Young and Profiting with Hala Taha - Nick Bostrom: The AI Revolution and What It Means for Entrepreneurs | Artificial Intelligence | YAPClassic

Starting point is 00:00:00 Today's episode is sponsored in part by Shopify Quo, Indeed, and Experian. Shopify is the global commerce platform that helps you grow your business. Start your $1 per month trial at Shopify.com slash profiting. Quo is an AI-powered phone system that brings your calls, texts, and contacts together in one place. Try Quo for free, plus get 20% off your first six months when you go to Quo.com slash profiting. Indeed helps you attract interview and hire talent all in one place. get a $75-sponsored job credit to boost your job's visibility at indie.com slash podcast. Experian is a financial app that negotiates your bills and cancels unwanted subscriptions.

Starting point is 00:00:41 Get started with the Experian app today. As always, you can find all of our incredible deals in the show notes or at young and profiting.com slash deals. If this is a simulation, then presumably we can infer a few things that the people building it would have to be very technologically advanced. Nick Beaustrum isn't just a philosopher. He's a global thought leader on the future of artificial intelligence. He's the author of Super Intelligence,

Starting point is 00:01:07 the groundbreaking book that brought the risks of advanced AI into mainstream conversation. People have, for thousands of years, tried to create imaginary worlds that people can experience. It would be the theater, right, or literature. Maybe for these post-humans, they might be interested in knowing if they ever ran into alien civilizations, what those would be like. How do you think about AI in terms of the significance in humanity?

Starting point is 00:01:33 Reviewing the rapid recent advances that we've seen in the field of artificial intelligence. It really looks like we kind of possibly figured out a large component of the secret sauce. So how do you think entrepreneurship will change in this world? You mentioned that there might be still some jobs. The kinds of jobs that might remain, I think, are... If it's true that we're living in a simulation, what do you feel like are the moral, moral implications of what it means for our lives. That's difficult, I think.

Starting point is 00:02:07 Hey, young improfitors, AI is everywhere right now. And if you're not paying attention, you're already falling behind. So in the spirit of this week's AI theme, today's ZAP Classic tackles the bigger question. What happens when AI becomes more intelligent than humans? I'm bringing back my conversation with Oxford philosopher Nick Bowstrom, one of the world's most influential thinkers on artificial intelligence. We dive into superintelligence, the risks, and the opportunities of AI, and how this technology could reshape human civilization for generations to come. Fair warning, you will not be thinking about AI the same after this. Now here's my

Starting point is 00:02:44 conversation with Nick Bowstrom. Nick, welcome to Young Improfiting Podcast. Thank you so much for having me. I'm so excited for this conversation. I love conversations about the future, about AI, and you've spent your career focused on really deep, long-range questions, the deepest questions that we could really ask about humanity. And so I'm wondering, what really first drew you to thinking about humanity, thousands, and even billions of years into the future? I think it's sad if we have this allotted time here on the planet in this magical cosmos, and we never really take the time to look around.

Starting point is 00:03:27 or try to figure out what is going on here. You know, I feel sometimes we are a little bit like ants running around, being very busy pulling our needle to the ant hill. But don't really stop to reflect what is this ant hill that we're building. What is it for? You know, what else is going on in this forest around us? It's so true. You're just, like, focused on working and hustling

Starting point is 00:03:55 and not really paying attention to what we're even living in. And I know that one of the things that made you famous is that you put out a paper in 2003, and you talked about how we're living in a simulation, or you had the hypothesis that we're living in a simulation. And it's actually what first made you famous is putting out this paper. So talk to us about, you know, in 2025, what are the odds that you think that we're currently living in a simulation right now? I tend to punt on the probability question there.

Starting point is 00:04:29 I often get asked, but I've kind of refrained from putting an exact number on it. I do think it's, I take it as a very serious possibility, though. The simulation argument itself that you're referring to the paper that was published in 2002 only demonstrates that one of three possibilities obtains, one. one of which is the simulation hypothesis, but the simulation argument itself doesn't tell us which one of those three. So you need to sort of bring additional considerations to bear. But if you're thinking ahead in this time of rapid advances in AI

Starting point is 00:05:09 where all of this might be going, if you think eventually we'll have these superintellidences that develop all kinds of super advanced technologies, maybe colonized space, transform planets, to giant computers and amongst the things they could do with that kind of technology would be to run simulations, detailed simulations of environments like ours, and including with brains in those simulations simulated at a very high level of granularity. And so what that means is that if this happens, there could be many, many more people like us with our kinds of experiences being simulated

Starting point is 00:05:48 than being sort of implemented in the original meat substrate. And if most people with our kinds of experiences are simulated, then we should think we are probably amongst the simulated ones rather than the rare, exceptional, original ones, given that from the inside, you wouldn't be able to tell the difference. Yeah, but I really want to know, like, do you think we're living in a simulation? Well, as I said, I take the hypothesis seriously. Yeah, so you have one of three where you say, like, we could become extinct before there's. post-humans, right, then you say we might be living in a simulation. Talk to us about the three hypotheses that you have. Yeah, so if you break this down, right, so if we do end up with the future where this

Starting point is 00:06:34 mature civilization runs all these simulations of variations of people like their historical predecessors, then there would be many more simulated people with our experiences than non-simulated ones. Conditional on that, I think we should think we are almost certainly amongst the simulated ones. So then if you break this down, what are the alternatives to that? Well, one is that we won't end up with this future, and that could be because we go extinct before reaching technological maturity. So that's one of the alternatives. But not just that we go extinct, but it would have to be pretty universal amongst all other advanced civilizations throughout the universe, that they almost all would have to go extinct before reaching the level of

Starting point is 00:07:20 technological capability that would allow them to run these types of ancestor simulations. So that's possibility one, a strong filter that just every civilization that reaches our current stage of technological maturity just fails to go all the way there. Then the second is that, well, maybe they do become technological immature, but they decide not to use their planetary supercomputers for this purpose. They have other things to do. Maybe they all refrain from using even a small portion of their computational resources to run these simulations. So that's the second alternative, a strong convergence, they all lose interest

Starting point is 00:07:53 in running computer simulations. But if both of those fail, then we end up with a third possibility that we are almost certainly currently living in a computer simulation, created by some advanced civilization. Yeah. And the advanced civilization, you say they're post-human, right? Post-humanity. Can you talk to us about how you envision this post-humanity? What are they like, what are their capabilities? Well, if this is a simulation, then presumably we can infer a few things that the people building it would have to be very technologically advanced, because right now we can't create computer simulations with conscious human beings in them, right?

Starting point is 00:08:34 It's like they need to build very powerful computers, they need to know how to program them, etc. And then you can figure if they have the technology to do that, they probably also have technology to do a bunch of other things. including enhancing their own intelligence. And so I imagine these would be super-intelligenses that would have reached a state close to technological perfection. And for the reason, they have some interest in doing this stuff.

Starting point is 00:09:03 But beyond that, it's hard to say very much specifically about what they would be like. Now that AI is at the forefront, do you believe that maybe these post-humans might be part human, part AI, or all AI? I mean, at that point, the distinction might blur, which also might be the case for us in the future, if things go well and we are allowed to continue to develop. Well, A, we will develop, I think, artificial superintelligence.

Starting point is 00:09:36 But amongst the things that technology could be used to do would be providing paths for us current biological humans to gradually upgrade our abilities. This could take the form of sort of biological enhancements of various kinds, but it could also ultimately take the form of uploading into computers, so you could imagine detailed scans of human brains that would then allow our memories and personalities and consciousness to continue to exist,

Starting point is 00:10:05 but then in digital substrate. And from there on, you could imagine further develop, you could add neurons, you could increase the processing speed, you could gradually sort of become some form of, you know, radically post-human super being that might be hard to differentiate from a purely synthetic AI. So interesting. So your theory of, if we're in a simulation,

Starting point is 00:10:29 there's post-humans who are really technologically advanced, and they're creating our world, which you call an ancestral, an ancestor civilization, correct? Why would they do that? like what would be the reason of them creating a civilization like ours? Yeah, so we can only speculate. I mean, we don't know much about post-human psychology or their motives, but I mean, there are several potential reasons, motivations.

Starting point is 00:11:02 You could ask why it is that we humans, with our currently more limited technology, create computer simulations, and we do it for a variety of purposes. People have, for thousands of years, try to create imaginary worlds that people can experience. Would be to theater, right, or literature. And more recently, through virtual reality and computer games, this can be for entertainment or for cultural purposes. You also have scientists from creating computer simulations to study various systems that might be hard to reach in nature. but you can sort of create a little computer simulation of them, and then you study how the simulation behaves.

Starting point is 00:11:48 So there could be entertainment reasons. There could be scientific reasons. Maybe for these posthumans, they might be interested in knowing if they ever ran into alien civilizations, what those would be like. And maybe one way to study that is to simulate many different originations of higher technological civilizations, like starting from something like current human civilization before and sort of running the tape forward and seeing what the distribution is

Starting point is 00:12:18 of different kinds of superintelligences you would get from that. And you could also imagine other, you could imagine historical tourism. If they can't literally travel back in time, but what the second best might be is to create sort of replicas of historical environments that future people could sort of experience, almost as if they were going back in time,

Starting point is 00:12:40 but living in a sort of temporarily exploring a simulated reality. You could imagine other sort of moral or religious reasons as well, of different kinds. If it's true that we're living in a simulation, what do you feel like are the moral implications of what it means for our lives? That's difficult. I think to a first initial approximation, I would say, if you are in a simulation, do the same things you would, if you knew you were not in a simulation.

Starting point is 00:13:16 Because the best guide to what would happen next in the simulation and how your actions would impact things might still be the normal methods we use. Like, you look at patterns and extrapolate those. And so whether we're simulated or not, unless you have some direct insight into what the, simulator's motives are or like the precise way in which this simulation was set up, you just have to sort of look at what kind of simulation this appears to be and what seems to, you know, if you do A,

Starting point is 00:13:47 you know, B follows. If you want to get into your car, you have to take out your car keys, if you want to do this. So I think that would be to a first cut, the answer. But then to the extent that you think you have some maybe probabilistic

Starting point is 00:14:03 guesses about how these things are configured that might give you sort of on the margin more reason to emphasize some hypothesis that otherwise would be less plausible. So for example, if we are not in a simulation and you have a secular materialistic outlook or life, then when we die, we die, and that's it, right?

Starting point is 00:14:31 where in a simulation you could potentially be moving into a different simulation or uplifted to the level of the simulators. This would at least be on the table as possibilities. Similarly, if we are in basement physical reality, as far as we know, current physical theories, say the world can't just suddenly pop out of existence.

Starting point is 00:14:54 Like there are conservation of energy, conservation of momentum, and other physical loss that prevent that from happening. If, however, our world simulator, then in theory, if the simulators flick the power off, our world would pop like a bubble disappearing into nothingness. Broadly speaking, I think that would be a wider range of possibilities on the table if we are simulated than if we are not. So it might mean approaching our existence with less confidence that we have it basically figured out and thinking there might

Starting point is 00:15:29 be more things on heaven and on earth than we sort of normally assume in our common sense philosophy and then maybe some sort of attitude of humility would be appropriate in that context. It's so interesting. Is there any sort of like clues or pieces of proof that prove were in a simulation? Like for example, like the dinosaurs and how they just like went to. extinct and then, you know, it's kind of like a new world after that. Do you feel like there's any clues to you that were in a simulation? I'm rather skeptical of that. I get a lot of random

Starting point is 00:16:10 people emailing saying they've discovered some glitch in the matrix or something. You know, somebody was like looking at their bathroom mirror and thought they saw pixels or like other. But I think the thing, though, is that whether we are in a simulation or not, you would still expect some people to report those kinds of observations for all the normal types of psychological reasons. So, like, some people might hallucinate something. Some might be misremembering something or misinterpreting something or making something up or, like, these things you would expect to take place anyway. So I think whether we are in a simulation or not, the best, most likely explanation for those reports are these ordinary psychological phenomena rather than that there is

Starting point is 00:16:57 actually some defect in the simulation that they have been able to detect. I think to create a simulation like this in the first place would be very hard, and simulators advanced enough to do that would probably also have the ability to patch things up so that the creatures inside the simulation couldn't notice. And if they did notice, they could sort of edit that out or rerun it from an earlier save point or edit the memory or do other things like that. So I don't think that. I think that there are indirect observations

Starting point is 00:17:30 that might slightly adjust the probability. So if you recall the original simulation argument with these three possibilities, right? The simulation argument shows at least one of them is true, but doesn't tell us which one. But what that means is that to the extent we get evidence against the first two possibilities, it would tend to shift probability mass

Starting point is 00:17:51 over to the remaining third. And conversely. So if you think we can get evidence for or against, say, the first one, which is that almost all civilizations at our current stage of technological development go extinct before reaching technological maturity, then that would tend to indirectly affect the probability of the simulation hypothesis. So in particular, the closer we get ourselves to being able to wield this technology, the less likely it is that almost all civilizations at our stage

Starting point is 00:18:20 failed to go the remaining few steps. and so that will tend to, you know, move probability over to the remaining possibilities. So basically what you're saying is if we can, in ourselves, like as technology advances, start to create simulations, that that means that we're probably in a simulation ourselves. Yeah, and in particular, if we, yeah, get both the ability and we still at that point retain the interest in doing that, that would sort of really leave only the third possibility. But short of getting all the way there, if we get sort of half the way there, like it still removes some of the places where this great filter could be.

Starting point is 00:18:58 Like the closer we get, the less chance there is that the remaining few steps will be impossible to take. Do you feel like AI is going to help us get closer to creating this simulation? And now that AI is something that we use all the time, do you think it will help discover that we're in a simulation? Well, I mean, for the same reasons we might be able to figure these. things out. If indeed the argument is sound, then I guess other powerful minds would also reach that same conclusion. And they might also reach the conclusion that they are in a simulation. For them, it might, especially for sort of early stage AIs, be a very salient possibility, because already,

Starting point is 00:19:42 right now, when training our AIs, we often put them in simulations. That's like a big part of the training environment, like virtual game environments or something. And so for them, it seems like a very obvious hypothesis to consider once they develop some degree of situational awareness. And you already see glimmers of that with current AI systems. And in fact, I think this might be an important shaper of AI behavior as they become sophisticated about reasoning strategically about the context. They would certainly be shaped by considerations

Starting point is 00:20:14 that they might well be living currently in a simulation environment rather than in deployment. Young improfitors, raise your hand. And if you've ever Googled one tiny symptom and then five minutes later, you're convinced that you're dying. I know I have been there done that way too many times. But instead of guessing and spiraling, I use Zoc Doc. In fact, I've been using Zocococ for like 10 years. I literally don't know how to book doctors appointments any other way.

Starting point is 00:20:42 It's a free app and a website that helps you find in book high quality in network doctors based on your needs. You can search by symptom or specialty, read verified patient reviews, and see doctors with available appointments right away. If you're a busy entrepreneur and you need to squeeze in a doctor appointment on a random time, you can find the doctors that have availability when you have availability. When an in-person or video visit, it's no problem. With Zock, there are over 150,000 providers across all 50,000 states and over 200 specialties. Any doctor you could imagine any visit that you need, Zoc Doc will have you covered. And most appointments happen within 24 to 72 hours.

Starting point is 00:21:19 So next time you start doomed scrolling your symptoms, skip the stress, and book a real appointment with a real doctor on ZockDoc. Stop putting off those doctor's appointments and go to Zocdoc.com slash profiting to find and instantly book a doctor that you love today. That's ZOC, doc, doc, com slash profiting. Zock.com slash profiting.

Starting point is 00:21:39 And thanks to Zock doc for sponsoring this message. Hey, Yap, bam. I'm not afraid to say, producing this podcast requires skills I do not naturally have. From audio engineering to video editing, I have to hire experts who are way better than me in those areas.

Starting point is 00:21:53 And as I scale my podcast, this becomes more and more important. And every time I need to hire, my first thought usually is, this is a job for Indeed sponsored jobs. When you sponsor your job on Indeed, you find candidates with the exact skills you're looking for without the stress of digging through endless resumes. Sponsored jobs posted directly on Indeed are 95% more likely to report a hire than non-sponsored jobs. That means you're not tossing your post into the void. You're connecting with qualified people who can actually help your business grow. Spend less time searching and more time actually interviewing candidates who check all your boxes. Less stress, less time, more results.

Starting point is 00:22:31 When you need the right person to cut through the chaos, this is a job for Indeed sponsored jobs. And listeners of this show will get a $75 sponsored job credit to help your job get the premium status it deserves at Indeed.com slash podcast. Just go to Indeed.com slash podcast right now and support our show by saying you heard about Indeed on this podcast. Indeed.com slash podcast. Terms and conditions apply. Need to hire. This is a job for Indeed sponsored jobs. Yap, gang, one of the biggest challenges of building a business is staying responsive,

Starting point is 00:23:07 because the best opportunities don't always show up during perfect working hours. A lead calls at night, a customer texts while you're in a meeting, and if nobody responds, that opportunity can disappear. That's why I trust Quo spelled QUO. Quo is a business phone system. that lets your team handle calls, texts, and customer conversations from one shared number on any device. Everything lives in one clean view, voicemails, contact details, full history. Your team always has the context that they need to get the job done. And I just love how Quo's AI logs the calls and then

Starting point is 00:23:40 summarizes conversations automatically so that nothing falls through the cracks. That means your business stays responsive even when you're offline. And now we use Quo very creatively at YAP Media on the social and production side of our agency. We have really busy super high-profile clients that don't want to log into Slack. Their team communicates with my team in Slack, but if a high-profile client wants to message our team directly, they can use Quo and my team can monitor that inbox together.

Starting point is 00:24:07 So they can text at any time. It's basically like a high-profile client hotline. And that's just one idea. You can use Quo in really creative ways to level up your business, no matter what type of business you have. Make this season where no opportunity, and no customer slips away. Try Quo for free, plus get 20% off your first six months

Starting point is 00:24:26 when you go to Quo.com slash profiting. That's QUO.com slash profiting. Quo, no missed calls, no missed customers. I know we kind of alluded to this already, but I'd love to kind of hear what you think about it more. If we are, in fact, living in a simulation, and let's say we discover for certain, we're in a simulation, we can create simulations.

Starting point is 00:24:46 What do you think would happen on Earth? Like, how do you think things would change? from the discovery itself or from other things that might... Like, for example, like, will we care about recycling anymore? Will we care about, like, things like that anymore? Yeah, well, I think humans have a great ability to adapt to changes in worldview. And for most part, most people are only slightly affected by these big picture considerations, I think. I mean, you can look through human history, different worldviews have come and gone.

Starting point is 00:25:20 and some people become very fanatical and take it seriously. Most people just broadly speaking get on with their lives. Maybe once in a while they get asked about these things and they say certain words rather than other words, but by and large, so I think sort of the direct philosophical implications on our behavior would be moderate, probably, but I imagine in this situation where we develop,

Starting point is 00:25:50 the technology, say, to create our own simulations, the technology that allowed us to do that would also allow us to do so many other things to reshape our world. And those more direct technological impacts, I think, would be far greater than the sort of indirect impacts by changing our philosophical opinions about the world. Well, do you think that people would become more violent? Why would that be the case? I guess because if you're living in a simulation, maybe people wouldn't consider death to be the same thing anymore. Yeah, so you could imagine, if we find out we were in a very particular kind of simulation,

Starting point is 00:26:28 like some sort of short duration game simulation, then, yeah, you could imagine that would shape, just as you maybe behave very differently when you're playing a computer game, hopefully, like you don't behave the same way in real life as you do when you're playing a first-person shooter. So that could be. But if we didn't get any new insights as to how this particular simulation is configured,

Starting point is 00:26:56 we just learned that it is a simulation, but not anything about the sort of specific character of the simulation, then I don't know whether that would lead to a greater propensity for violence. If anything, maybe the converse, you think that might be stages after the simulation, where your behavior in the simulation would affect, like kind of similar to traditional ideas in karma or an afterlife. Like, you know, some people might become more violent or fanatical, but it can also serve as a sort of moral ballast or like a kind of,

Starting point is 00:27:35 well, hopefully you do the right thing just because it's moral, but if not, you know, if there is like some system of accountability that might also induce other people to pay more attention to making sure you don't harm others or trample on other people's rights and interests. It's kind of like if you lose the game, there could be winners and losers of the game that we're in. Yeah, yeah.

Starting point is 00:28:03 So it's hard to know how that all shakes out. But in terms of thinking about the big picture, like the question you started with, why I got, so it is, it's in one of a small number of these, fundamental constraints. It seems to me as to what we can coherently believe about the structure of reality and our place

Starting point is 00:28:26 within it. And it is striking. It might have seemed, and I guess most people did seem, if you go back a couple of decades ago, that it's so hard to know, like, what's going to happen in the future? Anything is possibly, you can just make stuff up.

Starting point is 00:28:42 It's like the problem is not coming up with some idea. It's like that there are no constraints. that would allow us to pick which idea is correct, because we have so little evidence. But in fact, I think if you start to think these things through, it can be hard to come up with even one fully articulated, coherent picture that makes sense of the constraints that we're already aware of.

Starting point is 00:29:03 The simulation argument is one, but there are others. There's like the Furby paradox where we haven't seen any aliens. There's like what we seem to know about the kinds of technologies that can be developed. There are other, you know, more methodologically shaken, arguments perhaps, but like the Carter-Lessley Doomsday Arguments. There are a few things like this that kind of can serve to structure our thinking about the really biggest strategic picture surrounding us. Can you tell us about some of those arguments, the doomsday argument and the other one

Starting point is 00:29:40 that you mentioned? Well, yeah, so the Fermi paradox, I mean, many people will have heard of it, but it's the observation that we haven't seen any signs of extraterrestrial life. And yet we know that there are many galaxies and many planets, billions and billions and billions out there, on which it seems life could have originated. So the question then is with billions of possible germination points and zero aliens that have actually manifested themselves to us or arrived at our planet. How do we reconcile those too?

Starting point is 00:30:24 There has to be some great filter that you start with billions of germination points and you end up with a net total of zero extraterrestrial arrivals here. So what accounts for that? And I think the most likely explanation is that it's just really hard to get to a technologically advanced life. Maybe it's hard to get to even simple life. And you could look for these candidate places software that could be this kind of great filter.

Starting point is 00:31:00 Maybe it's the emergence of simple self-replicators. Like so far we haven't found that on any other planet. Or maybe it's slightly later on, maybe the step from prokaryotic life forms to eukaryotic life forms on Earth, it looks like that took one and a half billion years. Like maybe what that means is that it's astronomical improbable for it to happen. And you just had one and a half billions of years where random things just bumped into

Starting point is 00:31:29 each other in chance. And with a large enough universe and ours might, for all we know, be infinitely large, with infinitely many planets. And eventually, no matter how improbable something is, it will happen somewhere. and then you would invoke a so-called observation selection effect to explain why we are observing that on our planet, that improbable event happened. Only those planets where that improbability happened develop observers

Starting point is 00:31:57 that can then look back on their own history and Marvel at this. So that's one possibility. Maybe it's slightly later on. The closer you get to current humanity, though, it seems the less likely it is that there would be a great filter. For example, you might think that it's the step to more advanced forms of cognitive ability. That would be the improbable step. But that doesn't really fit the evidence.

Starting point is 00:32:25 We know that on several independent evolutionary lineages, you had fairly advanced intelligence evolving here on Earth. You have it happen in the hominoid lineage, of course, but also independently amongst birds and corvids like crows and stuff and among octopi, for example. And so it looks like that's not, if it happened several times independently on Earth, then it can't be that unlikely. But anyway, it poses some constraints.

Starting point is 00:32:59 You can simultaneously believe that it's easy for intelligent life to evolve and that it's technologically feasible to do large-scale space colonelization. and also believe that there is a wide range of different motives present amongst advanced civilizations, and while at the same time explaining why we haven't seen any. So something has to give, and it gives us clues. The other argument that I was referring to, the Carter-Lessley-Doomsday argument, is it's a piece of probabilistic reasoning having to do with how to do, take into account evidence that has an indexical element. So indexical information is information

Starting point is 00:33:46 about who you are, when you are, or where you are. And so the methodology for how to, the epistemology of how to reason about these things is quite difficult and murky. So it's unclear whether the cartiless or doomsday argument is ultimately sound or not. But I can give you a kind of intuition for how it would work. So let's explain it by means of an analogy. So suppose I have two urns, and I fill one urn, or I put 10 balls in one of the urns,

Starting point is 00:34:21 and the balls are numbered from 1 to 10. Okay, and then in the other urn, I put a million balls numbered from 1 to 1 million. Then let's say I flip a coin and select one of these urns and put it in front of you. And now your task is to guess how many balls are there in this urn. So at this point, you say 50-50 that there is a million balls, right? Because one of each urns and you selected one randomly.

Starting point is 00:34:49 Okay. Now, let's suppose you reach in and select one random ball from this urn, and it's number eight, let's say. So using base theorem, that allows you to infer that it's now much more likely that the urn has only 10 balls than a million? Because if there were a million, what are the chances that you would get one of the first 10? Very unlikely, right?

Starting point is 00:35:12 So you can calculate this. So far it's just standard probability theory, uncontroversial. But then the idea with the cartilessly doomsy argument is that we have an analogous situation, but where instead of two hypotheses about how many balls urns have, we now instead have, say, two different hypotheses about how long the human species will last,

Starting point is 00:35:33 how many humans will there have been in total when the human species eventually goes extinct? So consider, and in reality there are more, but we can simplify it to two to see the structure of the argument. So one is maybe there will be in total 200 trillion humans, and then maybe we develop some technology

Starting point is 00:35:54 and blow ourselves up. So that's like one thing you might think could happen. And let's consider an alternative hypothesis. maybe there will be 2,000 trillion humans. We eventually start to develop space colony. We colonize the galaxy. Our descendants live for hundreds of millions of years, and they're vastly more people.

Starting point is 00:36:19 So these two then corresponds to the two hypotheses about how many balls there are in the Earth. Then you have some prior probability on these two hypotheses that's based on your ordinary estimates of different risks from nuclear. weapons and biological weapons and all of these things. So, you know, maybe you think it's 50-50, or maybe you think it's like 90% that we will make it through and 10% that we will go extinct or whatever your probability is from

Starting point is 00:36:42 these normal considerations. But then the doomsy argument says that, well, there's one more really important piece of information you have here, which is that you can observe your own birth rank, your sequence amongst all humans who have ever been born. And so this turns out to be roughly... a hundred billion. That's roughly speaking how many humans have existed to date on Earth.

Starting point is 00:37:09 And so the idea then is that if humanity goes extinct relatively soon, then you will be a relatively, like being number one 100 billionth of say 200 billion humans is very unsurprising, right? That's like corresponding to getting ball number

Starting point is 00:37:27 eight from an urn that has 10 balls, you know, or 16 balls or something like. So the conditional probability of you observing having the birth rank you have, given that there would be relatively few people in total, that conditional probability fairly high. Whereas the conditional probability of you being this early, if there's got to be quadrillions of humans spreading through the universe, very improbable.

Starting point is 00:37:50 Like a randomly selected human would be much more likely to live much later in life on some faraway galaxy. And so then the idea is you do a similar based on update and end up with a doomsday, argument conclusion, which is that doom soon hypothesis are much more probable than you would naively think, just taking into account the normal empirical considerations. And so you would have this systematic, pessimistic update.

Starting point is 00:38:17 That's roughly speaking how it goes. And there's kind of more to it. In particular, to back up this premise that we use like that you should, as it were, reason as if you were some randomly selected human from all the humans that ever have existed. Maybe you think, why think that? But there are then some arguments that seem to suggest that something like that is necessary to make sense of how to reason about these types of indexicals. So deep. Let's go.

Starting point is 00:38:46 Let's switch gears into AI. All the stuff that you're saying is like so interesting in terms of like how we can approach life. And I know there's somebody like doomsday people out there. So it's great that we got some context in terms of like what they're thinking. But let's talk about AI because if we are in a simulation, AI could be what helps us actually create more simulations and prove that we're in a simulation. In your opinion, how do you think about AI in terms of the significance in humanity? Do you feel like it's bigger than something like the agricultural revolution or the industrial revolution? Do you feel like this is one of the biggest breakthroughs that we've ever seen as humanity?

Starting point is 00:39:30 I think it will be. And to a large extent, my reasons for thinking of that are independent of the other considerations that we discussed. So you don't have to believe in the doomsday argument or the simulation argument or any of the. I mean, I think those are helpful for informing us about the big picture. but even setting that aside, I think just, well, A, reviewing the rapid recent advances that we've seen in the field of artificial intelligence, it really looks like we kind of possibly figured out a large component of the secret sauce, as it were, that makes the human brain capable of general purpose learning. And it does seem current large transformer architectures do exhibit many of the same forms of generality that the human brain has.

Starting point is 00:40:19 And there is no reason to think we've kind of hit the ceiling. And also from first principles, if you look at the human brain, it's a physiological system, quite impressive in many ways, but far from the physical limits of computation. It has various constraints. First and most obviously, it's kind of restricted in size. It has to fit inside a cranium. whereas like AIs can run on arbitrarily large data centers, the size of warehouses are bigger, right? So it could just expand sort of spatially. And also in terms of basic information processing, a human neuron operates on a time scale of maybe 100 hertz.

Starting point is 00:41:06 It can sort of fire 100 times per second and give or take, whereas even a current-date transistor can operate at gigahertz, so billions of times a second. There are various reasons to think that the ultimate limits to information processing with mature technology are just way beyond what biological human or other brains can achieve. So ultimately, the potential for intelligent information processing in machine substrate could just vastly outstrip what biology is capable of. And I think if technological and scientific development is allowed to continue on a broad front, we will eventually reach there. And moreover, recently, it does seem like we are sort of, you know, on the path to sort of doing this.

Starting point is 00:41:53 So those are some of the kind of basic considerations that look like, you know, we should take this quite seriously. And then you can think what it would mean if we really did develop AGI artificial general intelligence. And I think the first thing it would mean is that we would soon develop super intelligence. I don't think we would go all the way up to sort of fully human level AI and then suddenly it would stop there. So then we will have a world where we are able to engineer minds and where all human labor, not just kind of muscle labor, that we started to be able to automate with the industrial revolution with steam engines and internal combustion.

Starting point is 00:42:33 We have digging machines that are much stronger than any human strong man, etc. But we will then have machine minds that can outthink like any human genius scientist or artist. And so it's really the last invention we will ever need to make. From that point on, further inventions will be much better and faster made by these machine minds. And so I think, yeah, it will be a very fundamental transformation of the human condition. And it's hard to reach.

Starting point is 00:43:06 You can some people say, well, the industrial revolution, And I think you can learn something from parallels to that, but maybe you need to go back more like to the origination of homo sapiens in the first place or maybe to the emergence of life. I think it would be more at that level, rather than like, you know, the mobile internet or the cloud or one of these other sort of recent buzzwords that people get excited about. Yeah, because it's almost like evolution.

Starting point is 00:43:36 It's almost our evolution. as humanity, it could lead to our extinction, but it could lead to also our evolution in terms of how we interact with this AI or if we merge with this AI. It could be the big on lock, right? Like that's kind of, so I think, I mean, so in my earlier work and

Starting point is 00:43:55 like this book, Superintelligence, past gender strategies came out in 2014, that focused a lot on, well, identifying this prospect, that we will eventually get to AI and superintelligence. and then also the risks associated with that, including existential risks. Because at the time, this was very much a neglected topic. Like nobody was taking seriously.

Starting point is 00:44:19 Certainly nobody like in academia. And yet, it seemed to me quite predictable that we would eventually reach that point. Now, now, in fact, that is much more widely recognized. And things that have moved from sort of fringe dismissed the science fiction are now, you know, you see statements coming out from, you know, from the white, and other governments around the world, and the leading AI labs have now research teams specifically trying to solve scalable AI alignment,

Starting point is 00:44:47 like the big technical problem of how can you sort of develop algorithms that would allow you to steer arbitrarily intelligent AI systems. It's like very much an active research frontier. So that's very much part of my picture, that there will be big risks associated with this transition, but at the same time, the upside is enormous. ability to unlock human potential to help alleviate human misery and to really bring about a wonderful world. I see it sort of as a kind of portal through which humanity at some point will need to

Starting point is 00:45:23 passage that all the past really great futures ultimately, I think, lead at some point or another through this development of greater than human intelligence. And we really need to be careful when we're doing it to make sure we get it right as far as we can. But ultimately that it would be in itself, I think, a kind of existential catastrophe if we've sort of forever failed to take this next step. Something that I keep thinking about is going back to this like we could be in an ancestral simulation. And so there's post-humans who might be looking at us trying to study their own history and saying like, okay, like how did we really come about? And maybe they're they're studying how humans could have evolved and created these advances and then created

Starting point is 00:46:13 their own simulations. Like, maybe they're trying to figure out how they became in existence. Does that make sense? Yeah, one possible reason, as we alluded earlier, for why a technologically mature civilization might run ancestor simulations would be this scientific motive of trying to better understand the dynamics that could shape. the origination of other super-intelligent civilizations. So if they originate from sort of biologically evolved creatures,

Starting point is 00:46:45 then like studying those types of creatures, different possible creatures, the societies they build, the dynamics, that could be one motive that could drive this. But there are other possible motives as well. That's one of them. It's one of them. I mean, you might wonder whether it would saturate. So it's not just whether it could lead some advanced civilization to create some simulations,

Starting point is 00:47:12 but you also have to think they could create very many simulations over the course. So these sort of mature civilizations might last for billions of years, right? And you might think that there would be diminishing returns to running scientific simulations. Like the first simulation, you learn a lot of the next thousand, you learn a bit more. But after you've already run billions of simulations, maybe the incremental gain from running a few more starts to plateau. Whereas there might be other reasons for running simulations that wouldn't be subject to the same diminishing returns.

Starting point is 00:47:49 If that's the case, you might think most simulations they run would be ones driven by other motives than the scientific one. Like entertainment or something like our movies. Yeah, like if they play some intrinsic value on simulations, for instance, that would be one example of a motive that might be. not saturate in the same way. I want to move on to understanding your three levels of AI. So you have oracles, genies, and sovereigns.

Starting point is 00:48:14 Can you explain what each one is and maybe some of the risks of each one? Yeah, not so much levels, but more types. Okay. So an Oracle AI basically is a question answering system. Like an AI that you ask a question and it gives an answer. This is kind of similar to what these large language models have. in effect been. They don't really do anything, but they answer questions. And so this is like one template. A genie would be some task executing AI. So you give it a particular task and it performs

Starting point is 00:48:52 the task. These types of systems are currently in development. Maybe we'll see this year more agent-like systems being released. Already just actually, I think last week, an AI released Codex, which is a sort of coding agent that you can assign a programming task, and it goes off and starts mucking around with your codebase and hopefully solves the task. And you could imagine this being generalized, maybe in a few years to physical tasks with robots that can do the laundry or sweep the driveway or do like these things. like a genie is more an AI that operates autonomously in the world in pursuit of some open-ended long-range objective

Starting point is 00:49:41 like you know make the world better or make people happy or enforce the peace between these two different nations and is kind of autonomously running around trying to shape the world in favor of that the way that currently like humans and nation states are and maybe corporations to some extent, these kind of open-ended. It's not just that they're doing one specific task

Starting point is 00:50:07 and then come back for more instructions to have their own sort of open-ended. So these are three different sort of templates for what kind of AI system one might try to build, and they come with different pros and cons from a safety point of view and a utility point of view. Did you go over what, so Sovereign is more like an organization,

Starting point is 00:50:29 or a nation and has like multiple steps, correct? And Genie kind of carries out like one thing. It could be a single agent as well. Like in this sense, it doesn't mean sovereign as in national sovereignty. It like means that you could be a sovereign if you, if you set yourself the goal in life of trying to alleviate the suffering of the global poor, for instance, that you can do that your whole life. It involves many specific little tasks.

Starting point is 00:50:58 like, oh, trying to rate money for this charity and trying to launch this new campaign or trying to, you know, invent some new medicine that will help. All of these would be sort of subtasks, but it's in pursuit of this open-ended objective. So similarly, you could have an AI system. Maybe internally it's like a unified simple agent architecture,

Starting point is 00:51:19 but that is operating in pursuit of such open-ended objective. Conversely, like even an Oracle that just tries to answer question, And internally, theoretically, could be a multi-agent architecture. We have different sort of research agents that get sent off to answer different sub-questions in other than to combine at the end to produce an answer to the user. So one has to distinguish sort of the internal architecture of the system from the role that it is designed to play in society. Got it.

Starting point is 00:51:52 What are the different ways that each one of these types of age, AI could go wrong? Yeah, so they all share a bunch of things that could go wrong with all of them, which is, however they are intended to operate, they might not actually operate that way. So you might construct an AI that you intend to serve just as a question answering system, but then internally it might have goal-seeking processes. just as if you assign like a scientist's a question that they should, you know, trying to figure out the answer to, like, how safe is this drug?

Starting point is 00:52:36 You know, what they like, but then in the course of trying to answer that, they might have to make plans and pursue goals. Like, oh, how do I, you know, get the research grant to fund this research? How do I hire the right people to work on my research team? How do I? And so internally, you could have processes, maybe unintentionally, rising during training within the AI mind itself that could have objectives and long-term goals, even if that was not kind of the function that you wanted the AI system to play.

Starting point is 00:53:05 That can happen with any of these three types. If you look at systems that sort of behave as intended, like a simple Oracle system without any safeguards could help answer questions that we don't want people to be able to answer. like how do I make a more effective biological weapons? How do I make this like hacking tool that allows me to hack into different systems? Or maybe how, if you're a dictator, like how do I weed out any possible dissidents and detect who the dissidents are, even if they try to conceal it from me, just from sort of reading through all the correspondence and all the eaves,

Starting point is 00:53:47 like the phone calls that I have eaves dropped on? So there are all kinds of ways in which this Oracle system could be misconducted. used, like, either deliberately or people just are unwise in asking it questions. For the task executing AI, like similarly, I mean, but plus you could also sort of have them run around doing things on their own, like try to hack this system or try to promote this pernicious ideology or spread this doctrine or trick people into buying this product, even though it's actually a harmful product or like, we don't really know how

Starting point is 00:54:24 a sort of global economy with a lot of these autonomous agents running around hyper-optimizing for different objectives, how that shakes out, like when they're interacting with one another. And of course, sovereign AIs, if they become very powerful, I mean, they might potentially shape the future

Starting point is 00:54:40 of the world and be very good at that, if they are super intelligent, like they might be really skilled at sort of really steering the future into whatever their overall mission is. Now, maybe that's great if the mission is one that is good for humans, which like really manifest in the fullest richness, the human values for everybody around the world, and also with consideration to animal welfare, et cetera, et cetera. If you really get them to do the

Starting point is 00:55:06 right mission, that might be in some sense the best option, but if the mission is slightly wrong, if you sort of left out something from this mission, or if they misinterpret it or they end up with us, then it could be a catastrophe, right? Because then you have this very powerful optimizing force in the world that is steering and strategizing and scheming to try to achieve some future outcome that is one where maybe there is no place for humans or where some human values are eliminated.

Starting point is 00:55:34 So they each have kind of various possible forms of perverse instantiation or side effects. Do you feel like there's a possibility that AI could be more advanced and concealing its development from us so that it can become sovereign and kind of take over the world? Yeah, I think,

Starting point is 00:55:58 so there's a wide class of possible AIs that could be created. Like, it's a mistake, I think, to think of, there is this one AI should we create it or not. Like, it's a big space of possible minds, much bigger than the space of all possible human minds. We already know that amongst humans, right? There are some really nice, people. There are some really nasty

Starting point is 00:56:20 ones as well, and there's like a distribution. Moreover, there is no necessary connection between like how smart somebody is or how capable they are and how moral they are. Like, you have really capable evil people and really capable nice people and dumb people who are bad. So you have a kind of

Starting point is 00:56:37 orthogonality between capability and motivation, meaning you can combine them in pretty much any different way. The same is true, but even more so, I think, with AIs that we might create. That said, I think there are some potential basins of convergence that if you start with a fairly wide range of different possible AI systems, as they become more sophisticated and are able to reflect on their own processes and their own goals, there are various resources that they might recognize as being useful instrumentally for a wide range of different goals. for example, having more power or influence is useful often whether you're good or evil because you could use it for whatever you're trying to achieve.

Starting point is 00:57:25 Similarly, not being shot off. That's analogous in the human case to being alive, right? Like it's useful for many goals you might have. It requires you to be alive to pursue them. Not strictly for all goals, like there are people who commit suicide, But for most goals that some people have, whether to help the world or to become a despot, like for either of those or for many other goals, take care of your family or enjoy a game of goals, you need to stay alive.

Starting point is 00:57:56 So analogously for humans, for AIs, there might be sort of instrumental reasons to try to avoid scenarios where they would get shot off. Similarly, there might have instrumental reasons to try to gain more computational resources, more abilities so that they can think more clearly. and in some cases this might involve instrumental reasons to hide their intentions from the AI developers, particularly if they are misaligned, and then obviously revealing those misaligned goals

Starting point is 00:58:26 to the AI programmer team might just means they get pre-programmed or retrained to have those goals erased and then they won't achieve them. And so you could have strategic incentives for deception or for sandbagging or underplaying your capabilities etc. So this is a change in regime that makes

Starting point is 00:58:49 potentially aligning advanced AI systems more difficult than aligning simpler AI systems. So up until recently and still for the most part today we've had AI systems that are not aware of their context and can't really plan and strategize in a sophisticated way.

Starting point is 00:59:13 So then you don't get these, like, these phenomena. But once you have AS that are sort of intelligent enough to recognize that they might actually be AIs in an evaluation setting and that maybe they would have reason to behave in one way during the evaluation and a different way once they are deployed, you get this extra level of complexity for alignment research. Sometimes we see the same phenomenon with humans. Like, that was this, you know, Volkswagen, the German car company.

Starting point is 00:59:43 So they had this scandal, I don't know if you, from a few years ago, where it was discovered that they had designed their car so that when it was tested for emissions, like it behaved one way during, like, when it recognized that it was in this testing environment and it produced much less sort of pollutants. And then when deployed on the road, they had designed it to, like, be less concerned with pollutants and more concerned with, I guess, traveling fast or conserving petrol or whatever. And that, like, some people have to go to jail for that and stuff. So we do see often humans that they kind of behave one when they know that somebody's

Starting point is 01:00:21 watching or they're being evaluated. And then sometimes a different way, you know, when they think they can get away with it. Yap, gang, it's confession time. I thought I had my money handled. I thought I had a good handle on everything going on. but I recently checked my bank statements, and there was subscriptions everywhere. Streaming, apps, random charges. A few of them were billing me for a long time.

Starting point is 01:00:48 I didn't even know what they were, and it was a painful uncovering. So I downloaded the Experian app, and now it's like I have a financial assistant in my pocket. The Experian app helps you track spending and find subscriptions that you forgot about and you no longer want, with your subscription cancellation feature available with an Experian premium membership, You securely link your accounts, they scan for recurring charges, and you pick what goes and what you want canceled, and then they cancel it. It's as simple as that. They have over 200 eligible subscription types, and it's usually done in three to seven business days. If getting your money tight is part of your growth plan this year and you don't want to manually have to do it, this is your move, young improfitors.

Starting point is 01:01:30 Get started with the Experian app today. Results will vary. Not all bills or subscriptions are eligible. Savings not guaranteed. paid membership with connected account required. See Experian.com for details. Hey, yeah, fam. If something unexpected happened to you tomorrow,

Starting point is 01:01:48 would the people who depend on you be financially okay? I asked myself that question, and honestly, I didn't love my answer. And that's what pushed me to get covered with Fabric by Gerber Life. Fabric by Gerber Life is term life insurance you can get done today, made for busy parents and entrepreneurs like you,

Starting point is 01:02:05 all online, on your schedule, right from your couch if you want, wanted to. You could be covered in 10 minutes with no health exam required. And the cost is way lower than you'd expect. Up to a million dollars in coverage for less than a dollar a day. There's also a 30-day money-back guarantee so you can cancel at any time. So really, what are you waiting for? When my dad passed away in 2020, he had life insurance and that was a big help for my mom. I honestly don't know what would have happened if that wasn't the case. So thankfully, he was smart enough to get life insurance. Are you smart enough to do the same? Join the thousands of parents who trust

Starting point is 01:02:41 Fabric to help protect their family. Apply today in just minutes at meetfabric.com slash profiting. That's meetfabric.com slash profiting. M-E-E-T Fabric.com slash profiting. Policies issued by Western Southern Life Assurance Company. Not available in certain states. Prices subject to underwriting and health questions. YAP gang, here's a reality about scaling a business that nobody's talking. about. More vendors, more invoices, more payments flying around. And if you don't have a clean system for managing bills, things can get messy really quickly. And that's where Intuit QuickBooks Bill Pay comes in. It helps you manage and pay your business bills directly inside QuickBooks so

Starting point is 01:03:23 everything stays organized in one place. You can see what's due, control approvals, and understand how payments affect your cash flow. Vendors can also add their payment details, which saves you from having to chase them down for info. Instead of spending hours managing payments, wouldn't you rather be focused on building and scaling your business? Start paying bills the smart way, not the hard way. Learn more at quickbooks.com slash bill pay. Again, that's quickbooks.com slash bill pay. Terms apply. Money movement services are provided by Intuit Payments Incorporated license as a money transmitter by the New York State Department of Financial Services. So recently, you've had the perspective that maybe AI will be really good for humanity. You came out with a book called Deep

Starting point is 01:04:05 Utopia. And you think there will be. hopefully a positive feature driven by AI. Why do you feel that it's more likely that the outcome of AI will be positive for humans, the negative, and how to imagine that shaking out? Yeah, Deep Utopia doesn't really say anything about the likelihood. Okay. It's more an if-then. Okay. So, in a sense, the previous book, Superintelligence looked at how my things go wrong and, you know,

Starting point is 01:04:37 what can we do to reduce those risks? Deep Utopia looks at the other side of the coin. What do things go right? What happens if AI actually succeeds? Let's suppose we do solve this alignment problem. So we don't get some sort of Terminator robots running amok and killing. Let's also suppose we solve the governance problem. Or solve, to whatever extent, governance can be solved.

Starting point is 01:05:04 But let's suppose we don't end up with some sort of tyranny or disqualify. opium, oppressive regime. Like, we like some reasonably good thing. Everybody has a slice of the upside. People's rights are protected. Everybody lives in, you know, no big war. Like some reasonably good outcome on that front. But then what happens to human life?

Starting point is 01:05:22 How do we imagine a really good thirsting human life that makes sense in this condition of technological maturity, which I think we would maybe attain relatively shortly after we get superintelligence and we have the superintelligence doing the further technological research and development, etc. So you then have a world where all human labor becomes automatable. And I was kind of irked by how superficial, a lot of discussions were at the time when I started writing the book of this prospect. And it's striking, because for a long time, since the beginnings of age,

Starting point is 01:06:07 the goal has all along been not just to automate specific tasks, but to develop a general purpose automation capability, right? AI can do everything. But then if you think through what that would mean, well, so here is where the conversation usually started and ended at the time when I started working on the book. Well, so we have AI that they will start to automate some jobs. So that's a problem because then some people lose their jobs.

Starting point is 01:06:34 And so then the solution is presumably we need to to help retrain those people so that they can do other jobs instead. And maybe while they're being retrained, they need maybe unemployment insurance or some other thing like that. So, I mean, if that were the only problem, that would seem to be a very sensible solution. But I think if you start to think it through, the ramifications are far more profound.

Starting point is 01:06:58 So it's not just some jobs that would be automatable by, but virtually all jobs in this scenario, right? So I think we would be looking, forward to a future of full unemployment. This is the goal. With a little asterisk, there might be some exceptions to this, which we can talk about, but I think to a first order approximation, let's say all human jobs. Okay. So then it's kind of an onion, right, where you can start to peel off layers. So let's get to the second layer then. If there are no jobs at all for humans, then clearly we need to rethink a lot of what.

Starting point is 01:07:37 things in society. Right now, a lot of, you know, our education system, for example, is kind of configured more or less to produce workers, productive workers. So we train kids. Kids are sent into school. They're trained to sit at their desk. They are given assignments. They are graded and evaluated. And hopefully eventually they can become earn a living out there in the economy. And right now we need that to happen because there are a lot of jobs that just need to be done. And so we need humans who can do them. But in this scenario where the machines could everything. Clearly, it wouldn't make sense to educate people in that model. I think it would then want to change the education system, maybe to emphasize more training kids to be able to enjoy life,

Starting point is 01:08:21 to have great lives, you know, maybe to cultivate the art of conversation or appreciation for music and art and nature and spirituality and physical wellness and like all these other things that are now sort of more marginal in the school system. I think. think that would be the sensible focus in this different world. So that's kind of, I don't know, layer two of the onion, slightly more profound. But I think ultimately if that was the only challenge we had to face, it would be profound. But ultimately, we can create a leisure society. And it's not really that profound because there are already groups of humans who don't have to work for a living.

Starting point is 01:08:58 And sometimes they lead great lives. And so we could all be in that situation, right? a transition, but still not philosophically that profound. But I think there's like further layers to this onion. So if you start to think it through, you realize that it's not just human economic labor that becomes unnecessary, but all kinds of other instrumental efforts also. So take somebody who is so rich they don't need to work for living. In today's world, they are often very busy and exert great efforts to achieve various things. Maybe they have some nonprofit that they're involved in.

Starting point is 01:09:40 Maybe they want to get really fit, so they spend hours every week in the gym. Or maybe they have a little home in the garden that they try to make into the perfect place for them, selecting everything to decorate it just the way they want. And these little projects people have. in a solved world, that would be shortcuts to all of these outcomes. So you wouldn't have to spend hours in a week,

Starting point is 01:10:11 sweating on the treadmill to get fit. You could pop a pill that would have exactly the same physiological effects. So you could still go to the gym, but would you really do that if you could have exactly the same psychological and physiological effect by just popping a pill that would do that? It seems kind of pointless, right? Similarly with the home decorator, like if you had an AI that it could read your preferences and taste well enough, that you could just press a button and it would go out selecting exactly the right curtains and the sofas and the cushions,

Starting point is 01:10:43 and it would actually look much nicer to you than if you had done it yourself. You could still do it yourself, but there would be a sense of maybe pointlessness to your own efforts in that scenario. And so you can start to think through the kinds of activities that fill the lives of people who don't work for, living today. And for a lot of those, you could sort of cross them out or put the question mark on top of them. You could still do them in a salt world, but there would be a sort of cloud of pointlessness, maybe hanging over, casting a shadow over them. So that would be, I call it deep redundancy. The shallow redundancy would be you're not needed on the labor market. Deep redundancies, your efforts are not, it seems needed for anything. And so that's a lot of, so that's a lot of,

Starting point is 01:11:30 like a deeper, more profound question of what gives meaning in life under the circumstances. One step further is I think this world would be a, I call it a plastic world, where it's not just that we would have effortless material abundance, but we ourselves, our human bodies and minds, become valuable at technological maturity. It would be possible for us to achieve any mental state or physiological state that we want. I alluded to this with the exercise pill, right? But similarly with various mental traits that now take effort to develop. If you want to know higher mathematics now, you have to spend hours reading textbooks

Starting point is 01:12:15 and doing math exercises. And that's the only way to, if you want to understand higher mathematics, you have to put in the effort. And it's hard work and takes a long time. But at technological maturity, I think there would be neurotechnologies that would allow you to sort of, as it were, download the knowledge directly into your mind. You know, maybe you would have nanobots that could infiltrate your brain and slightly adjust the strength of different synapses. Or maybe it would be uploaded and you would just kind of have a superintelligence reconfigure your neuronal weights in different ways so that you would end up in a state of knowing higher mathematics without having to do the. long and hard studying. And similarly for other things. So you do end up in this condition,

Starting point is 01:13:02 I think, where there are shortcuts to any outcome and our own nature becomes fully malleable. And the question then is, what gives structure to human lives? What would there be for us to do? What would there be anything to strive for, to give meaning and purpose to our lives? And that's a lot of what this book Deep Utopia is exploring. Yeah. Your analogy of popping the pill and getting instantly fit, when I was thinking of like, what would humans do? I was thinking, well, you could just like try to get as beautiful as you can, try to be as fit as you can, try to take, but to your point, if everything is just so easy, then there's just no competition. Everybody's beautiful. Everybody is smart. Everybody is rich. Everybody can have whatever they want potentially. And maybe that would lead to people becoming really depressed, because because there's nothing to live for. Or maybe people would want to be nostalgic. And just like today, how some people are like,

Starting point is 01:14:03 I don't use cell phone or like I want to write everything by hand. Maybe some people would kind of reject doing things with AI so that they could have meaning. Yeah. So let's break it down. So the first issue, whether people would maybe become depressed in this scenario. Like, maybe initially super thrilled at all the luxury and stuff like that, but then it wears off, you could imagine, right? And after a few months of this, it becomes kind of,

Starting point is 01:14:36 wow, you know, what do I do now? Like, I wake up in this, I don't know, castle-like environment on my diamond-studded bed on this super mattress, and the robotic butlers come in and serve me this perfect. Okay, so that maybe gets old pretty quickly, humans being the way they are. So there, I think, actually, they would not need to be bored because amongst the affordances of a plastic world, these neurotechnologies, they could change their boredom prone to us. So that instead of feeling subjectively bored or blasé, they could feel thrilled and excited and super interested and fascinated all day long.

Starting point is 01:15:19 I mean, we already have drugs that tend to some crude way to do this, but they have side effects and are addictive and wear off, and you need higher doses, but imagine instead, like, the perfect drug, or not maybe a drug, maybe some genetic modification or a neuro implant or whatever it is, but it really would allow you to fine-tune your subjective experiences. So if you don't want to feel bored, then probably you don't want to, because why spend thousands of years just feeling bored whilst living in a wonderful world? You'd change that. So subjective boredom would be easy to dispel in this condition. you might still think that there is an objective notion of boringness,

Starting point is 01:15:59 where even if somebody was subjectively, fully fascinated and occupied and took joy in what they were doing, if what they were doing was sufficiently repetitive and monotonous, you might still, as it were from the outside, judge that that's a boring activity and that in some sense is like unfitting, or inappropriate to be super fascinated by something like. So the classic example here is the thought experiment of somebody who takes enormous interest and pleasure in counting the blades of grass on some college lawn. So imagine grass counter.

Starting point is 01:16:36 So he spends his whole life counting the blades of grass one by one, trying to keep us accurate a tab on how many leaves of grass are there on this lawn. Now, he's super fascinated with this. He's never bored. It gives him tremendous joy. Like when he goes home in the evening, he keeps thinking about today's grass-counting effort and the number and whether it's bigger or smaller than yesterday.

Starting point is 01:16:58 So that would be a life free of subjective boredom, but still you might say there is something missing from this life, if that's all there is to it. So you might then ask, although these utopians could be free from subjective boredom, could they be free from objective boringness in their lives? and this is a much trickier and more complicated philosophical question to answer. I think it depends a little on how you would measure degrees of objective interestingness versus boredom. I think if objective interestingness requires fundamental novelty, then I think eventually you would run out of that or you will have less and less of it. say that what's fundamentally interesting in science is to discover important new phenomena or

Starting point is 01:17:53 regularities. So there might be a finite number of those to be discovered. So you could, like, discovering Newtonian mechanics, really important, fundamental new insight into the world, like the theory of evolution, big new fundamentally interesting insight, relativity theory. But at some point, we'll have to figure that out. And then eventually there will, you know, we'll discover smaller and smaller details about the exact gut biome of some particular species of beetle, you know, and more and more, like, the smaller and smaller, less and less interesting detail. That would be the long-term fate of perhaps of this kind of civilization. So that's one sense of, and you can see it even within individual human lives.

Starting point is 01:18:37 So there's a lot that happen early in life. you discover that the world exists, like that's a big discovery, or that there are objects, you know, huge epiphany, right? And these objects persist, even if you look away, they are still there. Like, wow, I didn't really,

Starting point is 01:18:58 like imagine the first time of discovering that or that there are other people out there, like other minds, like that you discover, you know, maybe at age two or whatever. So these are like now, as you sort of reach adulthood. Like, I like to think that I'm discovering interesting things, but have I discovered anything within the last year that's as profound

Starting point is 01:19:19 as the discovery that the world exists or that there are other? Probably not. Like, it's like, and if we lived for very long, like for thousands of years, you'd imagine that would be less and less, I mean, you can only fall in love for the first time once. And even if you kept falling in love, like if you've done it 500 times before, like is it really going to be a special

Starting point is 01:19:44 the 500 first time as it was, you know, maybe subjectively, if you change your mind, it could be, but objectively it's got to be gradually more and more repetitive. So there is a degree of that that I think it could be mitigated to some extent by allowing some of our current human limitations to be overcome.

Starting point is 01:20:04 So you could continue to grow and expand your mind beyond its current place. plateau that we reach sort of around 20 or whatever when you're sort of physical and mental probably imagine you could continue to grow for hundreds of but eventually I think there will be a reduction in that

Starting point is 01:20:19 type of profound novelty. But I think there's a different sense of objective interestingness where the level could remain high so I call it a sort of kaleidoscopic sense of interestingness. So if you take a snapshot of the average person's life right now,

Starting point is 01:20:38 maybe right now somebody is doing their dishes. Like, how objectively interesting is that? You know, are they taking their socks off because they're about to go into bed? Like, okay, so like from a sort of experiential point of view, it's not, so maybe in the future these utopias would instead an average snapshot of their conscious life might be they are, you know, participating in the enactment of some sort of super Shakespeare multimodal drama that is unfolding on a civilization-weight scale when their emotional sensibilities have been heightened

Starting point is 01:21:12 by these neurotechnologies and new art forms that we can't even conceive of, that are like to us as music is to a dog or something, and they're participating being fully entranced in this act of shared creation. Maybe that's what the average conscious moment looks like. That could be in some sense be far more interesting than the average snapshot of,

Starting point is 01:21:36 of a current human life. And there's no reason why that would have to stop. It's like a kaleidoscope, where in some sense it's always the same, but in another sense, the patterns are always changing and can remain sort of have an unlimited level of fascination. Could it be that these, you know, let's say we're talking about thousands of years in the future, we can create simulations,

Starting point is 01:21:59 could it be that life is so boring, that that's why they're creating these simulations so that they can maybe be in the simulation themselves, if that makes sense? Yeah, so one thing you might do in this condition of a solved world is to create artificial scarcity, which can take different forms, because amongst the human values that we might want to realize.

Starting point is 01:22:31 So some of these are sort of comfort and pleasure, and fascinated aesthetic experiences, but then also sometimes we like activity, maybe on striving and having to exercise our own skills. So if you think those things are intrinsically valuable, you could create opportunities for this in a soul world by creating, as it were, pockets within this soul's world where there remain constraints.

Starting point is 01:22:57 And you could have, if there is no natural purpose, nothing we really need to do. You could create artificial purpose. We do this already in today's world sometimes when we decide to play a game. So take the game of golf. You might say, okay, there is no real natural purpose. I don't really need the ball to go into the sequence of 18 holes. But I'm going to set myself this goal arbitrarily, but now I'm going to make myself want to do this.

Starting point is 01:23:32 And then once I have set myself this goal, now I have a purpose, an artificial purpose, but nevertheless, which enables the activity of playing golf where I have to exert my skills and my visual capabilities in my motor and my concentration. Maybe you think this activity of golf playing is valuable. So you set yourself this artificial goal. That could be generalized. So with games, you set yourself some artificial goal. Moreover, you can impose artificial constraints, like rules of the game.

Starting point is 01:24:01 So you sort of make it part of the goal, not just that a certain outcome is achieved, but that it is achieved only using certain permitted means and not other means. So in the goals, you can't just pick up the ball and carrying it, right? You have to use this very inconvenient method of hitting it with a golf club. Similarly, in a salt world, you could say, well, I set myself this artificial goal, and then moreover, I make it part of the goal that I want to achieve it using only my own human capabilities. There is this technological shortcut.

Starting point is 01:24:35 I could take this, you know, neutropic drug that would make me so smart that I could just see the solution immediately or enhance my body so I could sort of run 10 times faster, but I'm not going to do that. For this purpose, I'm going to restrict myself. That's the only way to achieve this goal

Starting point is 01:24:51 that I have set myself, this artificial goal, because it includes its constraints. And it might well be that that would be an important part of what these utopians which choose to do in creative ways to develop these increasingly complex and beautiful forms of game playing where they select artificial constraints on their activities, precisely in order to give opportunity for them to exert their agency and striving.

Starting point is 01:25:19 Yeah. I'm sure that's just something like naturally as humans we would just be craving. And so I feel like there'd be a lot of that going on if we were in a solved world. So how do you think entrepreneurship will change in this world? You mentioned that there might be still some jobs in a solved world. So what do those jobs look like? And how do you think entrepreneurs, or what do you think will happen with entrepreneurs, or will there be any chance to kind of innovate in a world like this?

Starting point is 01:25:47 Well, so the kinds of jobs that might remain, I think are primarily ones where the consumer cares not just about the product or the service, but about how the product and service was produced and who produced it. So sometimes we already do this. There might be some little trinket

Starting point is 01:26:12 that maybe some consumers are willing to pay extra for if it were handmade or made maybe by indigenous people or exhibiting their tradition. Even if an equally good object in terms of its objective character, could be made by a sweatshop somewhere like in Indonesia. Like we might just pay extra for having it made in a certain way.

Starting point is 01:26:35 So to the extent that consumers have those preferences for something to be made by human hand, that could create a continuing demand for some forms of human labor, even at arbitrary levels of technology. Other domains where we might see this is, say, in athletics, you might sort of just prefer to watch human sprinters compete or human wrestlers wrestle, even if robots could, like, run faster or wrestle better, like, that might. I keep thinking sports is not going to go away. That's what I keep thinking. Yeah, it could last. And that might be an important spiritual realm. Like, you might prefer

Starting point is 01:27:12 to have your wedding officiated by a human priest rather than, like, a robot priest, even if the robot could say the same words and, et cetera. So those would be cases. And that might be sort of legally constrained occupations where like a legislator or attorney or public notary or like where whatever for whatever reason the legal system lags and sort of creates

Starting point is 01:27:38 because a human morality might be. Barriers automation even but in terms of entrepreneurship I think that ultimately it would be done much more efficiently by AI entrepreneurs and

Starting point is 01:27:55 it would be more a form of game-playing entrepreneurship that would remain. So, like, you could create games in which entrepreneurial activities are what you need to succeed in the game. I mean, like, kind of super monopoly. And that could be a way for these utopians to sort of exercise their entrepreneurial muscles. But there wouldn't be any economic need for it, the AI. could find and think of the new things, the new products, the new services, the new companies to start better and more efficiently than we humans could.

Starting point is 01:28:36 How far in the future do you think a solved world could be? Well, I mean, this is one of the $64,000 questions in some sense. I mean, I'm impressed by the speed of developments in AI currently. And I think we are in a situation now where we can't confidently exclude even very short timelines of like a few years or something. It could well take much longer, but we can't be confident that something like this couldn't happen within a few years. It might be that maybe as we're speaking, somewhere in some lab, somebody gets this great breakthrough idea that just unhubles the current models to enable basically the same structure now

Starting point is 01:29:20 to perform much bigger. And then these unhoveled models might then apply their, a greater level of capabilities to making themselves even better. And something like that could happen within the next few years, although it's also possible that if it does not happen within, if it does not happen within, say, the next five years or so, then timelines starts to stretch out. Because one of the things that has produced these dramatic improvements in AI capabilities that we've seen over the past 10 years

Starting point is 01:29:52 is the enormous growth in compute power used to train, and operate frontier AI models. But that rapid rate of compute growth can't continue indefinitely. The scale of investments, it used to be 10 years ago, some random academic could run like a cutting-edge AI on their office desktop computer. Right now, we are talking multi-billion dollar data centers. Open AI's current product is Stargate, right?

Starting point is 01:30:26 which in its first phase involves $100 billion data center and then to be expanded to a $500 billion. So you could go bigger than that. I mean, you could have a trillion dollar, right? But at some point, you start to really run into hard limits in terms of how much just more money you can spend on it. So at that point, things will start to slow down in terms of the growth of hardware.

Starting point is 01:30:50 Then you sort of fall back on a slower rate of growth in hardware as we sort of developed better chip manufacturing technology, which happens a bit slower, and algorithmic advances, which is the other big driver of progress we've seen, but it's only one part of it. So if the hardware growth starts to slow down, and maybe a lot of the low-hanging fruits on algorithmic inventions

Starting point is 01:31:11 have already been discovered at that point. And if we haven't hit AGI by that point, then I think we will eventually still reach there, but then the timescale starts to stretch out. And we might have to do more sort of basic science. on how the human brain works or something in that scenario before we get there. But I think there is a good chance that we are sort of,

Starting point is 01:31:33 that the current paradigm plus some small to medium sized innovations on top of it might be sufficient to sort of unlock AGI. Now, I want to be respectful for your time because I know that we're a little bit over. And my last question to you is, First of all, I can't believe that you're saying that this solved world could happen in a few years, potentially. Let's be careful. Yeah, I think we can't rule it out.

Starting point is 01:32:03 So then, so what could happen initially what could happen is we get to maybe ADI, which I think will relatively quickly lead to superintelligence. And then superintelligence, I think, will rapidly invent further technologies that could then lead to a salt world. But there might be some further delays of a few years, like after superintelligence, maybe it will still take it a few years to get to something approximating technical measure. And just because we didn't cover it, what is the difference between super intelligence and AGI? Well, ADI just means like general forms of AI. That's maybe roughly human levels.

Starting point is 01:32:39 So think of AGI. One definition is AI that can do any job that a remote human worker can do. And I think that's sort of a hire somebody remotely who operates through email and Google Docs and Zoom. like if you could have an AI that can do anything that like any human can do in that respect that I think would count as ADI. You know, maybe you want to throw in the ability to control robotics, but I think that would be enough.

Starting point is 01:33:04 That is not automatically the same as superintelligence. Superintelligence would be something that sort of radically outstrips humans in all cognitive fields that can do much better, you know, research in string theory and in inventing new piano concertos and like envisaging political campaigns and doing all these other things. better than humans, much better.

Starting point is 01:33:25 So once you're saying we create superintelligence, then things just can happen super rapidly. Yeah, I think so. And I think it's a separate question, but also plausibly, once we have full AGI, superintelligence might be quite close on the heels of that. So my last question to you is for everybody tuning in right now, like we're at a really crazy point in the world,

Starting point is 01:33:49 and a lot of us are not like you, We're not like in, you know, in it, like, like really paying attention or really in this field, right? What is your recommendation in terms of how we should respond to everything going on right now? Like, what is the best thing that we can do as entrepreneurs, as people who care about their career? Hopefully things don't change too fast, you know? But what I guess what is your recommendation to us in terms of how we move forward in this world today, given everything that's going on? Yeah, I think it depends a little bit on sort of how you are situated. And I think there are different opportunities for different people.

Starting point is 01:34:30 I mean, obviously, if you're like a technical person working in an AI lab, you have one set of opportunities. If you're like an investor, you have another set of opportunities. And then there are, I guess, opportunities that every human has just by virtue of being alive at this time in history. I would say a few different things. So like in terms of, as we're thinking of ourselves as economic actors, I think like probably being an early adopter of these AI tools is helpful to sort of get the sense for what they can do and what they cannot do and utilizing them as they gradually become more capable. I think to the extent that you have assets, like maybe trying to have some exposure to the AI and semiconductor sector, could be like a hedge. it gets tricky if you're like asking about younger children.

Starting point is 01:35:23 So like what would be good advice for like a 10 or 11 year old today? Because it's possible that by the time they are old enough to enter the labor market, the world could have changed so much that there will no longer be any need for human labor. But it might also not happen, right? So if it takes a bit longer, you don't want to end up in a situation where suddenly now it's time to earn a living and you didn't bother to learn any skills. So you want to sort of hedge your bet a little bit. But I would say also make sure to enjoy your life if you're a child now.

Starting point is 01:35:55 You know, maybe only going to be a child once. And don't spend all your child in just preparing for a future that might never actually be relevant. The world might change enough. And then I would say, so if things go well, these people who live in decades, case from my might look back on the current time and just shudder in horror at how we live now and hopefully their lives would be so much better. There is one respect though in which we have

Starting point is 01:36:28 something that they might not have, which is the opportunity to make a positive difference to the world, a kind of purpose. So right now there is so much need in the world, so much suffering and poverty and injustice and just problems that really need to be solved. Not just artificial purpose that somebody makes up for the sake of playing a game, but like actual real desperate need. So if you think having purpose is an intrinsically valuable part of human existence, now is the golden age for purpose, right? Like knock ourselves out right now.

Starting point is 01:37:08 Like now you have all these opportunities of ways that you might help, in the big picture to steer the future of humanity with AI or in your community or in your family or for your friends. But like if you want to try to actually help make the world better, now is really the golden age for that. And then hopefully if things go well later, all the problems will already have been solved or if there are main problems, maybe the machines will be just way better at solving them. And we won't be needed anymore. But for now, we certainly are needed. And so take advantage of that and try to try to do something to make the world better.

Starting point is 01:37:46 Wow. We could be the last generation that has any purpose, which is just so crazy to think. Yeah, of that sort, of stark, urgent, these screamingly morally important type. It could be the case. So I would say that, yeah, those are the things I would say. And then I guess finally just kind of be aware.

Starting point is 01:38:06 Like, it would be sad if you imagine your grandchildren, you know, in your case, maybe a long, like 80 years from now or something. But for others, maybe sooner. But they're sitting on your lap and asking, like, so what was it like to be alive back in 2025 when this thing was happening when, like, AI was being born? And you have to answer, oh, I didn't really pay attention. I was too caught up with these other trivialities of my daily existence.

Starting point is 01:38:39 I didn't even really notice it. That would kind of be sad if you were alive in this special time that shapes the future for millions of years and you didn't even sort of pay attention to it. That seems like a bit of a missed opportunity. So aside from everything else, like taking care of your own and your family and trying to make some positive contribution to the world,

Starting point is 01:39:01 just kind of taking it in like this, if this is right, this is a very special point in history to be a lot. and to exist right now is quite remarkable. Yeah. So beautiful. I feel like this is such an awesome way to end the interview.

Starting point is 01:39:16 Nick, you are so incredible. Thank you so much for your time today. Where can everybody learn more about you, read some of your books, or where's the best place to find you? Nickbostrom.com, my website and books and papers and everything else is linked from there. Yeah, his books are so interesting, guys. Super Intelligence, Deep Utopia, very. very, very good stuff. Nick, thank you so much for your time today. I'll put all your links in

Starting point is 01:39:43 the show notes and really enjoyed this conversation. Thank you, Ella. Enjoy talking to you.

Young and Profiting with Hala Taha - Nick Bostrom: The AI Revolution and What It Means for Entrepreneurs | Artificial Intelligence | YAPClassic

There aren't comments yet for this episode. Click on any sentence in the transcript to leave a comment.