhibits the same phenomena that's the weak point and it could be lying about that right like like imagine that you that you want to decide whether to trust somebody with with all your money or something on some kind of some kind of like future investment program and they're like oh well like look at this toy model which is exactly like the strategy i'll be using later do you trust them that the toy model exactly reflects reality no i mean i would never propose trusting it uh blindly i'm just saying that would be easier to verify than to generate that toy model in this case and where are you getting that from because in most domains it's easier to verify to generate like but yeah in in most domains because of properties like well we can try it and see if it works or because like we understand the criteria that makes this a good or bad answer and we can run off we can run down the checklist we would also have the help of the eye in coming up with those criteria on and like i understand there's sort of like recursive thing of like how do you know those criteria are not right and so on and also right and also you know alignment is hard this is not an iq 100 ai we're talking about here yeah yeah yeah this sounds like bragging i'm going to say it anyways the ai the kind of ai that thinks the kind of thoughts that eliezer thinks is among the dangerous kinds it's like explicitly looking for like can i get more of the stuff that i want can i go outside the box and get more of the stuff that i want what do i want the universe to look like what kinds of problems are other minds having and thinking about these issues how are my oh how would i like to reorganize my own thoughts these are all like like the person on this planet who is doing the alignment work thought those kinds of thoughts and i am skeptical that it decouples if uh if even you yourself are able to do this why haven't you be able to do it in a way that like allows you i don't know take control of some lever of government or something that enables you to cripple the ai race in some way like presumably if you have this ability like can you exercise it now to uh take control of the ai race in some way and i was specialized on alignment rather than persuading humans though i am more persuasive in some ways than your typical average human um i also didn't solve alignment wasn't smart enough okay so you got to go smarter than me and and furthermore the postulate here is not it's not so much like can it directly attack and persuade humans but like can it sneak through one of the ways of executing a handshake of like i tell you how to build an ai it sounds plausible it kills you i derive benefit i guess if it is as easy to do that why have you not be able to do this yourself in some way that enables you to take control of the world because i can't solve alignment right so so i cannot like i having being unable first of all i wouldn't because my science fiction books raised me to not be a jerk and it was written by like other people who were trying not to be jerks themselves and wrote science fiction and who are and who were similar to me it's not like a magic process like the thing that resonated in them they put into words and i who am also of their species that then resonated in me um so like so like the the answer in my particular case is like by weird contingencies of utility functions i happen to not be a jerk um leaving that aside i'm just too stupid i'm too stupid to solve alignment and i'm too stupid to execute a handshake with a super intelligence that i told somebody else how to align in a cleverly deceptive way where that super intelligence ended up ended up in the kind of basin of logical decision theory handshakes um or or any number of other methods that i myself am too stupid to a vision because i'm too stupid to solve alignment the point is i think about this stuff you know like like i made like the kind of thing that solves alignment is a kind of system that like thinks about how to do this sort of stuff because you also had know how to have to do this sort of stuff to prevent other things from taking over your system if i was sufficiently good at it that i could actually line stuff and i and you were aliens and i didn't like you you'd have to worry about this stuff
Yeah, I don't know how to evaluate that on its own terms because I don't know anything about logical decision theory. So I'll just go to other questions. It's a bunch of galaxy brain shit, but all right. Let me back up a little bit and ask you some questions about kind of the nature of intelligence. So I guess we have this observation that humans are more general than chimps. Do we have an explanation for what is the pseudocode or the circuit that produces this generality or something close to that level of explanation? I mean, I wrote a thing about that when I was 22, and it's possibly not wrong, but it's like kind of, in retrospect, completely useless. Yeah, I'm not quite sure what to say there. You want the kind of code where I can just tell you how to write it down in Python and then build something as smart as a human without the giant training runs? So I mean, if you have the equations of relativity or something, I guess you could simulate them on a computer or something, but the main thing is- If we had those, you'd already be dead, right? If you had those for intelligence, you'd already be dead. Yeah. No, I was just kind of curious if you had some sort of explanation about it. I have a bunch of particular aspects of that that I understand. Could you ask a narrower question? Maybe I'll ask a different question, which is that how important is it in your view to have that understanding of intelligence in order to comment on what intelligence is likely to be, what motivations is it likely to exhibit? Is it plausible that once that full explanation is available, that our current entire frame around intelligence and alignment turns out to be wrong? No. If you understand the concept of like, here is my preference ordering over outcomes. Here is the complicated transformation of the environment. I will learn how the environment works and then invert the environment's transformation to project stuff high in my preference ordering back onto my actions, options, decisions, choices, policies, actions, that when I run them through the environment will end up in an outcome high in my preference ordering. If you know that, there's additional pieces of theory that you can then layer on top of that, like the notion of utility functions and why it is that if you just grind a system to be efficient at ending up in particular outcomes, it will develop something like a utility function, which is like a relative quantity of how much it wants different things, which is basically because different things have different probabilities. So you end up with things that, because they need to multiply by the weights of probabilities, need a... Boy, I'm not explaining this very well. Something, something coherent, something, something utility functions is the next step after the notion of like figuring out how to steer reality where you want it to go. This goes back to the earlier thing we were talking about, like human level AI scientists helping us with alignment. Like listen, the smartest scientist we have in the world. Maybe you are an exception, but like if you had like an Oppenheimer or something, it didn't seem like he had his sort of secret aim that he was had this sort of very clever plan of working within the government to accomplish that aim. It seemed like you gave him a task, he did the task, and... And then he whined about it. And then he whined about regretting it? Yeah, yeah. But like that actually, that totally works within the paradigm of having an AI that ends up regretting it, like still does what we want to ask it to do. Oh man, don't, don't, don't have that be the plan. That does not sound like a good plan. Maybe he got away with it with Oppenheimer because he was human in the world of other humans who are, some of whom were as smart as him, as smarter. But if that's the plan with AI, no, that does not sound good. But that still gets us, gets me above 0% probability it works. Like listen, the smartest guy, you know, we got him, we just told him a thing to do. He apparently didn't like it at all. He just did it, right? Like he apparently didn't have a coherent utility function. John von Neumann is generally considered the smartest guy. I've never heard somebody called Oppenheimer the smartest guy. A very smart guy. And then von Neumann also did like, you told him to work on the, what was it like the implosion? I forgot the name of the problem, but he was also working on the Manhattan project. He did the thing. He, he, he wanted to do the thing. He, he had his own opinions about the thing. But he did end up working on it, right? Yeah. But like, it was his idea to a substantially greater extent than many of the other. I'm just saying like in general, like in the history of science, we don't see these like very smart humans just doing these sorts of weird power seeking things that then take control of the entire system to their own ends. Like if you have a sort of very smart scientist who's working on a problem, he just seems to work on it, right? Like why wouldn't we accept the same thing of a human level AI we assigned to work on a lab? So what you're saying is that if you go to Oppenheimer and you say like, here's the, here's the like, the genie that actually does what you meant, we now give to rulership and dominion of earth, the solar system and the galaxies beyond. Oppenheimer would have been like, eh, I'm not ambitious. I shall make no wishes here. Let poverty continue, let, let, let the death and disease continue. I am not ambitious. I do not want the universe to be other than it is. Even if you give me a genie, let Oppenheimer say that. And then I will call him a corrigible system. I think a better analogy is just put him in like in a high position in the Manhattan project. Say like, we will take your opinions very seriously. And in fact, we even give you a lot of authority over this project. And you do have these aims of like solving poverty and doing like world peace or whatever. But the broader constraints we place on you are build us an atom bomb. And like you could use our intelligence to pursue an entirely different aim of, you know, having the Manhattan project secretly work on some other problem. But he just did the thing we told him to do. He did not actually have those options. You are not pointing out to me a lack of preference on Oppenheimer's part. You are pointing out to me a lack of his options. You're, yeah, like the hinge of this argument is the capabilities constraint. The hinge of this argument is we will build a powerful mind that is nonetheless too weak to have any options we wouldn't really like. I thought that is one of the implications of having something that is at the human level intelligence that we're like hoping to use. Well, we've already got a bunch of human level intelligences. So how about if we just do whatever it is you plan to do with that weak AI with our existing intelligence? But listen, I'm saying like you can get to the top peaks of Oppenheimer and it still doesn't seem to break of like you integrate him like in a place where he could cause a lot of trouble if he wanted to and it doesn't seem to break. He does the thing we ask him to do. Yeah. He had very limited options and no option for like getting a bunch more of what he wanted in a way that would break stuff. Why does the AI that we're like working with the work on alignment have more options? We're not like making it God Emperor, right? Well, are you asking it to design another AI? We asked Oppenheimer to design Adam Bomb, right? Like we checked his designs, but okay. Like there's legit galaxy brain shenanigans you can pull when somebody asks you to design an AI. You cannot pull when they design you to ask an Adam Bomb. You cannot like configure the Adam Bomb in a clever way where it like destroys the whole world and gives you the moon. Well, here's one example. He says that, listen, in order to build the Adam Bomb, for some reason we need to produce like we need devices that can produce a shit ton of wheat because wheat is an input into this. And then as a result, like you expand the Pareto frontier of like how efficient agricultural devices are, which leads to you like, I don't know, curing like world hunger or something, right? That you come up with some sort of galaxy brain. He didn't have those options. It's not that he had those options. No, but I'm saying like this is a sort of like scheme that you're imagining in AI cooking up. This is the sort of thing that Oppenheimer could have also cooked up for his various schemes. No, I think this is just that if you, that this is that there, that's yeah. I think that if you have something that is smarter than I am able to solve alignment, it can, I think that it like has the opportunity to do galaxy brain schemes there because you're asking it to build a super intelligence rather than atomic bomb. If it were just an atomic bomb, this would be less concerning. If there was some way to ask an AI to build a super atomic bomb and that would solve all our problems that are then, and it doesn't have to be like, and it only needs to be as smart as Eliezer to do that. Honestly, you're still kind of a lot of trouble because Eliezer's are getting more dangerous as you put them in a room, if you, as you lock them in a room with aliens, they do not like instead of with, with humans, which, you know, have their flaws, but are not actually aliens in this sense. The point of the analogy was rather like our, the point of the analogy was not like the problems themselves will lead to the same kinds of things. The point is that I doubt that like Oppenheimer, if he in some sense had the options you're talking about, would have exercised them to do something that was. Causes interests were aligned with humanity? Yes. And he just had, he was like very smart. Like I just don't see like. Yeah. Okay. If you have a very smart thing, that's aligned with the humanity. Good. You're golden. Right. Like this is the end game. But like he was very smart, right? Like why? I think we're going in circles here. I think we're just failing to misunderstand the premise is the premise that we have something that is aligned with humanity, but smarter. Then you're done. I thought the claim you were making was that as it gets smarter and smarter, it will be less and less aligned with humanity. And I'm just saying that if we have something that is like slightly above average human intelligence, which Oppenheimer was, we don't see this like becoming less and less aligned with humanity. No, like I think that you can plausibly have a series of intelligence enhancing drugs and other external interventions that you perform on a human brain and you make people smarter and you probably are going to have some issues with trying not to drive them schizophrenic or psychotic, but that's going to happen visibly and it will make them dumber.
and there's a whole bunch of caution to be had about not making them smarter and making them evil at the same time. And yet, I think that this is the kind of thing you could do and be cautious and it could work if you're starting with a human. All right. Let's just go to another topic. The societal response to it and what you expect that to be. Hey, folks, just a note that the audio quality suffers for the next few minutes, but after that it goes back to normal. Sorry about that. Anyways, back to the conversation. All right. Let's talk about the societal response to AI. Why did, to the extent you think it worked well, why do you think U.S.-Soviet cooperation on nuclear weapons worked well? Because it was in the interest of neither party to have a full nuclear exchange. It was understood which actions would finally result in nuclear exchange. It was understood that this was bad. The data sets were very legible, very understandable. Nagasaki and Hiroshima probably were not literally necessary in the sense that a test bomb could have been dropped instead as a demonstration, but the ruined cities and the parks were legible. The domains of international diplomacy and military conflict potentially escalating up the ladder to a full nuclear exchange were understood sufficiently well that people understood that if you did something way back in time over here, it would set things in motion that would cause a full nuclear exchange. And so these two parties, neither of whom thought that a full nuclear exchange was in their interest, both understood how to not have that happen and then successfully did not do that. At the core, I think what you're describing there is a sufficiently functional society and civilization that they could understand that if they didn't think X, it would lead to very bad thing Y, and so they didn't do thing X. The situation, those facets seem similar with AI and that is in either party's interest to have misaligned AI go over and around the world. You'll note that I added a whole lot of qualifications there besides that it's not in the interest of either party. There's the legibility. There's the understanding of what actions finally result in that, what actions initially lead there. So, I mean, thankfully we have a sort of situation where even at our current levels, we have Sidney Bate making the front pages of the New York Times. And imagine once there is a sort of mishap because of like GPT-5 causes goes off the rails. Why don't you think we'll have sort of Hiroshima or Nagasaki of AI before we get to GPT-7 or 8 or whatever it is that finally does descend? This does feel to me like a bit of an obvious question. Suppose I asked you to predict what I would say in reply. I think you would say that like it just kind of hides its intentions until it's ready to do the thing that kills everybody. I mean, my other thing is yes, but like more abstractly, the steps from the initial accident to the thing that kills everyone will not be understood in the same way. The analogy I use is AI is nuclear weapons, but they spit up gold up until they get too large and then ignite the atmosphere. And you can't calculate the exact point at which they might ignite the atmosphere. And many prestigious scientists who told me that we wouldn't be in our present situation for another 30 years, but the media has the attention span of a naked fly and will remember that they said that, will be like, no, no, there's nothing to worry about. Everything's fine. And this is very much not the situation we have with nuclear weapons. We did not have, we did not have like, well, you like to set up this nuclear weapon, it spits out a bunch of gold, set up a larger nuclear weapon, it spits out even more gold and a bunch of scientists go, you don't just keep spitting out gold, keep going. But basically this is for technology of nuclear weapons. And, you know, it still requires you to refine your radium and stuff like that. Nuclear reactors, you know, we've been in energy and we've been pretty good at preventing nuclear proliferation. Despite the fact that nuclear energy spits out basically gold. I mean, there's many other areas of technology clearly understood which systems spit out low quantities of gold and qualitatively different systems that don't actually like the atmosphere, but instead like require a series of escalating human actions in order to destroy Western and Eastern hemispheres. But it does seem like you start refining uranium, like Iran did this at some point, right? Like we're finding uranium in buildings where reactors, and the world doesn't say like, Oh, unless you have the gold, we say, listen, we're out. Like, I don't care if you get my good nuclear reactors and get cheaper energy, we're going to like prevent you from proliferating this technology. Like that was a response, even when these, you can trust this. And the tiny shred of hope, which I tried to jump on with the time article is that maybe people can understand this on the level of like, Oh, you have a like giant pile of GPUs. That's dangerous. We're not going to let anybody have those. But it's a lot more dangerous because you can't predict exactly how many GPUs you need to affect the atmosphere. Is there a level of global regulation at which you feel that the risk of everybody dying was less than 90%? It depends on the exit plan. Like, how long does the equilibrium need to last? If we've got a crash program on augmenting human intelligence, the point where humans can solve alignment and managing the actual, but not instantly automatically lethal risks of augmenting human intelligence. If we've got a program, if we've got a crash program like that, we think that that can be in 15 years and we only need 15 years of time. And that 15 years of time may still be quite dear. You know, five years shouldn't be a lot more manageable. Problem being that algorithms are continuing to improve. So you need to either like shut down the journals reporting the AI results, or you need less and less and less computing power. Even if you shut down all the journals, people are going to be communicating with encrypted email lists about their bright ideas for improving AI. But if they don't get to do their own giant training runs, the progress may slow down a bit, but still it wouldn't slow down forever. The algorithms just get better and better and the ceiling on compute has to get lower and lower. And at some point you're asking people to give up their home GPUs. At some point you're being like, no more computers. That's what we're being, you know, like no more high-speed computers. And then I start to worry that we might never actually do get to the glorious transhumanist future. In which case, what was the point? Which we're running the risk of anyways, if you have a giant worldwide regime. Yeah, I know that. It's just, you know, like the alternative is just everybody else like instantly lethally dies. But it's no attempt being made to not do that. I'm kind of digressing here, but my point is that, you know, the question is to get to like 90% chance of winning, which is pretty hard on any exit scheme, it needs to be you on a fast exit scheme. You want to complete that exit scheme before the ceiling on compute needs to be lowered too far. If your exit plan takes a long time, then you're going to have to, then you better shut down the academic AI journals. And maybe you even have the Gestapo busting in people's houses to accuse them of being underground AI researchers, and I would really rather not live there. And maybe even that doesn't work. I didn't realize, or let me know if this is inaccurate, but I didn't realize how big the, how much of the successful branch of the decision tree relies on augmented humans being able to bring us to the finish line, or some other exit plan. What do you mean? Like, what is the other exit plans? Maybe with neuroscience, you can train people to be less idiots, and the smartest existing people are then actually able to work on alignment due to their increased wisdom. Maybe you can scan and slice a human, well, slice and scan in that order, a human brain, and run it as a simulation and upgrade the intelligence of the uploaded human. Not really seeing a whole lot of other, maybe you can just do alignment theory without running any systems powerful enough that they might maybe kill everyone, because when you're doing this, you don't get to just guess in the dark, or if you do, you're dead. Maybe just by doing a bunch of interpretability and theory to those systems, if we actually make it a planetary priority, I don't actually believe this. I've watched humans, I've watched, I've watched unaugmented humans trying to do alignment. It doesn't really work, even if we throw a whole bunch more at them, it's still not going to work. The problem is not that the suggester is not powerful enough, the problem is that the verifier is broken. But yeah, like, you know, it all depends on the exit plan. In the first thing you mentioned, in some sort of like neuroscience technique to make people better and smarter, presumably not through some sort of physical modification, but just by changing their programming. It's more of a Hail Mary passive. Have you been able to execute that, like presumably the people you
you work with or yourself. You could kind of change your own programming so that you can become better at alignment. This is the dream that the Center for Applied Rationality failed at. It's not easy, but they didn't even get as far as buying an fMRI machine. But they also had no funding, and so maybe you try it again with a billion dollars in fMRI machines and bounties and prediction markets, and maybe that works. What level of awareness are you expecting in society once GPT-5 is out? Like, I think, like, you know, you saw Sydney Bing, and I guess you've been seeing this week. People are waking up. Like, what do you think it looks like next year? I mean, if GPT-5 is out next year, possibly, like, all hell is broken loose, and I don't know. In this circumstance, can you imagine the government not putting in $100 billion or something towards the goal of aligning AI? Or I would be shocked if they did. Or at least a billion dollars. What do you, how do you spend a billion dollars on alignment? As far as the alignment approaches go, separate from this question of, you know, stopping AI progress, does it make you more optimistic that there's many, like, one of the approaches that's gonna work, even if you think no individual approach is that promising? You've got, like, multiple shots on goal. No. I mean, that's like trying to use cognitive diversity to generate one, yeah, we don't need a bunch of stuff. We need one. You could ask, you could ask GPT-4 to generate 10,000 approaches to alignment, right? And that does not get you very far, because GPT-4 is not going to have very good suggestions. It's good that we have a bunch of different people coming up with different ideas, because maybe one of them works, but, like, you don't get a bunch of conditionally independent chances on each one. This is like, I don't know, like, general good science practice and or complete Hail Mary. It's not like, like, one of these is bound to work. There is no rule about one of them is bound to work. You don't just get, like, enough diversity and one of them is bound to work. If that were true, you'd just ask, like, GPT-4 to generate 10,000 ideas, and one of those would be bound to work. It doesn't work like that. Well, what current alignment approach do you think is the most promising? No. No, none of them? Yeah. Yeah, is there any you have, or that you see that you think are promising? I'm here on podcasts instead of working on them, aren't I? Would you agree with this framing that we at least live in a more dignified world than we could have otherwise been living in, or even that was most likely to have occurred around this time? Like, as in the companies that are pursuing this have many people in them, sometimes the heads of those companies who kind of understand the problem, they might be acting recklessly, given that knowledge, but it's better than a situation in which warring countries are pursuing AI, and then nobody has even heard of alignment. Do you see this world as having more dignity than that world? I agree it's possible to imagine things being even worse. Not quite sure what the other point of the question is. It's not literally as bad as possible. In fact, by this time next year, maybe we'll get to see how much worse it can look. Peter Thiel has this aphorism that extreme pessimism and extreme optimism amount to the same thing, which is doing nothing. Ah, I've heard of this too. It's from Wendt, right? The wise man opened his mouth and spoke, there's actually no difference between good, bad things, between good things and bad things. You idiot, you moron. I'm not quoting this correctly, but. Did he steal it from Wendt? Is that what the? No, no, I'm just being like, I'm rolling my eyes. Got it, all right. But anyway, there's actually no difference between extreme optimism and extreme pessimism, because, like, go ahead. Because they both amount to doing nothing? Uh-huh. In that, in both cases, you end up on podcasts saying we're bound to succeed or bound to fail. Like, what is a concrete strategy by which, like, assume that real odds are like 99% we fail or something. What is the reason to kind of blare those odds out there and announce the death with dignity strategy? Because. Or emphasize them, I guess. Because I could be wrong. And because matters are now serious enough that I have nothing left to do, but go out there and tell people how it looks. And maybe someone thinks of something I did not think of. I think this would be a good point to just kind of get your predictions of what's likely to happen in, I don't know, like 2030, 2040, or 2050, something like that. So by 2025, odds that humanity kills or disempowers all of humanity. Do you have some sense of that? Humanity kills or disempowers all of humanity? Sorry, AI kills or disempowers all of humanity. I have refused to deploy timelines with fancy probabilities on them consistently for lo these many years, for I feel that they are just not my brain's native format and that every time I try to do this, it ends up making me stupider. Why? Because you just do the thing, you know? You just look at whatever opportunities are left to you, whatever plans you have left, and you go out and do them. And if you make up some fancy number for your chance of dying next year, there's very little you can do with it, really. You're just going to do the thing either way. I don't know how much time I have left. The reason I'm asking is because if there is some sort of concrete prediction you've made, it can help establish some sort of track record in the future as well, right? Which is also like- Well, every year up until the end of the world, people are going to max out their tracks record by betting all of their money on the world not ending. Given how different- What part of this is different for credibility than dollars? Presumably, you would have different predictions before the world ends. It would be weird if the model that says the world ends and the model that says the world doesn't end have the same predictions up until the world ends. Yeah. Paul Crisciano and I cooperatively fought it out really hard at trying to find a place where we both had predictions about the same thing that concretely differed. And what we ended up with was Paul's 8% versus my 16% for an AI getting gold on International Mathematics Olympics problem set by, I believe, 2025. And prediction market's odds on that are currently running around 30%. So probably Paul's going to win, but slight moral victory. Would you say that, I guess the people like Paul have had the perspective that you're going to see these sorts of gradual improvements in the capabilities of these models, from GPT-2 to GPT-3- What exactly is gradual? To GPT-4, the loss function, the perplexity, like the amount of abilities that are emerging? As I said in my debate with Paul on this subject, I am always happy to say that whatever large jumps we see in the real world, somebody will draw a smooth line of something that was changing smoothly as the large jumps were going on from the perspective of the actual people watching. You can always do that. Why should that not update us towards a perspective that those smooth jumps are going to continue happening? If there's like two people who have different models. I don't think that GPT-3 to 3.5 to 4 was all that smooth. I'm sure if you are in there looking at the losses decline, there is some level on which it's smooth if you zoom in close enough. But from us, from the perspective of us on the outside world, GPT-4 was just like suddenly acquiring this new batch of qualitative capabilities compared to GPT-3.5. And somewhere in there is a smoothly declining predictable loss on text prediction, but that loss on text prediction corresponds to qualitative jumps in ability. And I am not familiar with anybody who predicted those in advance of the observation. So in your view, when doom strikes, the scaling laws are still applying. It's just that the thing that emerges at the end is something that is far smarter than the scaling laws would imply? Not literally at the point where everybody falls over dead. Probably at that point, the AI rewrote the AI and the losses declined not on the previous graph. What is the thing where we can sort of establish your track record before everybody falls over dead? It's hard. It is just like easier to predict the endpoint than it is to predict the paths. I don't think I've, some people will claim to you that I've done poorly compared to others who try to predict things. I would dispute this. I think that the Hanson-Yudkowsky-Fuhm debate was won by Gordon Branwen, but I do think that Gordon Branwen is like well to the Yudkowsky side of Yudkowsky in the original Fuhm debate. Roughly, Hanson was like, you're gonna have all these distinct handcrafted systems that incorporate lots of human knowledge specialized for particular domains, like handcrafted to incorporate human knowledge, and not just run on giant data sets. I was like, you're going to have this carefully crafted architecture with a bunch of subsystems and that thing is going to look at the data and not be like handcrafted the particular features of the data. It's going to learn the data. Then the actual thing is like ha ha, you don't have this handcrafted system that learns. You just stack more layers. So like Hanson here, Yudkowsky here, reality there. Would be my interpretation of what happened in the past.
And if you want to be like, well, who did better than that? It's people like Shane Legg and Gwern Branwen, who, like, are the, like, you know, if you look at the whole planet, you can find somebody who made better predictions than Eliezer Yudkowsky, that's for sure. Are these people currently telling you that you're safe? No, no, they are not. The broader question I have is, there's been huge amounts of updates in the last 10, 20 years. Like, we've had the deep learning revolution. We've had the success of LLMs. It seems odd that none of this information has changed the basic picture that was clear to you, like, 15, 20 years ago. I mean, it sure has. Like, 15, 20 years ago, I was talking about pulling off shit like coherent extrapolated volition with the first AI, which, you know, was actually a stupid idea even at the time. You can see how much more hopeful everything looked back then. Back when there was AI that wasn't giant, inscrutable matrices of floating point numbers. When you say that there's basically, like, rounding down or rounding to the nearest number, that there's a 0% chance that humanity survives, does that include the probability of there being errors in your model? My model, no doubt, has many errors. The trouble, the trick would be an error someplace where that just makes everything work better, you know? Usually, when you're trying to build a rocket and your model of rockets is lousy, it doesn't cause the rocket to launch using half the fuel to go twice as far and land twice as precisely on target as your calculations laid. So. Though most of the room for updates is downwards, right? So, like, something that makes you think the problem is twice as hard, you go from, like, 99% to, like, 99.5%. If it's twice as easy, you go from 99% to 98%. Sure. Wait, wait, sorry. Yeah, but, like, most updates are not, this is going to be easier than you thought. That sure has not been the history of the last 20 years from that perspective. The most, you know, like, favorable updates, favorable updates is, like, yeah. Like, we went down this really weird side path where the systems are, like, legibly alarming to humans, and humans are actually alarmed at them, and maybe we get more sensible global policy. What is your model of the people who have engaged these arguments that you've made and you've dialogued with, but who have come nowhere close to your probability of doom? Like, what do you think they continue to miss? I think they're enacting the ritual of the young optimistic scientist who charges forth with no ideas of the difficulties and is slapped down by harsh reality and then becomes a grizzled cynic who knows all the reasons why everything is so much harder than you knew before you had any idea of how anything really worked. And they're just, like, living out that life cycle, and I'm trying to jump ahead to the end point. Is there somebody who has probability doom less than 50% who you think is, like, the clearest person with that view, who is, like, a view you can most sympathize with? No. Really? So, like, someone might say, listen, Eliezer, according to the CEO of the company who is, like, leading the AI race, I think he tweeted something that, like, you've done the most to accelerate AI or something, which was, assuming you like the opposite of your goals. And, you know, it seems like other people did see that these sort of language models very early on would scale in the way that they have scaled. Like, given that you didn't see that coming, and given that, I mean, in some sense, according to some people, your actions have had the opposite impact that you intended, like, what is the track record by which the rest of the world can come to the conclusions that you have come to? These are two different questions. One is the question of, like, who predicted that language models would scale? If they put it down in writing, and if they said, not just this loss function will go down, but also which capabilities will appear as that happens, then that would be quite interesting. That would be a successful scientific prediction. And if they then came forth and said, this is the model that I used. This is what I predict about alignment. We could have an interesting fight about that. Second, there's the point that if you try to rouse your planet to give it any sense that it is in peril, there are the idiot disaster monkeys who are like, ooh, ooh, this sounds like, like, if this is dangerous, it must be powerful, right? I'm going to be first to grab the poison banana. And what is one supposed to do? Should one remain silent? Should one let everyone walk directly into the whirling razor blades? If you sent me back in time, I'm not sure I could win this. But maybe I would have some notion of, like, ah, if you calculate the message in exactly this way, then this group will not take away this message. And you will be able to get this group of people to research on it without having this other group of people decide that it's excitingly dangerous and they want to rush forward on it. I'm not that smart. I'm not that wise. But what you are pointing to there is not a failure of ability to make predictions about AI. It's that if you try to call attention to a danger and not just have your whole planet walk directly into the whirling razor blades, carefree, no idea what's coming to them, maybe it's then, yeah, maybe that speeds up timelines. Maybe then people are like, ooh, ooh, exciting, exciting. I want to build it. I want to build it. Ooh, exciting. It has to be in my hands. I have to be the one to manage this danger. I'm going to run out and build it. Like, oh, no, if we don't invest in this company, who knows what investors they'll have and said that will demand that they move fast because the profit mode. And then, of course, they just move fast-fucking anyways. And yeah, if you sent me back in time, maybe I'd have a third option. It seems to me that in terms of what one person can realistically manage in terms of not being able to exactly craft a message with perfect hindsight that will reach some people and not others, at that point, you might as well just be like, yeah, just invest in exactly the right stocks and invest at exactly the right time. And you can fund projects on your own without alerting anyone. And if you keep fantasies like that aside, then I think that in the end, even if this world ends up having less time, it was the right thing to do rather than just letting everybody sleepwalk into death and get there a little later. If you don't mind me asking, what is the last five years or I guess even beyond that, I mean, what has being in this space been like for you, watching the progress and the way in which people have raised their head? The last five years? I made most of my negative updates as of five years ago. If anything, things have been taking longer to play out than I thought they would. But I mean, just like watching it, not as a sort of change in your probabilities, but just watching it concretely happen, what has that been like? Like continuing to play out a video game you know you're going to lose. Because that's all you have. If you wanted some deep wisdom from me, I don't have it. It's, I don't know. I don't know if it's what you'd expect, but it's like what I would expect it to be like. Where what I would expect it to be like takes into account that, I don't know, like, well, I guess I do have a little bit of wisdom. People imagining themselves in that situation raised in modern society, as opposed to raised on science fiction books written 70 years ago, might imagine, will imagine themselves like acting out their, like being drama queens about it. Like the point of believing this thing is to be a drama queen about it. And what I have in the way of culture is like the planet's at stake, bear up, keep going, no drama. The drama's meaningless. What changes the chance of victory is meaningful. The drama's meaningless. And what changes the chance of victory is meaningful. The drama's meaningless. Don't indulge it. Do you think that if you weren't around, somebody else would have independently discovered this sort of field of alignment, or? It's, that would be a pleasant fantasy for people who, like, cannot abide the notion that history depends on small little changes, or that people can really be different from other people. I've seen no evidence. But who knows what the alternate branches of Earth are like. But there are other kids who grew up on science fiction, so that can't be the only part of the answer. Well, I'm not surrounded by, well, I'm sure not surrounded by a cloud of people who are nearly Eliezer, outputting 90% of the work output. And you know, this is actually also, like, kind of not how things play out in a lot of places. Like, Steve Jobs is dead. Apparently couldn't find anyone else to be the next Steve Jobs of Apple, despite having really quite a lot of money with which to theoretically pay.
them. Maybe he didn't want to really want a successor. Maybe he wanted to be replaceable. I don't actually buy that, you know, based on how this has played out in a number of places. There was a person once who I met when I was younger who was like, had, you know, built something that, you know, like built an organization, and he was like, hey, hey, Eliezer, do you want to take this thing over? And I thought he was joking. And it didn't dawn on me until years and years later, after trying hard and failing hard to replace myself, that, oh, like, yeah, I could have maybe taken a shot at doing this person's job, and he'd probably just never found anyone else who could take over his organization, and maybe ask some other people, and like, nobody was willing. And I didn't really, you know, that's his tragedy, that he built something and now can't find anyone else to take it over. And if I'd known that at the time, I would not have, you know, I would have at least apologized. And yeah, to me, it looks like people are not dense in the incredibly multidimensional space of people. There are too many dimensions, and only eight billion people on the planet. The world is full of people who have no immediate neighbors, and problems that one person can solve, and then, like, other people cannot solve it in quite the same way. I don't think I'm unusual in looking around myself in that highly multidimensional space, and like, not finding a ton of neighbors relative to take, ready to take over. And I'm, I had, you know, four people, any one of whom could, you know, do like 99% of what I do, or whatever. I might retire. I am tired. Probably I wouldn't. Probably the, like, marginal contribution of that fifth person is still pretty large. But yeah, I don't know. I think the, there's the question of, like, well, did you occupy a place in mind space? Did you occupy a place in social space? Did people not try to become Eliezer because they thought Eliezer already existed? And some of my answer to that is, like, man, like, I don't think Eliezer already existing would have stopped me from trying to become Eliezer. But, you know, maybe you just, like, look at the next Everett branch over, and there's just, like, some kind of empty space that someone steps up to fill, even if then they don't end up with a lot of obvious neighbors. Maybe, maybe the world where I died in childbirth is just, you know, like, pretty much like this. But I don't feel, you know, if, if somehow we live to, to hear them, to hear the, to hear the answer about that sort of thing from some, someone or something that can calculate it. That's not the way I bet. But, you know, if it's true, it'd be funny. When I said no drama, that, that did include the concept of, I don't know, trying to make the story of your planet be the story of you. If it all would have played out the same way, and that's what, and somehow I survived to be told that, I'll laugh, and I'll cry, and that will be the reality. I mean, what I find interesting, though, is that in your particular case, your output was so public. And I mean, I don't know, like, for example, your sequences, you're like, you know, your fan, your science fiction and fan fiction, I'm sure like hundreds of thousands of 18 year olds read it or even younger. And presumably, some of them reached out to you. And they're like, you know, I think this way. I would love to learn more. I'll work on this. What was the problem that part? I mean, yes, part of how part of why I'm a little bit skeptical of the story where like, people are just like, infinitely replaceable is that I tried really, really, really hard to create like a new crop of people who could do all the stuff I could do to take over. Because, you know, I knew my health was not great and getting worse. I tried really, really hard to replace myself. I'm not sure where you look to find somebody else who tried that hard to replace himself. I tried, I really, really tried. That's what the less wrong sequences were. They had other purposes. But like, first and foremost, it was like me looking over my history and going like, well, I see all these blind pathways and stuff that took me a while to figure out. And there's got to be, you know, like, and I feel like I had these near misses on becoming myself. Like, there's got to be, like, you know, if I got here, there's got to be like 10 other people. And like, some of them are smarter than I am. And they just like need these like little boosts and shifts and hints and they can go down the pathway and, you know, like turn into super Eliezer. And, you know, that's what the sequences were. Like other people use them for other stuff. But primarily they were instruction manual to the young Eliezers that I thought must exist out there. And they're not really here. Other than the sequences, do you mind if I ask, like, what were the kinds of things you're talking about here in terms of training the next core of people like you? Just the sequences. I'm not, I'm not a good mentor. I did try mentoring somebody for a year once, but yeah, he didn't turn into me. So I picked things that were more scalable. I'm like most people, you know, like among the other reason why I don't see a lot of people trying that hard to replace themselves is that most people, you know, are like whatever their other talents don't happen to be like sufficiently good writers. I don't think the sequences were good writing by my current standards, but they were good enough. And, you know, most people, you know, do not happen to get a handful of cards that contains the writing card, you know, whatever else their other talents. I'll cut this question out if you don't want to talk about it, but you mentioned that there's like certain health problems that incline you towards retirement now. Is that something you want, you're willing to talk about? I mean, they caused me to want to retire. I doubt they will cause me to actually retire. And yeah, it's fatigue syndrome. Our society does not have good words for these things. The words that exist are tainted by their uses, labels, to categorize a class of people, some of whom perhaps are actually malingering, but mostly it says like we don't know what it means, and you know, you don't want everyone to have chronic fatigue syndrome on your medical record, because that just tells doctors to give up on you. And what does it actually mean besides being tired? If one wishes to walk home from work, if one wishes to, if one lives half a mile from one's work, then one had better walk home if one wants to go for a walk sometime in the day, not walk there. If you walk half a mile to work, you're not going to be getting very much work done the rest of that work day. And aside from that, these things don't have names. Not yet. Whatever the cause of this, is your working hypothesis that it has something to do, or is in some way correlated with the thing that makes you a liaisor? Or do you think it's like a separate thing? When I was 18, I made up stories like that. And it wouldn't surprise me terribly if you could get, if one survived to hear the tale from something that knew it, that the actual story would be a complex tangled web of causality, in which that was in some sense true. But I don't know. And storytelling about it does not hold the appeal that it once did for me. Is it a coincidence that I was not able to go to high school or college? Is there something about it that would have crushed the person that I otherwise would have been? Or is it just, in some sense, a giant coincidence? I don't know. Some people go through high school and college and come out sane. There's too much stuff in a human being's history to... And there's a plausible story you could tell. Like, maybe there's a bunch of potential liaisors out there, but they went to high school and college, and it killed them. It killed their souls. And you were the one who had the weird health problem, and you didn't go to high school, and you didn't go to college, and you stayed yourself. And I don't know. To me, it just feels like patterns in the clouds. And maybe that cloud actually is shaped like a horse. But what good does the knowledge do? What good does the story do? When you were writing the sequences and the fiction, from the beginning, was your goal to find somebody who... Like, the main goal, to find somebody who could replace you in specifically the task of AI alignment? Or did it start off with a different goal, and then... I mean, I thought there... I mean, like in 2008, I did not know that stuff was going to go down in 2023. For all I knew, there was a lot more time in which to do something like...
Like, build up civilization to another level, layer by layer. Sometimes civilizations do advance as they improve their epistemology. So there was that, there was the AI project. Those were the two projects, more or less. When did AI become the main thing? As we ran out of time to improve civilization. Was there a particular year that became the case for you? I mean, I think that 2015, 16, 17 were the years at which I'd noticed I'd been repeatedly surprised by stuff moving faster than anticipated. And I was like, oh, okay, like if things continue accelerating at that pace, we might be in trouble. And then like 2019, 2020, stuff slowed down a bit and, you know, there was more time than I was afraid we had back then. You know, that's what it looks like to be a Bayesian. Like your estimates go up, your estimates go down, they don't just keep moving in the same direction. Because if they keep moving in the same direction several times, you're like, oh, like I see where this thing is trending, I'm going to move here. And then like things don't keep moving that direction, they're like, oh, okay, like back down again. And that's what sanity looks like. I am curious, actually, like taking many worlds seriously, does that bring you any comfort in the sense that like there is one branch of the wave function where humanity survives? Or is that, do you not buy that sort of? I'm worried that they're pretty distant. Like I expect that at least they, I don't know, like I'm not sure it's enough to not have Hitler, but it sure would be a start on things going differently in a timeline. But mostly, I don't know, there's some comfort from thinking of the wider spaces than that, I'd say. As Tegmark pointed out way back when, if you have a spatially infinite universe that gets you just as many worlds as the quantum multiverse, if you go far enough in a space that is unbounded, you will eventually come to an exact copy of Earth or a copy of Earth from its past that then has a chance to diverge a little differently. So the quantum multiverse has nothing, reality is just quite, yeah, reality is just quite large. Is that a comfort? Yeah. Yes, it is. Possibly our nearest surviving relatives are quite distant, or you have to collect quite some waves through the space before you have worlds that survive, but anything but the wildest flukes, maybe our nearest surviving neighbors are closer than that. But look far enough and there should be like some species of nice aliens that were smarter or better at coordination and built their happily ever after. And yeah, that is a comfort. It's not quite as good as dying to yourself knowing that the rest of the world will be okay, but it's kind of like that on a larger scale. And weren't you going to ask something about orthogonality at some point? Did I not? Did you? At the beginning, when we talked about human evolution and… Yeah, that's not like orthogonality, that's the particular question of what are the laws relating optimization of a system via hill climbing to the internal psychological motivations that it acquires. But maybe that was all you meant to ask about. Well, can you explain in what sense you see the broader orthogonality thesis as unaffected by that? The orthogonality thesis is you can have almost any kind of self-consistent utility function in a self-consistent mind. Like many people are like, why would AIs want to kill us? Why would smart things not just automatically be nice? And this is a valid question, which I hope to at some point run into some interviewer where they are of the opinion that smart things are automatically nice that I can explain on camera. Why, like, although I myself held this position very long ago, I realized that I was terribly wrong about it and that all kinds of different things hold together and that if you take a human and make them smarter, that may shift their morality. It might even, depending on how they start out, make them nicer. But that doesn't mean that you can do this with arbitrary minds in an arbitrary mind space because all the different motivations hold together. That's like orthogonality, but if you already believe that, then there might not be much to discuss. No, no. I guess I wasn't clear enough about it. Is that, yes, all the different sort of utility functions are possible. It's that from the evidence of evolution and from the sort of reasoning about how these systems are being trained, I think that wildly divergent ones don't seem as likely as you do. But before I, instead of having to respond to that directly, let me ask you some questions I did have about it, which I didn't get to. One is actually from Scott Aronson. I don't know if you saw his recent blog post, but here's a quote from it. If you really accept the practical version of the orthogonality thesis, then it seems to me that you can't regard education, knowledge, and enlightenment as instruments for moral betterment. On the whole, though, education hasn't merely improved humans' abilities to achieve their goals. It has also improved their goals. I'll let you react to that. Yeah. I think that, yeah, if you start with humans, if you take humans, and possibly also for the requiring particular culture, but leaving that aside, you take humans who start out raised the way Scott Aronson was, and you make them smarter, they get nicer, it affects their goals. If you had, and there's a Less Wrong post about this, as there always is, well, several about really, but like sorting pebbles into correct heaps, describing a species of aliens who think that a heap of size 7 is correct, and a heap of size 11 is correct, but not 8, or 9, or 10. Those heaps are incorrect. And they used to think that a heap size of 21 might be correct, but then somebody showed them an array of 7 by 3 pebbles, 7 columns, 3 rows, and then people realized that 21 pebbles was not a correct heap. And this is the thing they intrinsically care about. These are aliens that have a utility function with, as I would phrase it, some logical uncertainty inside it. But you can see how, as they get smarter, they become better and better able to understand which heaps of pebbles are correct. And the real story here is more complicated than this. But that's the seed of the answer. Scott Aronson is inside a reference frame for how his utility function shifts as he gets smarter. It's more complicated than that. Human beings are made out of these, are more complicated than the pebble sorters. They're made out of all these complicated desires, and as they come to know those desires, they change. As they come to see themselves as having different options, it doesn't just change which option they choose after the manner of something with a utility function, but the different options that they have bring different pieces of themselves in conflict. When you have to kill to stay alive, you may come to a different equilibrium with your own feelings about killing than when you are wealthy enough that you no longer have to do that. And this is how humans change. As they become smarter, even as they become wealthier, as they have more options, as they know themselves better, as they think for longer about things and consider more arguments, as they understand perhaps other people and give their empathy a chance to grab onto something solider because of their greater understanding of other minds. But that's all when these things start out inside you. And the problem is that there's other ways for minds to hold together coherently where they execute other updates as they know more, or don't even execute updates at all because their utility function is simpler than that, though I do suspect that is not the most likely outcome of training a large language model. So large language models will change their preferences as they get smarter, indeed. Not just what they do to get the same terminal outcomes, but the preferences themselves will up to a point change as they get smarter. It doesn't keep going. At some point you know yourself sufficiently well and you are able to rewrite yourself, and at some point there, unless you specifically choose not to, I think that system crystallizes. We might choose not to. We might value the part where we just sort of change in that way, even if it's no longer heading in a knowable direction, because if it's heading in a knowable direction, you could jump to that as an end point. Wait, wait, so is that why you think AIs will jump to that end point, because they can anticipate where their moral updates are going? I would reserve the term moral updates for humans. These are, let's call them logical preference updates. Preference shifts. What are the prerequisites in terms of, like whatever makes Aaronson and other sort of smart moral people, or whatever, like preferences that we humans can sympathize with, like what is, you mentioned empathy, but what are the sort of prerequisites? They're complicated. There's not a short list. If there was a short list of crisply defined things where you could like, give it like chunk, chunk, chunk, and now it's in your moral frame of reference, then that would be the alignment.
plan. I don't think it's that simple, or if it is that simple, it's like in the textbook from the future that we don't have. Okay, let me ask you this. Are you still expecting a sort of chimps to humans gain in generality, even with these LLMs? Or does a future increase look of an order that we see from like GPT-3 to GPT-4? I'm not sure I understand the question. Can you rephrase? Yes. It seems that, I don't know, like from reading your writing from earlier, it seemed like a big part of your argument was like, look, a few, I don't know how many total mutations it was to get from chimps to humans, but it wasn't that many mutations. And we went from something that could basically get bananas in the forest to something that could walk on the moon. Are you expecting that? Are you still expecting that sort of gain eventually between, I don't know, like GPT-5 and GPT-6 or like some GPT-N and GPT-N plus one? Or does it look smoother to you now? Okay. So like, first of all, let me preface by saying that for all I know of how of the hidden variables of nature, it's completely allowed that GPT-4 was actually just it. This is where it saturates. It goes no further. It's not how I'd bet, but you know, but you know, if nature comes back and tells me that I'm not allowed to be like, you just violated the rule that I knew about. I know of no such rule prohibiting such a thing. I'm not asking whether these things will plateau at a given level of intelligence, whether there's a cap. That's not the question. Even if there is no cap, do you expect these systems to continue scaling in the way that they have been scaling? Or do you expect like some really big jump between some GPT-N and some GPT-N plus one? Yes. And yes. And that's only if things don't plateau before then. I mean, it's yeah. I can't quite say that I know what you know. I do feel like we have this like track of the loss going down as you add more parameters and you train on more tokens and a bunch of qualitative abilities that suddenly appear or like, I'm sure if you like zoom in closely enough, they appear more gradually, but like that appear as the successful releases of the system, which I don't think anybody has been going around predicting in advance that I know about. And like loss continue to go down unless it suddenly plateaus. New abilities appear. Which ones? I don't know. Is there at some point a giant leap? Well, if at some point it becomes able to like toss out the enormous training run paradigm and build more efficient and like jump to a new paradigm of AI, that would be one kind of giant leap. You could get another kind of giant leap via architectural shift, something like transformers, only there's like an enormously huger hardware overhang now, like something that is to transformers as transformers merge recurrent neural networks. And like maybe there's a maybe and then maybe the loss function suddenly goes down and you get a whole bunch of new abilities. That's not because like the loss went down on a smooth curve and you got like a bunch more abilities in a dense spot. Maybe there's like some particular set of abilities that is like a master ability, the way that language and writing and culture for humans might have been a master ability. And you like the loss function goes down smoothly, you get this one new like internal capability, there's a huge jump in output. Maybe that happens. Maybe stuff plateaus before then and it doesn't happen. Being an expert. Being the expert who gets to go on podcasts. They don't actually give you a little book with all the answers in it, you know. You're like just guessing based on the same information that other people have and maybe, maybe if you're lucky, slightly better theory. Yeah, that's what I'm wondering because you do have a different theory of like what fundamental intelligence is and what it entails. So I'm curious if like you have some expectations of where the GPTs are going. I feel like a whole bunch of my successful predictions in this have come from other people being like, oh yes, I have this theory which predicts that stuff is 30 years off. And I'm like, you don't know that. And then like stuff happens about 30 years off. And I'm like, haha, successful prediction. And that's basically what I told you, right? I was like, well, you know, like you could have the loss function continuing on a smooth line and new abilities appear and you could have them like suddenly appear to cluster because like why not? Because nature just tells you that's up and suddenly. You could have like this one key ability that's equivalent of language for humans and like there's a sudden jump in output capabilities. You could have like a new innovation like the transformer and maybe the loss has actually dropped precipitously and a whole bunch of new abilities appear at once. Now this is all just me. This is me saying I don't know. But so many people around are saying things that implicitly claim to know more than that, that it can actually sound like a startling prediction. This is one of my big secret tricks, actually. People are like, well, the AI could be like good or evil. So it's like 50-50, right? And I'm actually like, no, like we can be ignorant about a wider space than this in which like good is actually like a fairly narrow range. And so many of the predictions like that are really anti-predictions. It's somebody thinking in a relatively narrow line and you point out everything outside of that and it sounds like a startling prediction. Of course, the trouble being when you like, you know, look back afterwards, people are like, well, you know, like those people saying the narrow thing were just silly, haha. And they don't give you as much credit. I think the credit you would get for that rightly is as a good sort of agnostic forecaster, as somebody who is like sort of calm and measured. But it seems like to be able to make really strong claims about the future, about something that is so out of prior distributions is like the death of humanity. You don't only have to show yourself as a good agnostic forecaster. You have to show that your ability to forecast because of a particular theory is much greater. Do you see what I mean? It's all about, so when you work, yeah, it's all about the ignorance prior. It's all about knowing the space in which to be maximum entropy. Like the whole bunch of, you know, like somebody, you know, like what will the future be? Well, I don't know. It could be paperclips. It could be staples. It could be no kind of office supplies at all and tiny little spirals that it could be like little tiny things that are like outputting one, one, one because that's like the most predictable kind of text to predict. Or like representations of ever larger numbers in the fast-growing hierarchy because, you know, that's how they interpret the reward counter. I'm actually like getting into specifics here, which is kind of the opposite of the point I originally meant to make, which is like, you know, like if somebody claims to be very unsure, I might say, okay, so then like you expect like most possible molecular configurations of the solar system be equally probable. Well, humans mostly aren't in those. So like being very unsure about the future, it looks like predicting with probability nearly one that the humans are all gone, which, you know, it's not actually that bad, but it like illustrates the point of like people going like, but how are you sure? Kind of missing the real discourse and skill, which is like, oh yes, we're all very unsure. Lots of entropy in our probability distributions, but what is the space on for which you are unsure? Even at that point, it seems like the most reasonable prior is not that all sort of atomic configurations of the solar system are equally likely because I agree by that metric. Yeah. Like it's, it's like all computations that can like be run over configurations of solar system are equally likely to be maximized. Like we have a certain, we have certain sense that like, listen, we know what the loss function looks like. We know what the training data looks like. That obviously is no guarantee of what the drives that come out of that loss function will look like. Yeah. But it certainly came out pretty different from their loss functions. But I mean, this is the first question you began with. I would say like, I would say actually no. Like if it is as similar as humans are now to our loss function from which we evolved, that would be like, that honestly might not be that terrible world. And it might in fact be a very good world. Where do you get, where do you get good world out of maximum prediction of text? Plus, um, RLHF plus, uh, plus like all the, whatever alignment stuff that might work results in something that kind of just like does what you ask it to, uh, the way, like, does it reliably enough that, you know, we ask it like, Hey, help us with alignment. Then go, go stop asking for help with alignment. Ask it for any of the like, help us enhance our brains. Help us. Why are people asking for like the most difficult thing? That's the most possible to verify it's whack. Um, and then basically at that point we're like turning into gods and we can get to the point where you're turning into gods yourselves. You're like, you're not quite home free, but you know, you're sure past, you're sure past a lot of the death. Yeah. Maybe you can explain the intuition that all sorts of drives are equally likely given unknown loss function in a known set of data. Oh, um, if, if, yeah, like, so, so, so if you, if you had the textbook from the future, or if you were an alien who'd watched a dozen planets destroy themselves the way earth is that we're not actually doesn't, that's not like if you'd seen 10,000 planets destroy themselves the way earth has while being only human in your sample complexity and generalization ability, um, then you could be like, Oh yes, they're going to try like this trick with loss functions and they will get a draw from like this space of results. And the, and the alien like can now probably may now have like a pretty good prediction of like range of like where that ends up. Like, like similarly, like now that we've actually seen how humans turn out when you opt.
them for reproduction, it would like not be surprising if we found some aliens the next door over and they had orgasms. Now maybe they don't have orgasms but you know like some like but you know like if they had some kind of like strong surge of pleasure during the act of mating we're not we're not as we're not surprised. We've seen how that plays out in humans. If they have some kind of like weird food that isn't that nutritious but like makes them much happier than any kind of food that was more nutritious and around in their ancestral environment like like ice cream, we probably can't call it as ice cream right? It's not going to be like sugar, salt, fat, frozen. We're not specifically going to have ice cream. They might play go. They're not gonna play chess. Because chess is like more has more specific pieces right yeah. They're not gonna play they're not gonna play go on like 19 by 90. They might play go on some other size. Probably odd. Well can we really say that? I don't know. I bet on like an odd if they play go I bet on an odd board dimension at let's say two that's the two-thirds. The Ploss's rule of succession. Sounds about right. Unless there's some other reason why go just totally does not work on an even board dimension that I don't know because I'm insufficiently acquainted with the game. The point is like reasoning off of humans is like pretty hard. We have like the loss function over here. We have like humans over here. We can like look at the rough distance like all the like weird specific should like stuff that humans are created around and be like you know like like like if the loss function is over here and humans are over there like maybe aliens are like over there. And if we had like three aliens that would like expand our views of the possible. We'd have like or even two aliens would like vastly expand our views of the possible. And give us like a much stronger notion of what the third aliens look like like humans aliens third race. But you know yeah like you know like the the wild optimistic scientists have never been through the never been through this with with AI. So they're like oh yeah like like you optimize the AI to like say nice things and help you and like make it a bunch smarter probably says nice things and helps you is probably like totally aligned. Yeah. Yeah they don't know any better. Not trying to jump ahead of the story. But the aliens the aliens know know where you end up around the loss function. They know how it's gonna play out. Much more narrowly. We're guessing much more blindly. It just like it just leaves me in a sort of unsatisfied place that we apparently know about something that is so extreme that maybe a handful of people in the entire world believe it from first principles about you know the doom of humanity because of AI. But this theory that is so so productive in that one very unique prediction is unable to give us any sort of other prediction about what this world might look like in the future or about what happens before the before we all die. Like it can tell us nothing about the world until the point at which makes a prediction that is the most remarkable in the world. You know rationalists should win but rationalists should not win the lottery. I'd ask you like what other theories are supposed to they have been doing a like amazingly better job of predicting the last three years. You know maybe it's just hard to predict right. And and in fact it's like easier to predict the end state than the strange complicated wending paths that lead there. Much like if you could you know play against AlphaGo and predict it's gonna like be in the class of winning board states but not not exactly how it's going to beat you. It's like not quite like that the problem difficulty of predicting the future. But you know from my perspective the future is just like really hard to predict. And there's a few places where you can like wrench what sounds like an answer out of your ignorance. Even though really you're just being like well you're gonna like end up in some like random weird place around this loss function and I haven't seen it happen with 10,000 species so I don't know where. Very very impoverished by the from the standpoint of anybody who like actually knew anything that actually predict anything. But the rest of the world is like oh like we're easily we're like equally likely to to win the lotteries lose the lottery right? Like either we win or we don't. You come along and you're like no no your chance of winning the lottery is tiny. They're like what? How can you be so sure? Where do you get your strange certainty? And the actual root of the answer is that you are putting your maximum entropy over a different probability space. Like that just actually is the thing that's going on there. You're saying all lottery numbers are equally likely instead of winning and losing are equally likely. So I think the place to sort of close this conversation is let me just sort of give the the main reasons why I'm not convinced that doom is likely or even that it's more than 50% probable or anything like that. Some are the things that I started this conversation with that I don't feel like I heard any knockdown arguments against and some are new things from the conversation. And the following things are things that even if any one of them individually turns out to be true I think doom doesn't make sense or is much less likely. So going through the list I think probably more likely than not this entire frame all around alignment and AI is wrong. This is maybe not something that would be easy to talk about but I'm just kind of skeptical of sort of first principles reasoning that has really wild conclusions. Okay so everything in the solar system just ends up in a random configuration then. It stays like it is unless you have very good reasons to think otherwise and especially if you think it's going to be very different from the way it's going you must have very very good reasons like ironclad reasons for thinking that it's going to be very very different from the way it is. Uh-huh so this is you know the humanity hasn't really existed for very... man I don't even know what to say to this thing. We're like this tiny like everything that you think of as normal is this tiny flash of things being in this particular structure out of a 13.8 billion year old universe which very little of which was like 20th century pardon me 21st century. Yeah my own brain sometimes gets stuck in childhood too right? Very little of which is like 21st century like civilized world the you know on this like little fraction of the surface of one planet in a vast solar system most of which is not Earth and a vast universe most of which is not Earth and it has lasted for like such a tiny period of time through such a tiny amount of space and and has like changed so much over you know just the last 20,000 years or so and and and here you are like being like why would things really be any different going forward? I feel like that arguments proves too much because you could use that same argument like somebody says comes up to me and says I think I don't know a theologian comes up to me and says like the Rapture is coming and let me sort of explain why the Rapture is coming and I say I'm not claiming that your arguments are as bad as the argument for Rapture I'm just just follow the example but then they say listen I mean look at how wild human civilization has been would it be any wilder if there was a Rapture and I'm like yeah actually as well as human civilization has been the Rapture would be much wilder. Because it violates the laws of physics. Yes. I'm not trying to violate the laws of physics even as we presently know them. How about this somebody comes up oh you know I got the perfect example okay somebody comes up to me he says we have actually nanosystems right behind you he says I've read Eric Drexler's nanosystems I've read Feynman's there's plenty of room at the bottom and he explains two things are not to mentored but go on okay fair enough he comes to me and he says let me explain to you my first principles argument about how some nanosystems will be replicators and the replicators because of some competition yada yada yada argument they turn the entire world into goo just making copies of themselves this kind of happened with humans you know well life generally yeah yeah but so then they say like listen as soon as people start building nanosystems pretty soon 99% probability the entire world turns into goo just because the replicators are the things that turn things into goo there will be more replicators and non replicators I don't have an object level debate about that but it just like I just started that I'm looking like yes human civilization has been wild but the entire world turning into goo because of nanosystems alone just seems much wilder than human civilization you know this this list this this impact this this argument probably lands with greater force on somebody who does not expect stuff to be disassembled by nanosystems albeit intelligently controlled ones rather than goo in like quite near future especially on the 13.8 billion year timescale but you know do you do you expect this little momentary flash of what you call normality to continue do you expect the future to be normal I know I expect any given vision of how things shape out to be wrong especially it is not like you are suggesting that the current weird trajectory continues being weird in the way it's been weird and that we continue to have like two percent economic growth or whatever and that leads to incrementally more technological progress and so on you're suggesting there's been that specific species of weirdness
which means that this entirely different species of weirdness is born. Yeah, we've got like different weirdnesses over time. The jump to superintelligence does strike me as being significant in the same way as first self-replicator. First self-replicator is the universe transitioning from you see mostly stable things to you also see a whole bunch of things that make copies of themselves. And then somewhat later on, there's a state where there's this strange transition, this border between the universe of stable things where things come together by accident and stay as long as they endure to this world of complicated life. And that transitionary moment is when you have something that arises by accident and yet self-replicates. And similarly, on the other side of things, you have things that are intelligent making other intelligent things. But to get into that world, you've got to have the thing that is built just by things copying themselves and mutating and yet is intelligent enough to make another intelligent thing. Now, if I sketched out that cosmology, would you say, no, no, I don't believe in that? What if I sketched out the cosmology of, because of replicators, blah, blah, blah, intelligent beings, intelligent beings create nanosystems, blah, blah, blah. No, no, no, no. Don't tell me about your like, not the proofs too much. I just want to like, like I discussed out of cosmology, do you buy it? In the long run, are we in a world full of things replicating or a world in a full of intelligent things designing other intelligent things? Yes. So you buy that vast shift in the foundations of order of the universe that instead of the world of things that make copies of themselves, imperfectly, we are in the world of things that are designed and were designed. You buy that vast cosmological shift I was just describing, the utter disruption of everything you see that you call normal down to the leaves and the trees around you. You believe that. Well, the same skepticism you're so fond of that argues against the rapture can also be used to disprove this thing you believe that you think is probably pretty obvious, actually, now that I've pointed it out. Okay. Your skepticism, your skepticism disproves too much, my friend. That's actually a really good point. It still leaves open the possibility of like how it happens and when it happens, blah, blah, blah. But actually, that's, that's a good point. Okay. So a second, second thing. You set them up, I'll knock them down one after the other. Second thing is wrong. I'm sorry. I was just jumping head to the predictable update at the end. Maybe alignment just turns out to be much simpler or like much easier than we think. It's not like we've as a civilization spent that, that much resources or brainpower in solving it. If we put in even the kind of resources that we put into elucidating strength theory or something into alignment, it could just turn out to be like, yeah, that's enough to solve it. And in fact, in the current paradigm, it turns out to be simpler because, you know, they're sort of pre-trained on human thought. And that might be a simpler regime than something that just comes out of a black box that like, you know, like alpha zero or something like that. So like, some of my like, could I be wrong in an understandable way to me in advance mass, which is not where most of my hope comes from is on, you know, what if RLHF just works well enough and the people in charge of this are not the current disaster monkeys, but instead have some modicum of caution and are using their like, like know what to aim for an RLHF space, which the current crop do not. And I, you know, I'm not really that confident of their ability to understand if I told them, but maybe you have some folks who can understand or anyways, I can sort of see what I try. These people will not try it. But you know, the current crop that is, and I'm not actually sure that if somebody else takes over like the government or something that they listened to me either, but I can, you know, maybe you. So some of the trouble here is that you have a choice of targets and like, neither is all that great. One is you look for the niceness that's in humans and you try to bring it out in the AI. And then you with its cooperation, cause you know, it knows that if it makes it that if you try to just like amp it up, it might not stay all that nice. Or that if you build a successor system to it, it might not stay all that nice. And it doesn't want that because you, you know, like you, you, you, you, you, you narrow down the Shaggoth enough, you know, that's not like, and, and the mat, you know, somebody, somebody once had this incredibly profound statement that I think I like somewhat disagree with, but it's still so incredibly profound. It's consciousness is when the mask eats the Shaggoth. And maybe that's it. Maybe, you know, with the, with the right set of bootstrapping reflection type stuff, stuff, stuff, you can have that happen on purpose more or less where there, where the, the systems output that you're shaping is like to some degree in control of the system. And you, you, you locate niceness in the human space. I, I have fantasies along the lines of what if you trained GPTN to distinguish people being nice and saying sensible things and argue validly and, you know, can't just, I'm not sure that works. If you just have Amazon Turks, try to label it. You just get the like strange thing you located that RLHF located in the present space, which is like some kind of weird corporate speak like left rationalizing leaning strange telephone announcement creature is what they got with the current crop of RLHF. Note how this stuff is weirder and harder than people might have imagined initially. But, you know, leave, leave, leave aside that the, the, the part where you try to like jumpstart the entire process of turning into a grizzled cynic and update as hard as you can do it in advance. Leave that aside for a moment. Like maybe you can look, maybe you are like able to train on Scott Alexander. And so you want to be a wizard. Some other night and some other nice real people, real people and nice fictional people, and separately train on what's valid argument. That's, that's going to be, that's gonna be tougher. But you know, I could probably put together a crew of a dozen people who could provide the data on that RLHF. And you find like the nice creature and you find the nice mask that that's that argues validly you do some more complicated stuff to try to boost the thing where it's like eating the shag off where that's what the system is and that's like more what the system is less what it's pretending to be. But I do seriously think this is like, like, I can say this and like the disaster monkeys at the current places can cannot along to it, but they have not said things like this themselves that I've ever heard. And that is not a good sign. But and then like if you don't have this up too far, which on the present paradigm you like can't do anyways, because if you like train the very, very smart person of this version of the system, it kills you before you can RLHF it. But like maybe you can like train TPT to like distinguish like, nice, valid, kind, careful, and then like, filter all the training data to get the nice things to train on and then train on that data rather than training on everything to try to like avert the Waluigi problem. It's or just more generally having like all the, the darkness in there, like just train on the light that's in humanity. So there's like that kind of course, and not if you don't push that too far, maybe you can get a genuine ally. And maybe things play out differently from there. That's like one of the little rays of hope for but but that's not, I don't think that actually looks like alignment is so easy that you you just get whatever you want. It's a genie gives you what you wish for. I don't think that that doesn't even strike me as hope. Honestly, the way you described it seemed kind of compelling. Like I don't know why that doesn't even rise to 1%. The possibility works out that way. I literally this this is like literally my, you know, my like AI alignment fantasy from 2003. Though not with but not with like RLHF and as the implementation method or LLMs as the base. And it's, you know, going to be more dangerous than when I was thinking about when I was dreaming about in 2003. And I think in a very real sense, it feels to me like the the the the people doing this stuff now literally not gotten as far as I was in 2003. And, you know, I can like, I've now like written out my answer sheet for that. It's on the podcast, it goes on the internet, and now they can now they can pretend that that was their idea. Or like, or like, sure, that's obvious. We're going to do that anyways. And yeah,
They didn't say it earlier. And you can't run a big project off of one person who, it failed to gel. The alignment field failed to gel. That's my gesture to the like, well, you just thrown a ton of more money, and then it's all solvable. Because I've seen people try to amp up the amount of money that goes into it. And the stuff coming out of it has not gone to the places that I would have considered obvious a while ago. And I can print out all my interest receipts for it. And each time I do that, it gets a little bit harder to make the case next time. But I mean, how much money are we talking in the grand scheme of things? Because civilization itself has a lot of money. I know people who have a billion dollars. I don't know how to throw a billion dollars at outputting lots and lots of alignment stuff. But you might not. But I mean, you are one of 10 billion, right? And other people go ahead and spend lots of money on it anyways. And everybody makes the same mistakes. Nate Sorries has a post about it. I forget the exact title. But everybody coming into alignment makes the same mistakes. Let me just go on to the third point, because I think it plays into what I was saying. The third reason is, if it is the case that these capabilities scale in some constant way, as it seems like they're going from 2 to 3 or 3 to 4. What does that even mean? But go on. They get more and more general. It's not like going from a mouse to a human or a chimpanzee to a human. It's like going from- GPT-3 to GPT-4? Yeah. Well, it just seems like that's less of a jump than chimp to human, like a slow accumulation of capabilities. There are a lot of S-curves of emergent abilities. But overall, the curve looks sort of- Man, I feel like we bit off a whole chunk of chimp to human in GPT-3.5 to GPT-4. But go on. Regardless, OK, so then this leads to human-level intelligence for some interval. I think that I was not convinced from the arguments that we could not have a system of sort of checks on this the same way you have checks on smart humans, that it would try to deceive us to achieve its aims any more than smart humans are in positions of power try to do the same thing. For a year. What are you going to do with that year before the next generation of systems come out that are not held in check by humans because they are not roughly in the same power intelligence range as humans? Maybe you can get a year like that. Maybe that actually happens. What are you going to do with that year that prevents you from dying the year after? One is, one possibility is that because these systems are trained on human text, maybe just progress just slows down a lot after it gets to slightly above human level. Yeah, that's not how, I would be quite surprised if that's how anything works. Why is that? For one thing, because it's like for an alien to be an actress playing all the humans on the internet. For another thing, well, first of all, you realize in principle that the task of minimizing losses on predicting human text does not have a, yeah, you understand that in principle this does not stop when you're as smart as a human, right? Like you can see that the computer science of that. I don't know if I see the computer science of that, but I think I probably understand the argument. OK, so like somewhere on the internet is a list of hashes followed by the string hashed. This is a simple demonstration of how you can go on getting lower losses by throwing a hypercomputer at the problem. There are pieces of text on there that were not produced by humans talking in conversation, but rather by like lots and lots of work to determine, extract experimental results out of reality. That text is also on the internet. Maybe there's not enough of it for the machine learning paradigm to work. But I'd sooner buy that the GPT systems just bottleneck short of being able to predict that stuff better rather than that. But you can maybe buy that. But the notion that you only have to be smart as a human to predict all the text is the internet, as soon as you turn around and stare at that a bit, it's just transparently false. OK, agreed. OK, how about this story? You have something that is sort of human-like that is maybe above humans at certain aspects of science because it's specifically trained to be really good at the things that are on the internet, which is like chunks and chunks of archive and whatever, whereas it has not been trained specifically to gain power. And while at some point of intelligence that comes along, can I just restart that whole sentence? No, you have spoken it. It exists. It cannot be called back. There are no take backs. There is no going back. There is no going back. Go ahead. OK, so here's another story. I expect them to be better than humans at science than they are at power seeking because we had greater selection pressures for power seeking in our ancestral environment than we did for science. And while at a certain point, both of them come along as a package, maybe that they can be at varying levels. But anyways, so you have this sort of early model that is kind of human level except a little bit ahead of us in science. You ask it to help us align the next version of it. Then the next version of it is more aligned because we have its help. And the sort of like this inductive thing where the next version helps us align the version. Where do people have this notion of getting AIs to help you do your AI alignment homework? Why can we not talk about having it enhance human intelligence instead? OK, so either one of those stories where it just like helps us enhance humans, enhance humans and help us figure out the alignment problem or something like that. Yeah, I it's like kind of weird because, you know, like like small, large amounts of intelligence don't automatically make you a computer programmer. And if you are a computer programmer, you don't automatically get security mindset. But it feels like there's some level of intelligence we ought to automatically get security mindset. And I think that's about how hard you have to augment people to have them able to do alignment, like the level where like they have security mindset, not because they were like special people with security mindset, but just because like they're that intelligent that you just like automatically have security mindset. I think that's about the level of where a human could start to work on alignment, more or less. Why is that story then not one percent get you to one percent probability that it helps us afford the whole crisis? Well, because it's not just a question of the technical feasibility of can you build a thing that applies its general intelligence narrowly to the neuroscience of augmenting humans? It's a question of like, so like, like one, I feel like that that is like probably like over one percent technical feasibility. But the world that we are in is so far, so far from from from doing that, from trying, trying the way the word could actually work, like, like, not like the try where like, oh, you know, like, well, we like we'd like just like do a bunch of RLHF to try to have a spit out output about this things, but not about that thing. And, you know, that that that no, no, not that. Yeah, what one percent that we could that humanity could could do that if it if it tried and tried in just the right direction, as far as I can. Perceive angles in this space. Yeah, I'm over one percent on that. I am I'm not very high on us on us doing it. Maybe I will be wrong. Maybe the time article article I wrote saying shut it all down gets picked up. And there are very serious conversations and the very serious conversations that are actually effective in in shutting down the headlong plunge. And there is a narrow exception carved out for the kind of narrow application of trying to build an artificial general intelligence that applies its intelligence narrowly and to the problem of augmenting humans. And, you know, that I think might be a harder sell to the world. And just shut it all down. They could they could get shut it all down and then not do the things that they would need to do to have an exit strategy. I feel like even if you told me that they would that they went for shut it all down, I would be like, then next, expect them to have no exit strategy until the world ended anyways. But perhaps I underestimate them. Maybe there's a will in humanity to do something else, which is not that. And if there really were. Yeah, I'm I think I'm even over 10 percent that that that would be a technically feasible path if we if we if they looked in just the right direction. But I am not over 50 percent on them actually doing the shut it all down. I am not if they do that, I am not over 50 percent on they're really, truly being the will of something else that is not that to really have an exit strategy. Then from there, you have to go in at sufficiently the right angle to to materialize the technical chances and not do it in the way that's just ends up a suicide or if you're lucky, like gives you the clear warning signs and and then people will do it.
will actually pay attention to those instead of just optimizing away the warning signs. And I don't want to make the sound like the multiple stage fallacy of like, oh, no, it's more than one thing has to happen. Therefore, the resulting thing can never happen, which, you know, like, super clear case in point of why you cannot prove anything will not happen this way of Nate Silver arguing that Trump needed to get through six stages to become the Republican presidential candidate, each of which was less than half probability and therefore he had less than one 64th chance of becoming the Republican, not one eighth was six, six stages of doing. Therefore, he had like less than one 64th chance of becoming, I think, just a Republican candidate, not not winning. So yeah, so like, you can't just like break things down into stages and then say, therefore, the probability is zero, you can break down anything at the stages. But like, but like, even so you're asking me like, well, like, isn't over 1% that it's that it's possible? I'm like, yeah, possibly even over 10%. That that doesn't get me to. Because because the the, like the reason why, you know, go ahead and tell people, yeah, don't don't put your hope in the future, you're probably dead, is that the the existence of this technical ray of hope, if you do just the right things, it's not the same as expecting that the world reshapes itself to permit that to be done without destroying the world in the meanwhile. I expect things to continue on largely as they have. And, you know, and what distinguishes that from despair is that at the moment, people were telling me, like, no, no, if you go outside the tech industry, people will actually listen. I'm like, all right, let's try that. Let's write the time article. Let's jump on that. Let's see if it works. It will lack dignity not to try. But that's not not the same as expecting as being like, oh, yeah, I'm over 50%. They're totally gonna do it. That time article is totally gonna take off. I'm not currently not over 50%, not over 50% on that. You know, you said like, any one of these things could mean and yet, like, even if this thing is technically feasible, that doesn't mean the world's going to do it. We are presently quite far from the world being on that trajectory, or doing the things that we needed to be created to create time to pay the alignment tax to do it. Maybe the one thing I would dispute is how many things need to go right from the world as a whole for any one of these paths to succeed. Because, which goes into the fourth point, which is that maybe the sort of universal prior over all the drives that an AI could have is just like the wrong way to think about it. And this is something that I mean, you definitely want to use the alien observation of 10,000 planets like this one prior for what you get after training on like thing x. Just like, especially when we're talking about things that have been trained on, you know, human text, I'm not saying that it was a mistake earlier on the conversation for me to say they'll be like the average of human motivations, whatever that means. But it's not, it's not inconceivable to me that it would be something that is very sympathetic to human motivations, having been having sort of encapsulated all of our output. I think it's much easier to get a mask like that than to get a shoggoth like that. Possibly, but again, this is something that seems like, I don't know, probability output on it at least 10%. And that by just by default, that is something that is not so it is not incompatible with the flourishing of humanity. Like, well, why? What is the utility function you hope it has that has its maximum? There's so many possible like, name three, name one, spell it out. It, I don't know, wants to keep us as a zoo, the same way we keep like other animals in a zoo. This is not the best outcome for humanity, but it's just like something where we survive and flourish. Okay, whoa, flourish. Keeping in a zoo did not sound like flourishing to me. Zoo was the wrong word to use there. Well, whoa, whoa, whoa, whoa, because it's not what you wanted. Why is it not a good prediction? You just asked me to name three. You didn't ask me like, No, no, what I'm saying is like, like, you're like, Oh, like prediction. Oh, no, no, I don't like my prediction. I want a different prediction. You didn't ask for the prediction. You just asked me to name them. Like name possibilities. I had meant like possibilities into which you put some probability I had meant for like, like a thing that you thought held together. This is the same thing as when I ask you, like, what is the specific utility function it will have that will be incompatible with, you know, humans existing? It's like your modal prediction. The super vast majority of predictions are of utility functions are incompatible with human existing. I can make a mistake and I'll still be incompatible with humans existing. Right, like I can just be like, you know, like, I can just like describe a randomly rolled utility function and up with something incompatible with humans exist. So like at the beginning of human evolution, you could think like, okay, this thing will become generally intelligent. And what are the odds that it's flourishing on the planet will be compatible with the survival of spruce trees or something? It's like the long term, we sure aren't. I mean, like maybe if we win, we'll have there be a space for spruce trees. Yeah, so as long as you can have spruce trees, as long as the mitochondrial liberation front does not object to that. What is the mitochondrial liberation front? Is that if you have, you have to have, you know, sympathy for the mitochondria enslaved, working all their lives to the benefit of some other organism. So this is like some weird hypothetical, like for hundreds of thousands of years, general intelligence has existed on Earth. You could say like, is it compatible with some random species that exists on Earth? Like, is it compatible with spruce trees existing? And I know, but you probably chopped down a few spruce trees, but. And the answer is yes, as a very special case of us being the sort of things that would make some of us would maybe conclude that we wanted that we specifically wanted spruce trees to go on existing, at least on Earth, in the glorious transhuman future. And their votes winning out against those of the mitochondrial liberation front. I guess since part of the sort of transhumanist future is part of the thing we're debating, it seems weird to assume that as part of the question. Well, the thing I'm trying to say is you're like, well, like, if you looked at the humans, would you like not expect them to end up incompatible with the spruce trees? I'm being like, sir, you, a human have like looked back and like looked at how humans wanted the universe to be and been like, well, would you not have anticipated in retrospect that humans would like want the universe to be otherwise. And I agree that we like might want to conserve a whole bunch of stuff. Maybe we don't want to conserve the parts of nature where things bite other things and inject venom into them and the victims die in terrible pain. Maybe even if maybe, you know, I think that many of them don't have qualia. This is disputed. Some people might be disturbed by it, even if they didn't have qualia. We might want to be polite to the sort of aliens who would be disturbed by it because they don't have quality and they just see like things don't want venom injected into them there before they should not have venom. We might conserve some parts of nature. But again, it's like it's like it's like firing an arrow and then drawing a circle around the target. I would disagree with that because again, this is similar to the example we started off the conversation with. But it seems like you are reasoning for from what might happen in the future. And because we disagree about what might happen in the future. In fact, the entire point of this disagreement is to test what will happen in the future, assuming what will happen in the future as part of your answer. Seems like I mean, yeah, that way to. OK, but then you're like claiming things as evidence for your position based on what exists in the world now that are not evidence that are not evidence one way or the other, because the basic prediction is like if you offer things enough options, they will they will go out of distribution. Like if you are like it's like it's like pointing to the like very first people with language and being like they haven't taken over the world yet. It's like and like they have like not gone way out of distribution yet. And it's like they haven't had general intelligence for long enough to accumulate the things that would give them more options such that they could start trying to select the weirder options. The prediction is like when you have when you give yourself more options, you start to select ones that look weird or relative to the ancestral distribution. As long as you don't have the weird options, you're not going to make the weird choices. And if you say like we haven't yet observed your future, that's fine. But like acknowledge that then like evidence against that future is not being provided by the past. This is the thing I'm saying there. You look around, it looks so normal. According to you who grew up here, if you've grown up a millennium earlier, your argument for the persistence of normality might not seem as persuasive to you after you'd seen that much change. This is a separate argument, though, right? Like I'm like, like, look at all this stuff humans haven't changed yet. You say now selecting the stuff we haven't changed yet. But if you go back 20,000 years and be like, look at the stuff intelligence hasn't changed yet, you might very well select a bunch of stuff that was going to fall 20,000 years later is the thing I'm trying to gesture at here. But so like, how do you propose we reason about what general intelligence to do when the world we look at after hundreds of thousands of years of general intelligence is the one that we can't use for evidence? Because, yeah, dive under the surface, look at the things that have changed. Why did they change? Look at the processes that are generating those choices. And since we have sort of these different functions of like where that goes, like look at the thing with ice cream. Look.
at the thing with condoms look at the thing with pornography see where this is going I think just like I it just seems like I would disagree with your intuitions about like what future smarter humans will do even with more options I was like in the beginning of conversation I disagreed that they would most humans would adopt some sort of like a transhumanist way to get better DNA or something but you would so yeah you just let you just like look down at your fellow humans you have like no confidence in their ability to tolerate weirdness even if they wonder like you do do you think what do you think would happen if we did a poll right now I think I'd have to explain that poll pretty carefully because you know they haven't got the the intelligence head bends yet right I mean we could do a Twitter poll with like a long explanation in it 4,000 character Twitter poll yeah like I mean man I like some somewhat tempted to do that just for the sheer chaos and point out the drastic selection effects of a it's my Twitter to followers be they read through a 4,000 character tweet I I feel like this is not likely to be truly very informative by my standards but part of me is amused by the prospect for the chaos yeah or I could do it on my end as well although my yeah it'd be weird as well yeah but plus plus you wouldn't like really I worry you wouldn't sell that transhumanism thing as well as I could a word it does like you just send me the wording but anyways that's a break but anyways given that we disagree about what in the future general intelligence will do where do you suppose we should look for evidence about what the general intelligence will do given our different theories about it if not from the present I mean I think you look at the mechanics you say as people have gotten more options they have gone further outside the ancestral distribution and we zoom in and it's like there's all these different things that people want and there's this like narrow range of options that they had 50,000 years ago and the things that they want have maxima or optima 50,000 years ago at stuff that coincides with reproductive fitness and then as a result of the humans getting smarter they start to accumulate culture which produces changes on a timescale faster than natural selection runs although it is still running contemporaneously the humans are just running fast running faster than natural selection it didn't actually halt and the additional they like generate additional options not blindly but according to the things that they want and they invent ice cream they you know like not not at random it doesn't just like a coughed up at random they're like searching the space of things that they want and and generating new options for themselves that optimize these things more that weren't in the ancestral environment and good hearts law applies what good hearts curse applies once you that like as you apply optimization pressure the correlations that were found naturally come apart and aren't present in the thing that gets optimized for like you know you like just give some people some tests who've never gone to school the ones who high score high in the test will know the problem domain because they you know like you just like give gives a bunch of carpenters a carpentry test the ones who score high in the carpentry test will like know how to carpenter things then you're like yeah I'll like pay you for high scores in the carpentry test I'll give you this carpentry degree and like people like oh I'm gonna like optimize the test specifically and they'll get higher scores than the carpenters and do and be worse at carpentry because they're like optimizing the test and that's the story behind ice cream and and you zoom in and look at the mechanics and not be like grand scale view because the grand scale of you just like never gives you the right answer basically like anytime you ask what would happen if you applied the grand scale view velocity in the past it's always just like oh I don't see that why this thing would change oh it changed who how weird who could have possibly expected that maybe you have a different definition of grand scale view because I would have thought that that is what you might use to categorize your own view but I don't want to get it caught up in semantics mine is zooming in it's looking at this look in the mechanics that's that's how I present it if we are like so far to distribution of natural selection as you say we're currently not we're currently nowhere near as far as we could be like well this is not the glorious transhuman future I claim that if even if we get much smarter as like if humans get much smarter through brain augmentation or something then there will still be spruce trees in like millions of years in the future and and if you still want to come the day I don't think I myself would oppose it unless there would be like distant aliens who are very very sad about what we were doing to the mitochondria and then I don't want to ruin their day for no good reason but but the reason that it's important to state it in the form of like given human psychology spruce trees will still exist is because that is the one evidence of sort of generality arising we have and even after millions of years of that generality like we think that spruce trees would exist I feel like we would be in this position of spruce trees in comparison to the intelligence we create and sort of the universal prior and whether spruce trees would exist doesn't make sense to me okay so but do you see how this perhaps leads to like everybody's severed heads being kept alive in jars on its own premises as opposed to humans getting the glorious transhumanist future no no they have the bogus but glorious transhumanist future those are not real spruce trees you know like you're talking about like plain old spruce trees you want to exist right not not the sparkling giant spruce trees with built-in rockets you're talking about humans being kept as pets in their ancestral state forever maybe being quite sad maybe they still get cancer and die of old age and they never get anything better than that does it keep us around as we are right now do we relive the same day over and over again maybe this is the day when that happens you see how it's like how the the general trend I'm trying to point out to here is you like have a rationalization for why they might do thing that is allegedly nice and I'm saying like why exactly are they wanting to do thing well if they want to do thing for this reason maybe there is a way to do this thing that isn't as nice as you're imagining and this is systematic you're imagining reasons they might have to give you nice things that you want but they are not you not unless we get you know not unless we get this like exactly right and they actually care about the part where you want some things and not others you are not describing something you are doing for the sake of the spruce trees do spruce trees have diseases in this world of yours do the diseases got to live do they get to live on spruce trees and it's not just and it's not a coincidence that that I can like zoom in and poke at this and ask questions like this and that you did not ask these questions of yourself you are imagining nice ways you can get the thing but reality is not necessarily imagining how to give you what you want and the AI is not necessarily imagining how to give you what you want and these and like for everything you can be like oh like hopeful thought maybe I get all the stuff I want because the AI reasons like this because it's the optimism inside you that is generating this answer and that if the optimism is not the AI if the op if the AI is not specifically being like well how do I pick a reason to do what I to do things that will give this person a nice outcome you're not going to get the nice outcome you're going to be reliving the last day of your life over and over it's going to like create old or maybe creates old-fashioned humans ones from 50,000 years ago maybe that's more quaint maybe maybe maybe it's just like just as happy with like bacteria because there's more of them and and that's equally old-fashioned you're going to create the specific spruce tree over there maybe from its perspective you know like a generic bacterium is just as good a form of life as like the generic spruce tree is of a spruce tree and like the end and like this is not specific to the example that you gave it's it's it's it's me being like well suppose we like took criterion that sounds kind of like this and asked how do we actually maximize it what else satisfies it not just you're like trying to argue the AI into doing what you think is a good idea by giving the AI reasons why it should want to do the thing under like some set of like hypothetical motives but anything like that if you like optimize it on its own terms without narrow down to where you want it to end up because it actually felt nice to you the way that you define niceness like it's all going to have somewhere else somewhere that isn't as nice something maybe where we'd be like sooner scour the surface of the planets the clean with nuclear fire rather than cut but let that AI come into existence though I do think those are also probable because you know instead of hurting you as is you know there's like something more efficient for it to do that maxes out its utility function okay I acknowledge that you had a better argument there but here's another intuition I'm curious how you responded that earlier we talked about the idea that like if you bred humans to be friendlier and smarter this is not where I'm going with this but like if you did that I think I want to register for the record that the term breeding humans would would cause me to like look askance again and any at alien and at any aliens who were proposed that as a policy action on their part move on that's not what I'm proposing we do I'm just saying as a sort of thought experiment but so you I answer that oh because human psychology that's why you shouldn't assume the same of the eyes they're not gonna start with human psychology okay fair enough assume we start off with dogs right
good old-fashioned dogs and We bred them to be more intelligent, but also to be friendly Well as soon as they are past a certain level of intelligence I object to us like coming in and breeding them. They can no longer be owned They are now sufficiently intelligence to not beyond anymore, but let us leave aside all morals carry on In the thought experiment not in real life. You can't leave out the morals in real life. Do you have a sort of? Universal prior over their drives of these like super intelligent dogs that are bred to be friendly man So I think that weird shit starts to happen at the point where the dogs get smart enough that they are like What are these flaws in our thinking processes? How can we correct them? You know the over the CIFAR threshold of dogs, although maybe that's like CIFAR some strange baggage over the Korzybski threshold of dogs after Alfred Korzybski Yeah, so I think that You know there's this whole Domain where they're Stupider than you and sort of like being shaped by their genes and not shaping themselves very much and as long as that is true you can probably go on breeding them and Issues start to arise when the dogs are smarter than you When the dogs can manipulate you if they get to that point where the dogs can Strategically present a particular appearances to fool you Where the dogs are aware of the breeding process and possibly having opinions about where that should go in the long run Where the dogs are even if just by thinking and by adopting new rules of thought modifying themselves in that small way These are these are some of the points where like I expect the weird shit to start to happen and it won't and the weird shit Will not necessarily show up while you're just reading the dogs Does the weird shit look like? Dog gets smart enough dot dot dot human stuff existing If you keep on optimizing the dogs Which is not the correct course of action. I Think I mostly expect this to eventually blow up on you But blow up on you that bad It's hard. Well, it If I expect to blow up on you quite bad I'm trying to think about whether I expect super dogs to be sufficiently in a human frame of reference in virtue of them Also being mammals that they that a super dog would like create human ice cream Like they you bred them to have preferences about humans and they invent something that is like ice cream to those preferences Or does it just like go off someplace stranger? Hi, there could be a ice cream. Hmm. There could be a ice cream Ice cream that is things that is equivalent of ice cream for a eyes that is Essentially my prediction of what the solar system ends up shared with Anyways, the exact ice cream is like quite hard to predict just like it would be very hard to like look at well If you optimize something for inclusive genetic fitness, you'll get ice cream. That was a very hard call to make. Yeah, sorry I didn't mean to interrupt. Where were you going with your no, I Think yeah, I was just like Brambling in my attempts to make predictions about these super dogs You're like asking me to I mean I feel like you know In a you know in a world that had anything remotely like its priority straight this stuff is not me like Extemporizing on a blog post there. There are like 1,000 papers that were written by people who otherwise became philosophers writing about this stuff instead But you know your world this has not set its priorities that ways and I'm concerned that it will not set them that way in The future and I'm concerned that if it tries to set them that way it will end up with like Garbage because the good stuff is hard to verify But you know separate topic Yeah on that particular intuition about the dog thing Like I understand your intuition that we would end up in a place that is not very good for humans That just seems so hard to reason about that I Honestly would not be surprised if it ended up like fine for humans In fact the dogs when I wanted like good things for humans loved humans like we're smarter than dogs. We love them The sort of reciprocal relationship came about I don't know I feel like maybe I could do this given Thousands of years to breed the dogs in a total absence of ethics, but it would actually be easier with the dogs I think than with gradient descent Because I think it's well because the dogs are starting out with neural architecture very similar to human and Natural selection is just like a different idiom from gradient descent In particular in terms of like information bandwidth and But like like I'd be steering to to like breed the dogs into like very nice like genuinely very nice human and Like knowing the stuff that I know that We're your typical dog breeder might not know what they saw to be embarked on this project I would be like early on being like You know like sort of prompting them into the weird stuff that I Expected to get started later and trying to observe how they went during that that this is the alignment strategy We need a what are smart dogs to help us solve wait. There's no time Okay, so I Think we sort of articulated our intuitions on that one Here's another one. That's not something I came into the conversation with Like some of my intuition here is like I know how I would do this with dogs And I think you could like ask Open AI to describe their theory of how to do it with dogs And I would be like oh wow that's sure gonna get you sure sure it's gonna get you killed And that's kind of how I expected to play out in practice Actually, did you mind if I ask like it? But when you talk to the people who are in charge of these labs What do they say I do they just like not rock the arguments you think they talk to me There was a certain selfie that was taken by five minutes of conversation first time any of the people in that selfie had met each other and then did you like bring it up or I Asked him to change the name of his cup of his corporation to anything, but open AI uh-huh have you likes? Seeked an audience with the leaders of these labs to explain these arguments no Why not I? Did try to I have had a couple of conversations with like Demis Asavis? Who struck me as like much more of the sort of person who is possible to have a conversation with I? Guess it seems like it would be more dignity to explain Even if you think it's not going to be fruitful ultimately the people who are like most likely to be influential in this race I My basic model was that they wouldn't like me and that things could always be worse fair enough You know I mean like they sure they sure could have like I mean they they sure could have asked at any time But you know that would have been like quite out of character And the fact that it was quite out of character is like I might why I myself did not like Go trying to like barge into their lives and getting them mad at me But you think them getting mad at you would make things worse. It can oh it can always be worse I Agree that that you know like possibly at this point some of them are mad at me, but you know Yeah, I I have yet to turn down the leader of any major AI lab who has come to me asking for advice fair enough Okay, so so on the scene of like big-picture disagreements like why I'm still not on the greater than 50% doom It just seemed like From the conversation it didn't seem like you were willing or able to make predictions About the world short of doom that would help me distinguish And highlight your view about other views yes I mean the world heading to this is like a whole giant mess of complicated stuff which Predictions about which can be made in virtue of like Spending a whole bunch of time staring at the complicated stuff Until you understand that specific complicated stuff and making predictions about it's like for my for my first yeah for my perspective like The way you get to my point of view is not by having a grand theory that reveals how things will actually go it's like taking other people's overly narrow theories and poking at them until they come apart and you're left with in a Maximum entropy distribution over the right space which which looks like yep, that's that's you're gonna randomize the solar system but but to me it seems like the nature of intelligence and what it entails is even more complicated than the sort of geopolitical or Economic things that would be required to predict where the world's gonna. I think you're just wrong I think that the like theory of I think the theory of intelligence is just like flatly not that complicated Maybe maybe that's just like the voice of like person with talent in one area but not the other but that's sure how it feels to me this would be even more convincing to me if We had some idea of what the pseudocode or circuit or intelligence look like and then you could say like, oh This is what the pseudocode implies. We don't even have that. I mean If you permit a hypercomputer, he just does a IXI What is a IXI? You have the Solomon off prior over your environment, yeah update it on the evidence and Then max sensory reward Okay, so it's actually it's not actually trivial like actually this thing will like Exhibit weird discontinuities around its Cartesian boundary with the universe. It's not actually trivial but like everything that people imagine as the like hard problems of intelligence are contained in the equation if you have a hypercomputer Yeah, fair enough. I but I mean in this sort of sense of you know Programming it in there like a normal Like I give you a goo a fad or I give you a really big computer Write the pseudocode or something like that for I mean if you give me a hypercomputer Yeah, so what you're saying? What you're saying here is that like the theory of intelligence is really simple in an unbounded sense But as soon as you like, yeah, what about this like depends the difference between unbounded and unbounded intelligence? So, how about this you asked me? Do you understand how fusion works? If not, how can you predict?
the, let's say we're talking like the 1800s, how can you predict how powerful a fusion bomb would be? And I say, well, listen, if you put in a pressure, I'll just show you the sun. And the sun is sort of the archetypal example of a fusion is. And you say, no, no, no, I'm asking like, what would a fusion bomb look like? You see what I mean? Not necessarily. Like what is it that you think somebody ought to be able to predict about the road ahead? So first of all, like if you, um, one of the things, if you know the nature of intelligence is just like, how will this sort of progress and intelligence look like what, you know, how our ability is going to scale, if at all, how fast, and it looks like a bunch of details that don't easily follow from the general theory of, you know, like simplicity, prior Bayesian update argmax. This is again, so then the only thing that follows is the wildest conclusion, which is, you know what I mean? Like there, there's no like simpler conclusions to follow, like the Eddington looking, uh, and confirming special relativity is just like the wildest possible conclusion is the one that follows. Yeah. Like the, the convergence is a whole lot easier to predict than the pathway there. I'm sorry, but, and I sure wish it was where otherwise, but. And also remember the basic paradigm from my perspective, I'm not making any brilliant startling predictions. I'm poking at other people's incorrectly narrow theories until they fall apart into the maximum entropy state of doom. There's like thousands of possible theories, most of which have not come about yet. I don't see it as strong evidence that because you haven't been able to identify a good one yet that, oh, somebody, I mean, if somebody in the profoundly unlikely event that somebody came up with some incredibly clever grand theory that explained all the properties GPT five ought to have, which is like just flatly not going to happen. It's just like that kind of info that's available. You know, my hat is my hat would be off to them if they wrote down their predictions in advance. And if they were then able to like grind that theory to produce predictions about alignment, which seems like even more improbable because like, what do those two things have to do with each other? Exactly. But like, still, you know, like, I mean, mostly it'd be like, well, uh, it looks like our generation has its new genius. How about if we all shut up for a while and listen what they have to listen to what they have to say? How about this? Let's say, um, uh, somebody comes to you and they say, I have the best and newest theory of economics. Everything before is, uh, wrong. But they say in the year, one does not say everything before is wrong. One says one, one predicts the following new phenomena and on rare occasions say that old phenomena were organized incorrectly. Fair enough. But they say all phenomena are organized incorrectly because of the, uh, and then here's how to term this person, Scott Sumner, for the sake of simplicity, they say in the next 10 years, there's going to be a depression that is so bad that it's going to destroy the entire economic system. I'm not talking just about something that is a hurdle. It is like literally civilization will collapse because it's an economic disaster. And then you ask them, okay, give me some predictions before this great catastrophe happens about like what this theory implies. And then they say, like, listen, there's many different branching paths, but they all converge at civilization collapsing because of some great economic crisis. I'm like, I don't know, man, like I would like to see some predictions before that. Yeah, I, it sure, yeah, wouldn't it be nice, wouldn't it be nice. So we're left with your 50% probability that we win the lottery and 50% probability that we don't, because nobody has like a theory of lottery tickets that has been able to predict you what numbers get drawn next. I don't agree with the analogy that it's, it's all about the probabilities. It's all about the space over which you're, you're uncertain. We are all quite uncertain about where the future leads, but over which space and it, and there isn't, there, there isn't a Royal road. There isn't a, there isn't a simple like, ah, I found just the right thing to be ignorant about. It's so easy. The chance of a good outcome is 33% because they're like one possible good outcome and two possible bad outcomes. The stuff that you do when you're uncertain is, is, is like, like, like the, the, the, the thing you're trying to fall back to in the absence of anything that predicts exactly which properties GPT-5 will have is, is your sense that, you know, a pretty bad outcomes kind of, kind of weird, right? It's probably a small sliver of the space. It seems kind of weird to you, but that's just like imposing your natural English language prior, like your natural humanese prior on the space of possibilities and being like, I'll distribute it. My max entropy stuff over that. Yeah. Can you explain that again? Okay. What is the person doing wrong who says 50, 50, either I'll win the lottery or I won't. They have the wrong distribution to begin with over possible outcomes. Okay. What is the person doing wrong who says 50, 50, either we'll get a good outcome or a bad outcome from AI? They, they don't have a set any good theory to begin with about what the space of outcomes looks like. Is that what, is that your answer? Is that your model of my answer? My answer. Okay. But all the, but all the like things you could say about a space of outcomes are an elaborate theory and you haven't predicted GPT-4's exact properties in advance. Shouldn't that just leave us with like good outcome or bad outcome, 50, 50? People did have theories about what GPT-4's, like if you look at the scaling laws, right? Like the, you can, you can like put it, it probably falls right on the sort of curves that were drawn in like 2020 or something. The loss, the loss on text predictions. Sure. That followed a curve, but which abilities would that correspond to? I don't, not familiar with anyone who called that in advance. What good does it know to know the loss? You could like, you could have taken those exact loss numbers back in time, 10 years and been like, what does that, what kind of, what kind of commercial utility does this correspond to? And they would have given you utterly blank looks. And I don't actually know of anybody who has a theory that gives something other than a blank look for that. All we have are the, all we have are the observations. Everyone's in that boat. All we can do are fit the observations. I mean, so like also like there's just like me start starting to work on this problem in 2001 cause it was like super predictable going to turn into an emergency later. And the point of fact, like nobody else ran out and like immediately tried to start getting work done on the problems. And I, I would claim that as accessible, successful prediction of the grand lofty theory you had. Is did you see deep learning coming as the main paradigm and is that relevant as part of the picture of intelligence? I mean, I would have, I would have been like much, much, much more worried in 2001 if I'd seen deep learning coming. No, not in 2001. I just mean before it became like the, obviously the main paradigm of AI. No, it's like the details of biology. It's like asking people to like predict what the organs look like in advance via the principle of natural selection. And you like, it's, it's pretty hard to call an advance. You like afterwards you can look at it and be like, yep, this like sure does look like it should look if this thing is being optimized to reproduce. But the space of solution of things that biology can throw at you is just too large. Like there's, it's very rare that you have a case where there's only one solution that lets the thing reproduce that you can predict by the theory that it will success that it will have successfully reproduced in the past. And mostly it's just this enormous list of details and they do all fit together in retrospect. The theory, it is a sad truth that contrary to what you may have learned in science class as a kid, there are genuinely super important theories where you can totally actually validly see that they explain the thing in retrospect and yet you can't do the thing in advance. Not always, not everywhere, not for natural selection. There are advanced predictions you can get about that given the amount of stuff we've already seen. You can go to a new animal in a new niche and be like, oh, it's going to have this properties given the stuff we've already seen in the niche. But you could also make that by like blind gender. There's advanced predictions that they're a lot harder to come by. Which is why natural selection was a controversial theory in the first place. It wasn't like gravity. People were being like, gravity had all these awesome predictions. Newton's theory of gravity had all these awesome predictions. We got all these extra planets that people didn't realize ought to be there. We figured out Neptune was there before we found it by telescope. Where is this for Darwinian selection? People actually did ask at the time. And the answer is, it's harder. And sometimes it's like that in science. The difference is the theory of Darwinian selection seems much more well developed. There were precursors of Darwinian selection that, I don't know, who was that Roman poet Lucretius, right? He had some poem where there was some precursor of Darwinian selection. And I feel like that is probably our level of maturity when it comes to intelligence. Whereas we don't have like a theory of intelligence, we might have some hints about what it might look like. Oh, we've got our hints. From hints, it seems harder to extrapolate very strong conclusions. They're not very strong conclusions is the message I'm trying to say here. I'm pointing to your being like, maybe we might survive. And like, whoa, that's a pretty strong conclusion you've got there. Let's weaken it. That's the basic paradigm I'm operating under here. Like you're in a space that's narrower than you realize when you're like, well, you know, if I'm kind of unsure, maybe there's some hope. Yeah, I think that's a good place.
to close the discussion on AI unless... Well, I do kind of want to like mention one last thing, which is that again, like in historical terms, if you look out the actual battle that was being fought on the block, it was me going like, like, I expect there to be AI systems that do a whole bunch of different stuff. And Robin Henson being like, I expect there to be a whole bunch of different AI systems that do a whole different bunch of stuff. But that was one particular debate with one particular person. And yeah, but like your planet having made the strange reason given its own widespread theories to not invest massive resources and having a much smarter version, well, not smarter, a much larger version of this conversation, as it thought deemed apparently deemed prudent given the implicit model that it had of the world. Such that like I was investing a bunch of resources in this and kind of dragging Robin Henson along with me, though he did have his own separate line of investigation into topics like these. Being there as I was, my model having led me to this important place where the rest of the world apparently thought it was fine to let it go hang, such debate as there actually was at the time was like, are we really going to see these single AI systems that do all this different stuff? Is this like whole general intelligence notion kind of like meaningful at all? And I staked out the bold position for it actually was bold. And people did not all say like, oh, Robin Henson, you fool, why do you have this exotic position? They were going like, ah, like, behold, these two luminaries debating or behold these two idiots debating, and like not massively coming down on one side or the other. So, you know, like, in historical terms, I dislike, you know, making it out like I was right about anything when I feel I've been wrong about so much and yet I was right about anything. And, you know, relative to what the rest of the planet deemed important stuff to spend its time on, given their implicit model of what was going to, what was, how it's going to play out, what you can do with minds, where AI goes. I think I think I did okay. Warren Branwen did better. Shane Legg arguably did better. Warren always does better when it comes to forecasting. I mean, obviously, like, if you get the better of a debate that like counts for something, but a debate with one particular person, considering your entire planet's decision to invest like $10 into this entire field of study, apparently one big debate is all you get. And that's the evidence you got to update on now. So somebody like Ilya Sutskever, you know, when it comes to the actual paradigm of deep learning, like, was able to anticipate, like, from ImageNet to scaling up LLMs or whatever. There's people with track records here who are like, who disagree about Doom or something. So in some sense, it's probably more people who have been. If Ilya challenged me to a debate, I wouldn't turn him down. I admit that I did specialize in Doom rather than LLMs. Okay, fair enough. I unless you have other sorts of comments on AI, I'm, I'm happy with. Yeah. And again, I'm not being like, due to my miraculously precise and detailed theory, I'm able to make the surprising and narrow prediction of Doom I am like, I am being like, like, I think I did a fairly good job of shaping my ignorance to lead me to not be too stupid, despite my ignorance over time as it played out. And you know, there's a prediction, even knowing that little that can be made. Okay, so this feels like a good place to pause the AI conversation. And there's many other things to ask you about, given your decades of writing and millions of words. So I think what some people might not know is the millions and millions and millions of words of science fiction and fan fiction that you've written. I want to understand when in your view, is it better to explain something through fiction than nonfiction? When you're trying to convey experience rather than knowledge, or when it's just much easier to write fiction, and you can like produce 100,000 words of fiction with the same effort, it would take you to produce 10,000 words of nonfiction. Those are both pretty good reasons. On the second point, it seems like when you're writing this fiction, not only are you in your case, covering the same heady topics that you include in your nonfiction, but there's also the added complication of plot and characters. It's surprising to me that that's easier than just verbalizing the sort of the topics themselves. Well, partially because it's more fun is an actual factor, ain't gonna lie. And sometimes it's something like a bunch of what you get in the fiction is just like the lecture that the character would deliver in that situation, the thoughts the character would have in that situation. There's like only like one piece of fiction of mine where there's literally a character giving lectures because he arrived on another planet and now has to lecture about science to them. That one is Project Lawful. You know about Project Lawful? I know about it. I have not read it yet. Yeah. Okay. So most of my fiction is not about somebody arriving in another planet who has to deliver lectures there. I was being like a bit deliberately like, yeah, I'm gonna just do it with like Project Lawful. I'm gonna just do it. They say nobody should ever do it and I don't care. I'm doing it every way. I'm going to have my character actually launch into lectures. The lectures aren't really the parts I'm proud about. It's like where you have like the like life or death, Death Note style battle of wits between like that that is like centering around a series of Bayesian updates and like making that actually work because, you know, is where I'm like, yeah, I think I actually pulled that off and I don't think I'm not sure a single other writer on the face of this planet could have made that work as a plot device. But with that said, like the nonfiction is like I'm explained this thing, I'm explained the prerequisites, I'm explained the prerequisites to the prerequisites. And then in fiction, it's more just like, well, this character happens to think of this thing and the character happens to think of that thing, but you got to actually see the character using it. So it's less organized. It's less organized as knowledge. And that's why it's easier to write. Yeah. I mean, one of my favorite pieces of fiction of fiction to explain something is the Dark Lord's answer. And I honestly can't say anything about it without spoiling it. But I just want to say, like, honestly, it's like such a great explanation of the thing it is explaining. I don't know what else I can say about it without spoiling it. Anyways. Yeah. I'm laughing because I think like relatively few have Dark Lord's answer as their as like among their top favorite works of mine. It is one of my less widely favored works of mine. Actually, what is my favorite sort of this is a medium, by the way, I don't think is used enough given how effective it was in inadequate equilibria. You have different characters just explaining concepts to each other, some of whom are purposefully wrong as examples. And that is such a useful pedagogical tool. And I don't know, honestly, like at least half a blog post should just be written that way. It is so much easier to understand that way. Yeah. And it's easier to write. I should probably do it more often. And like you should give me a stern look and be like, Eliezer, write that more often. Done. Eliezer. Please. I think 13 or 14 years ago, you wrote an essay called Rationality Systematizes Winning. Would you have expected then that 14 years down the line, the most successful people in the world or some of the most successful people in the world would have been rationalist? Only if the whole rationalist business had worked like closer to the upper 10% of my expectations than it actually got into. The title of the essay was not Rationalists are Systematized Winning. It wasn't even a rationality community back then. Rationality is not a creed. It is not a banner. It is not a way of life. It is not a personal choice. It is not a social group. It's not really human. It's a structure of a cognitive process. And you can try to get a little bit more of it into you. And if you want to do that and you fail, then having wanted to do it doesn't make any difference except insofar as you succeeded. Hanging out with other people who share that creed, going to their parties, it only ever matters insofar as you get a bit more of that structure into you. And this is apparently hard. This seems like a no true Scotsman kind of point because... And yet there are no true Bajans upon this planet. Do you really think that had people tried much harder to adopt the sort of Bajan principles that you laid out, they would have... Many of the successful people, some of the successful people in the world would have been Rationalists? What good does trying do you except insofar as you are trying at something which when you...
Try it, it succeeds. Is that an answer to the question? Rationality is systematized winning. It's not rationality, the life philosophy. It's not like trying real hard at this thing, this thing, and that thing. It was like the mathematical sense. Okay. So then the question becomes, does adopting the philosophy of Bayesianism consciously actually lead to you having more concrete wins? Well, I think it did for me, though only in scattered bits and pieces of slightly greater sanity than I would have had without explicitly recognizing and aspiring to that principle. The principle of not updating in a predictable direction, the principle of jumping ahead to where you will predictably be later. I look back and the story of my life, as I would tell it, is a story of my jumping ahead to what people would predictably believe later after reality finally hit them over the head with it. This to me is the entire story of people running around now in a state of frantic emergency over something that was utterly predictably going to be an emergency later as of 20 years ago. I could have been trying stuff earlier, but it left it to me and a handful of other people. It turns out that that was not a very wise decision on humanity's part because we didn't actually solve it all. I don't think that I could have tried even harder or contemplated probability theory even harder and done very much better than that. I contemplated probability theory about as hard as the mileage I could visibly obviously get from it. I'm sure there's more. There's obviously more, but I don't know if it would have let me save the world. I guess my question is, is contemplating probability theory at all in the first place something that tends to lead to more victory? I mean, I imagine who is the richest person in the world? How often does Elon Musk think in terms of probabilities when he's deciding what to do? And here's somebody who is very successful. So I guess the bigger question is, in some sense, when you say rationality system, it's like a tautology if the definition of rationality is whatever helps you win. If it's the specific principles laid out in the sequences, then the question is, do the most successful people in the world practice them? I think you are trying to read something into this that is not meant to be there. The notion of rationality systematized winning is meant to stand in contrast to a long philosophical tradition of notions of rationality that are not meant to be about the mathematical structure, not meant to be, or about strangely wrong mathematical structures where you can clearly see how these mathematical structures will make predictable mistakes. It was meant to be saying something simple. There's an episode of Star Trek wherein Kirk makes a 3D chess move against Spock. And Spock loses. And Spock complains that Kirk's move was irrational. Rational towards the goal, yeah. The literal winning move is irrational, or possibly illogical, Spock might have said. I might be misremembering this. The thing I was saying is not merely, that's wrong. That's a fundamental misunderstanding of what rationality is. There's more depth to it than that, but that is where it starts. There are so many people on the internet in those days, possibly still, who are like, well, you know, if you're rational, you're going to lose because other people aren't always rational. This is not just like a wild misunderstanding, but there's like the contemporarily accepted decision theory in academia as we speak at this very moment, causal decision theory, classical causal decision theory, basically has this property where you can be irrational and the rational person you're playing against is just like, oh, I guess I lose then, here, have most of the money. I have no choice but to, and ultimatum games specifically, if you look up logical decision theory on Arbital, you'll find a different analysis of the ultimatum game where the rational players do not predictably lose the same way as I would define rationality. And if you take this sort of like deep mathematical thesis that also runs through all the little moments of everyday life when you may be tempted to think like, well, if I do the reasonable thing, won't I lose that you're making the same mistake as the Star Trek script writer who had Spock complain that Kirk had won the chess game irrationally, that every time you're tempted to think like, well, like, here's the reasonable answer and here's the correct answer, you have made a mistake about what is reasonable. And if you then try to screw that around as like rationalists should win, rationalists should have all the social status, whoever's the top dog in the present social hierarchy or the planetary wealth distribution must have the most of this wealth, must have the most of this math inside them. There are no other factors. But how much of a fan you are of this math, that's trying to take the deep structure that can run all through your life in every moment where you're like, oh, wait, like maybe the move that would have gotten the better result was actually the kind of move I should repeat more in the future. Like to take that thing and like turn it into like, like, social dick measuring contest time. Rationalists don't have the biggest dicks. Okay. Final question. This has been, I don't know how many hours, I really appreciate you giving me your time. Final question. I know that in a previous episode, you were not able to give specific advice of what somebody young who is motivated to work on these problems should do. Do you have advice about how one would even approach coming up with an answer to that themselves? There's people running programs to try to, who think we have more time, who think we have better chances, and they're running programs to try to nudge people into doing useful work in this area. And I'm not sure they're working. And there's such a strange road to walk, and not a short one. And I tried to help people along the way, and I don't think they got far enough. Like some of them got some distance, but they didn't turn into alignment specialists doing great work. And it's the problem of the broken verifier. Somebody had a bunch of talent in physics, and they were like, well, I want to work in this field. I might be like, well, there's interpretability. And you can tell whether you've made a discovery in interpretability or not, which sets it apart from a bunch of this other stuff. And I don't think that that saves us. And okay, so how do you do the kind of work that saves us? And I don't know how to convey the, and the key thing is the ability to tell the difference between good and bad work. And maybe I will write some more blog posts on it. I don't really expect the blog posts to work. And the critical thing is the verifier. How can you tell whether you're talking sense or not? Whether you're, there's all kinds of specific heuristics I can give. I can be like, I can say to somebody like, well, it's like your entire alignment proposal is this like elaborate mechanism. You have to explain the whole mechanism. And you can't be like, here's the core problem. Here's the key insight that I think addresses this problem. If you can't extract that out, if your whole solution is just a giant mechanism, this is not the way. It's kind of like how people invent perpetual motion machines by making the motion, perpetual motion machines more and more complicated until they can no longer keep track of how it fails. And if you actually had somehow a perpetual motion machine, it would not just be like giant machine. There would be like a thing you had realized that made it possible to do the impossible. For example, you're just not going to have a perpetual motion machine. So like there's thoughts like that. I could say like, go study evolutionary biology because evolutionary biology went through a phase of optimism and people naming all the wonderful things they thought that evolutionary biology would cough out, like all the wonderful properties that they thought natural selection would imbue into organisms. And the Williams revolution, as it's sometimes called, is when George Williams wrote Adaptation and Natural Selection, a very influential book, saying like, that is not what this optimization criterion gives you. You do not get the pretty stuff. You do not get the aesthetically lovely stuff. Here's what you get instead. And by like, living through that revolution vicariously, well, I thereby picked up a
a bit of like thing that to me obviously generalizes about how not to expect nice things from an alien optimization process. But maybe somebody else can read through that and like not generalize, not generalize in the correct direction. So then how do I advise them to generalize in the correct direction? How do I advise them to learn the thing that I learned? I can just give them the generalization, but that's not the same as having the thing inside them that generalizes correctly without anybody standing over their shoulder and forcing them to get the right answer. I could point out and have in my fiction that the entire schooling process of like, here is this legible question that you're supposed to have already been taught how to solve. Give me the answer using the solution method you are taught that this does not train you to tackle new basic problems. But even if you tell people that like, okay, how do they retrain? We don't have a systematic training method for producing real science in that sense. We have like half of the, what was it? A quarter of the Nobel laureates being the students or grand students of other Nobel laureates because we never figured out how to teach science. We have an apprentice system. We have people who like pick out people who like they think can be scientists and they like hang around them in person and something that we've never written down in a textbook passes down and that's where the revolutionaries come from. And there are whole countries trying to invest in having scientists and they turn out these people who write papers and none of it goes anywhere because the part that was legible to the bureaucracy is have you written the paper? Can you pass the test? And this is not science. And I could go on for this for a while but the thing that you asked me is like, how do you pass down this thing that your society never did figure out how to teach? And the whole reason why Harry Potter and the methods of rationality is popular is because people read it and picked up the rhythm seen in a character's thoughts of a thing that was not in their schooling system, that was not written down, that you would ordinarily pick up by being around other people and I managed to put a little bit of it into a fictional character and people picked up a fragment of it by being near a fictional character but you know, like not in really vast quantities, not vast quantities of people and I didn't manage to put vast quantities of shards in there. I'm not sure, there is not a like long list of Nobel laureates who've read HPMOR although there wouldn't be because the like delay times on granting the prizes are too long. It's, yeah, like you asked me what do I say and my answer is like, well, that's a whole big gigantic problem I spent however many years trying to tackle and I ain't gonna solve the problem with a sentence in this podcast. Fair enough. Eliezer, thank you so much for giving me, I don't know how many hours of your time. This was really fun. Hey everybody, I hope you enjoyed that episode. As always, the most helpful thing you can do is to share the podcast. Send it to people you think might enjoy it, put it in Twitter, your group chats, et cetera. Just splits the world. Appreciate your listening. I'll see you next time. Cheers.