Joseph Carlsmith - Utopia, AI, & Infinite Ethics

1. 0:00 – 0:55

   Preview

   1. JCJoseph (Joe) Carlsmith0:00

      So utopia for me just means a kind of profoundly better future. And I think it's important because I think it's just actually possible. I just think it's actually something that we could do. If, if we sort of play our cards right, we could just build a world that is radically better than the world we live in today. (screen swooshes) Infinite ethics is ethics that tries to grapple with how we should, uh, act with respect to kind of infinite worlds. (screen swooshes) There's a middle ground between "I shall ignore this completely" and "I shall, you know, be a Jane," um, which is recognizing that this is a, this is a real trade-off, there's uncertainty here, and, and taking responsibility for how you're responding to that. (screen swooshes) The future is a big thing to try to model with this tiny mind, and so, you know, o- of necessity, you need to use these extremely lossy abstractions. (cheerful music)

2. 0:55 – 3:42

   Introduction

   1. DPDwarkesh Patel0:55

      Today, I have the pleasure of interviewing Joe Carlsmith, who's a senior research a- analyst at Open Fo- uh, Philanthropy and a doctoral student in philosophy at the University of Oxford. Um, it, uh, j- uh, Joe has a really interesting blog that I got to check out, uh, called Hands and Cities, um, and that's the reason that I wanted to have him on the podcast, 'cause it has a bunch of thought-provoking and insightful, uh, uh, posts on there about philosophy, morality, ethics, uh, the future. And yeah, so I, I really wanted to talk to you, Joe, uh, but are you... Do, do you wanna give a, a bit of a longer intro on what you're up to?

   2. JCJoseph (Joe) Carlsmith1:31

      Sure. So I work at Open Philanthropy on existential risk from artificial intelligence, um, and so, you know, I think about what's gonna happen with AI, how can we make sure it goes well, and in particular, how can we make sure that advanced AI systems are safe? Uh, and then, uh, I have a side project, which is this blog, uh, where I write about philosophy and, uh, and the future and things like that. And that emerges partly from the, um, a sort of, my background, which is, um, I was, I was... Before, before getting into, uh, into AI and working at Open, Open Philanthropy, I was in, uh, academic philosophy.

   3. DPDwarkesh Patel2:07

      Okay, yeah. That, that's, uh, that, that, that's a quite, quite an ambitious side project. I mean, g- given the length and the regularity of those posts, it's, it's actually quite stunning. Um, do, do you want to talk more about what you're working on a- about AI at Op- uh, at, uh, Open Philanthropy?

   4. JCJoseph (Joe) Carlsmith2:22

      So it's a mix of things. Right now, I'm thinking about AI timelines and what's called takeoff speeds, sort of, sort of how fast the transition is from pretty impressive AI systems to AI systems that are, uh, kind of radically transformative, um, and I'm trying to use that, uh, to provide more perspective on the probability that, um, that everything goes terribly wrong.

   5. DPDwarkesh Patel2:41

      I see. Okay. Um, I, I didn't really say, but what, what are the implications? I suppose it's, uh, higher or lower than I would expect? Um, I guess if it's higher, maybe I should work on AI 11, but other than that, what is, wha- wha- what are the implications of that, that, that figure changing?

   6. JCJoseph (Joe) Carlsmith2:55

      I think there are a number of implications just from understanding, uh, timelines with respect to how you prioritize and what, um... You know, just to some extent, the sooner something is, then, uh, you need to be planning for it coming sooner and, and kind of cutting more corners or just, you know, um, counting less on having more time. Um, and yeah, I think overall the higher you think, uh, the probability of catastrophe is, the, um, the easier it is for this to, uh, to become kind of the most important priority. Uh, I do think there is a range of probabilities where it maybe doesn't matter that much, um, but I think, uh, the difference between, say, uh, 1 and 10%, I think, is, uh, is quite substantive, and, um, (laughs) the difference between 10 and 90 is quite substantive (laughs) , and, and, um, I, you know, I know people in all of those ranges.

   7. DPDwarkesh Patel3:40

      Gotcha. Okay, interesting.

3. 3:42 – 10:08

   How to define a better future?

   1. DPDwarkesh Patel3:42

      Um, yeah, so let, let, let's, let's back up here and, uh, talk a bit more about the philosophy motivating this. So, I think you identify as, uh, a long-termist. Um, yeah, so may, may, maybe a broad picture question here is, um... You have an interesting blog post about what the future, looking back on us, might think about, uh, the 21st century given the risk we're taking. Um, uh, so, wh- wha- what do you think about the possibility that we are potentially giving up resources, potentially dedicating... Well, I'm not, uh, but you're dedicating your career, um, to, uh, you know, buil- building a future that, you know, maybe given, you know, gi- given the fact that you're alive now, you might find strange or disturbing or disgusting? I mean, um, uh, so I guess to add more context to the question, if from a utilitarian perspective, the present is clearly much, much better than the past, but somebody from the past might think that, you know, uh, there, there's a lot of bad things about the present that are kind of disturbing. I mean, they, they might not like the configuration of how maybe isolating a, a, a modern city might be. They might find the kinds of free to cheap information that you can access on your phone, uh, ki- kind of disturbing. Yeah, so how, how do you think about that?

   2. JCJoseph (Joe) Carlsmith4:49

      So yeah, a few comments there. So one, um, I do think that if you took... You know, for most people throughout history, if you brought them to the present day, uh, they would... My, my guess is that fairly quickly and depending on exactly the circumstances, they would, um, come to prefer living, uh, living in the present day to, to the past even if there are sort of a bit of, uh, future shock and, and a bit of, um, uh... Some things are alienating or disturbing. Um, and, but that said, I think the distance, the sort of gap between historical humans and the present is actually much, much smaller, um, both in terms of time and kind of, um, other factors than the gap I envision between present day humans and the future humans who are living, living ide- ideally in, in, um, a kind of radically better situation. Um, and so I do expect s- sort of greater distance and possibly greater alienation when you first show up. My personal view is that, uh, the best, the best futures are, um, uh, going to be such that if you really understood them and if you really experienced what they're like, which, which may be, um, a big step and might require sort of extensive, uh, engagement and possibly sort of, um, changes to your capacities to understand and experience, then you would think it's really good. Um, and so, uh... And, and I think that's the relevant standard. So for me, I, I worry less if the future is sort of initially alienating, um, and the question for me is, how do I feel once I've really, um, really understood what's going on?

   3. DPDwarkesh Patel6:17

      I see. Um, so I- I- I- I wonder how much we should value that kind of inside view you would get into the future from being there. If you think about, um, uh, I don't know, many- many existing ideologies, uh, like I don't know, think of an Islamist or something who- who might say, "Listen, if- if you could just, like, come to Iraq and feel the bliss of- of- of fighting for the caliphate, uh, you- you would understand better than you can understand from the outside view, just, you know, sitting on a couch eating Doritos, wha- what, you know, what it's like to fight for a cause." And may- maybe that their experience is kind of blissful in some kind of way, but, um, uh, I- I- I feel like the outside view is more useful than the inside view there.

   4. JCJoseph (Joe) Carlsmith6:54

      Well, so I think there's a couple different questions there. One is, what would the experience be if you had it from the inside? Um, and then there was this, I think a subtly different question, which was what, which is what would your take on this be if you kind of fully understood? Where fully understanding is not just, um, a matter of having the internal experience of being in, you know, in a certain situation, but it's also a matter of understanding what that situation is causing, what sort of beliefs are structuring, um, the ideology, whether those beliefs are true, um, and all sorts of other factors. And it's the latter thing that I have in mind. So I'm not just imagining, "Oh, the future will feel good if you're there," um, uh, because, you know, sort of by hypothesis, the people who are there at least (laughs) one hopes they're enjoying it or one hopes they're thumbs up. If- if the people who are there aren't thumbs up, that's a strange- a strange utopia. Um, but I'm thinking more that in addition to, uh, their perspective, there's a sort of more holistic perspective, which is the sort of full understanding, uh, and that's the perspective from which you would endorse, uh, endorse this s- situation.

   5. DPDwarkesh Patel7:56

      I see. Um, and then, yeah, so ano- another respect in which, uh, it's interesting to think about what they might think of us is, you know, like, well, what will they think of the crazy risk we're taking, um, by not- not optimizing for existential risks? And, um, I... So, you- you know, one analogy you could offer, I think, Will MacAskill does this in his new book, is to think of us as, uh, you know, teenagers in our civilization's history, and then, you know, think of the crazy things you did as a teenager and how y- y- um... And, yeah, so, uh, I mean, maybe there is an aspect of which, like, one would wish they could take back the crazy things they did as a teenager. But my impression is that most adults probably think that while the crazy things were, um, kind of risky, um, they were, they're very formative and important, and, um, they feel nostal- nostalgic about the things they did in the past. Do- do you think that the- the future, looking back, they are going to, um, regret the- the way we're living in the 21st century, or, uh, or will they look back and think, "Oh, you know, that- that was kind of a cool time"? I mean, I guess this is kind of conditional on there being a future, which takes away a lot of the mystery here, but...

   6. JCJoseph (Joe) Carlsmith8:59

      I doubt that they will look back with, um, uh, with pleasure at, uh, the sort of risks and, uh, and horrors of the- of the 21st century. I mean, if you just think about how, uh, we, or at least I, tend to think about, uh, something like the Cuban Missile Crisis or, uh, World War II-

   7. DPDwarkesh Patel9:19

      (laughs)

   8. JCJoseph (Joe) Carlsmith9:19

      ... I don't personally have a kind of nostalgia, "Oh, you know, sure it was risky, but it made- it made me who I am," or something like that. Um, I also want to say, you know, I think it's true that when you look back on your teenage years, there is often a sort of, you know, let's say you w- you did something like crazy, you and your friends used to race, you know, around, and you'd play chicken or something at the local quarry, uh, and it's like, "Oh, ra-ra-ra-ra." You, but, you know, you survived, right? And- and the real reason not to do that is the, like, chunk of probability where you just died. Um, and so I think there's a- a, you know, to some extent, the... it's- it's, um, the ex post perspective of looking back on certain sorts of risks is not the right one, uh, for, especially for death risks. That's not the right, uh, perspective to use to kind of calibrate your understanding of how to feel about it overall.

   9. DPDwarkesh Patel10:06

      I see. Um,

4. 10:08 – 26:01

   Utopia

   1. DPDwarkesh Patel10:08

      okay, so I think you brought up utopia, and you have a really interesting post about, uh, the concept of utopia. Uh, so yeah, do you wanna talk a little bit more about this concept and why it's important? And, um, and also, why- why do we have so much trouble thinking of a compelling utopia?

   2. JCJoseph (Joe) Carlsmith10:24

      Yeah, so utopia for me just means a kind of profoundly better future. And I think it's important because I think it's just actually possible. I just think it's actually something that we could do. Uh, we- we could make... I- if we sort of play our cards right, in- in sort of non-crazy ways, we could just build a world that is radically better than the world we live in today. Um, and in particular, I think, uh, we often, in thinking about, um, that sort of possibility, underestimate just how big the difference in value could be between, um, our current situation and- and kind of, um, what's available. Uh, and I think often utopias are kind of anchored too hard on the status quo and sort of changing it in- in small ways, but imagine, imagining our kind of fundamental situation basically unaltered. Um, and I think... So such that it's a little bit like the difference between, you know, you have a kind of a crappy job or, like, a beach vacation. And utopia is like everyone has beach vacation. Uh, and (laughs) you know, I don't know how you feel about beach vacations, um, but I think it's much... I think the difference is more like being asleep and being awake, uh, or sort of a, uh, it's- it's more, um, uh... Yeah, it's sort of, it's like living in a cave or living- living in the, in, under the open sky. I think, I think it's like a really big, a really big difference, and, um, and that that matters a lot.

   3. DPDwarkesh Patel11:48

      I... That's interesting because I remember in the essay you had, um, you had a section where, um, uh, uh, you mentioned that you expect utopia to be recognizable, uh, like, to- to a person alive now. Um, the- the... I guess the way you put it just earlier made it seem like it would be a completely different category of experience than we would be familiar with. Um, yeah, so did... Is there a contradiction there, or maybe I'm missing something?

   4. JCJoseph (Joe) Carlsmith12:14

      So I think there's at least a tension. And- and the way I see the tension...... uh, playing out or, you know, kind of being reconciled, but specifically, uh, via the notion I referenced earlier of kind of you would, if you truly understood, come to see, uh, the utopia as genuinely good. But I think that process... I mean, ideally, I think the way we end up building utopia is we go through a long, um, patient process of becoming wiser and better and more capable as a species, um, and, and it's in virtue of that process kind of culminating that we are in a position to build, um, to build a civilization that is sort of profoundly good and radically, radically different. Um, but that's a long process. And so I, I do think, you know, if... Yeah, as I say, if I just transported you right there and you skipped, you skipped the process, then you might not like it. Um, but, uh... And, and, and it is quite alien in some sense, but I still... But if you went through the process of, like, really understanding and kind of becoming wiser, um, uh, you would, you would endorse.

   5. DPDwarkesh Patel13:15

      Uh-huh. That's, um, that's interesting to me that you think, uh, the process to get to utopia is more of a sort of a- a- a- maybe I'm, uh, misconstruing it, but when you mentioned wha- it's a process of us getting wiser. And, um, um, yeah, so it sounds like it's a more philosophical process, rather than, I don't know, this, we, we figure out how to con- convert everything to hedonium and, you know, it's eternal bliss from then on. Uh, yeah, so, uh, uh, am I getting it right that you think it's more a philosophical process, and then wh- why is it that you think so?

   6. JCJoseph (Joe) Carlsmith13:44

      Yeah, so I definitely don't sit around thinking that utopia... we sort of know what utopia is right now, and it's hedonium. Um, I'm not especially into the notion of hedonium, but I think it's possible to, um (laughs) ... I think it's, I think the brand is bad. Um, I think, I think, you know, people, uh, talk about pleasure with this kind of dismissive attitude sometimes, and, you know, hedonium implies this kind of sterile, um, uniformity, uh, y- y- you know, and you're sort of tiling... People are talking about they're, like, gonna tile the universe with hedonium, and it's like, "Wow, this sounds, this sounds rough." Um, whereas I think actually, you know, the relevant perspective when you're thinking about something like hedonium is the kind of internal perspective from which, uh, the sort of experience of the subject is something kind of, uh, joyful and, you know, boundless and kind of, uh, energizing, you know, whatever, whatever pleasure is actually like. Pleasure is not a trivial thing. I think pleasure is a, a profound thing in a lot of ways. But I really don't... I don't assume that that's what utopia is about at all. I think we're at... I think, A, my, you know, my own values seem to be quite complicated. I don't think I just value pleasure. I value a lot of different things. Um, and more broadly, I have a lot of uncertainty about how, how I will think and feel about things if I were to go through a kind of process of significantly, uh, increasing my capacity to understand. Um, I don't... I think sometimes when people imagine that, they imagine, oh, we're gonna sit around and do a bunch of philosophy, and then we'll have, like, solved normative ethics, and then we'll implement our solution to normative ethics. Um, and that's not what I'm imagining by, uh, kind of wisdom. I'm imagining something, um, richer and also that involves, uh, importantly a kind of enhancement to our, uh, cognitive capacity, so sort of really... Uh, you know, I think we have, we have very small... We- we're really limited in our ability to understand the universe right now. We have kind of, um... And I think there's just a huge amount of uncharted territory in terms of what minds can b- and do and see, and so I wanna sort of chart that territory before we start, uh, making kind of big and irreversible decisions about what sort of civilization we wanna build in the long term.

   7. DPDwarkesh Patel15:47

      I see. Um, and then I... Uh, another, uh, maybe concerning part of, uh, u- the utopia is that, um, yeah, as you mentioned in the piece, many m- many of the worst ideologies in history have had, uh, elements of utopian thinking in them. Um, to the extent that EA and utilitarianism generally are compatible with utopian thinking, maybe they don't, uh, advocate utopian thinking but they are compatible with it. Um, d- do you see that as a problem for, uh, the, the movement's health and potential impact?

   8. JCJoseph (Joe) Carlsmith16:16

      Is the question something like, uh, is this a red flag? (laughs) Kind of, ah, you know, we, we look at, we look at other ideologies throughout history, and they've been, uh, compatible with utopian thinking, um, and, and maybe sort of, um, effective altruism or, or, or, uh, utilitarianism is something that's similarly compatible, so should we, should we worry in the same way? Is that the question?

   9. DPDwarkesh Patel16:35

      Uh, yeah, partly, and also, um, another part is, um, may- maybe the... May- maybe, maybe it's still right, uh, that, like, morally speaking, yeah, uh, utopia is compatible with this worldview and the worldview is correct. Uh, but that, that, that the implications, uh, are that, you know, some- somebody misunderstands what is best, um, they identify as an EA and this leads to bad consequences when they try to implement their scheme.

   10. JCJoseph (Joe) Carlsmith17:00

       Yeah, so I think there are certainly reasons to be cautious, uh, in this broad vein. Um, I don't see them as very specific to EA or utilitar- I, I don't identify as a utilitarian, but, um, the... to utilitarianism. Um, I see them as more... or sort of better understood as, uh, risks that come from believing that something is very important at all (laughs) . Um, and I think it's true that many, um, acting from a space of, of conviction, um, especially where, uh, that conviction has, has sort of a flavor of... You know, it's, it's interesting what exactly constitutes an ideology, but I think it's, I think it's reasonable to look at, at EA and sort of be like, "This has... This looks like an ideology." And I think, you know, and I think, um, that's... I think that's right, and I think, uh, that's sort of important to, to, you know, have the sort of relevant red flags about. Um, I think it's pretty hard to have a view of the world that doesn't in some sense imply that it could be a lot better, um, or at least a plausible view of the world. And, and when I say utopia, I don't really mean anything much different from that. You know, I think it's sort of... Um, I'm not saying a perfect thing. I'm not... Y- you know, I- I do have sort of a more specific view about exactly how much better things could be. But more broadly, it seems to me many, many people believe in the possibility of a much better world and are fighting for that in different ways. Um, and, uh...So, I wouldn't, I wouldn't pin the red flag specifically to the belief that sort of things can be better. Um, I think it would have more to do with, uh, sort of what degree of rigidness are you, um, you know, relating to that belief with how are you, uh, kind of how are you acting on it in the world? How much are you willing to kind of, um, kind of break things or kind of act in uncooperative ways in virtue of, of that sort of conviction? And there, I think, um, uh, caution is definitely warranted.

   11. DPDwarkesh Patel18:52

       I see. Yeah. So I, I, I'm, I'm, uh, uh, I'm not sure I agree that, um, most people have a view, uh, or an ideology that implies, um, uh, anywhere close to the kind of utopia that, uh, wanting a utopian thinking one can have. Like, if you think of modern political parties in a developed, uh, democracy, um, like in the United States, for example, if you think of, uh, what is like a utopian vision that either party has, it's like, it's actually quite, uh, quite banal. It's like, "Oh, we'll have universal healthcare," or, I don't know, "GDP will be higher in the next, uh, couple of decades," um, which is, uh, which doesn't seem utopian to me. It just seems... And it does seem, um, it, it does seem like a limited worldview where they're not really thinking about how much better or worse things could be, but it doesn't exactly seem utopian. Uh, yeah, yeah. I'll, I'll let you react to that.

   12. JCJoseph (Joe) Carlsmith19:42

       I think that's a good point. So maybe the relevant notion of utopian here is something like to what extent is a concept of a radically better world kind of operative in your day-to-day, uh, engagement? Uh, you know, to some extent, what I meant is that I think, I think if I sat down and talked with most, uh, you know, most people, um, you know, we could eventually (laughs) with some kind of constraints on reasonableness come to agree that things could be a lot better in the world. Like, we could just cure cancer. We could cure, (laughs) you know, X, Y, Z disease. We could just go through a few things like that. We could talk about, um, the degree of abundance that could be available. Um, and I think, uh, you know, so, but the question is whether that's like a kind of structuring or important dimension to how people are relating to the world, and I think you're right that it's often not. And that's part of maybe, um, the thing I'm hoping to, uh, kind of push back against with that post is actually I think this is a really important feature of our situation. Um, I think it's true that it's, it can be dangerous, and if you're wrong about it or if you're acting, um, in the right, uh, in a sort of, um, unwise way with respect to it, that can be really bad. But I also think it's just, it's just a really basic fact, and I think we just sort of need to learn to deal with it maturely, and kind of pretending it's not true I think isn't the way to do that.

   13. DPDwarkesh Patel20:58

       I see. Uh, uh, but, uh, to me at least, utopian or utopia sounds like a, some sort of peak. Um, and maybe you didn't mean it this way, but, uh, so are, are you saying in the essay and generally that you think there is some sort of carrying capacity to how much good things can get or that beyond a certain, certain point things can keep getting in, um, indefinitely better, uh, but at this point we're willing to say that we have reached utopia?

   14. JCJoseph (Joe) Carlsmith21:28

       Yeah. So I mean, I certainly don't have a kind of hard threshold, ah, you know, here's, here's exactly where, where I'm gonna call it utopia. Um, you know, I mean something that is profoundly better. Uh, I do think that if you have a finite... So, you know, very basic level. If there's only a finite number of states that, uh, the sort of affectable universe can be in, um, and your, your ranking of these states in terms of how good they are is, uh, transitive and complete, um, then there will be a sort of top, (laughs) um, a top. And, and, uh, you know, I don't think that's an important thing to focus on from the perspective of just getting it, just, you know, taking seriously that things could be radically better at all. I think like talking about, ah, but exactly how good and what's the perfect thing is, is often kind of, um, distracting in that respect. And it gets into these issues about like, oh, you know, um, how much suffering is good to have. And, and, and a lot of this sort of discourse on utopia I think gets distracted from basic facts about like, at the very least, we can do just a ton better. Um, and that's important to keep in mind.

   15. DPDwarkesh Patel22:34

       I see. I see. You, you point out in the piece that many religions and spiritual movements have done the most amount of thinking on what a utopia could look like. And, you know, there, there, there's a very interesting, um, essay by Nick Bostrom in 2008 where he lays out his vision of what somebody speaking from the future utopia ta- talking back to us would sound like. And when you read it, it sounds very much like a sort of mystical, uh, mystical essay, the kind of thing that, uh, change a few words and a Christian could write, like C.S. Lewis could have written about like what it's like to speak down from heaven. Um, yeah. So, so to what, what extent is there, uh... And I don't, I don't mean this pejoratively, but, uh, to, uh, what, to what extent is there some sort of like, uh, spiritual or religious dimension to utopian thinking, uh, th- that relies on some amount of faith that things can get in sort of, uh, indescribably better in some sort of ephemeral indescribable way?

   16. JCJoseph (Joe) Carlsmith23:27

       So I think there are definitely analogs and similarities between some ways of relating to the notion of utopia and, uh, attitudes and orientations that are common in religious contexts and spiritual contexts. And I think it's, um... And I think personally, uh, I, so I don't think it needs to be that like that. I, as I say, I think, I don't think it requires faith. I don't think it requires anything mystical. Um, I don't think... I think this is, it's just a basic fact, um, about our kind of current, uh, you know, our current cognitive situation, our current civilizational situation that, um, things could be radically better. Um, and, uh, it's a f- you know, it's ephemeral in the sense that it's quite hard to imagine, especially, you know, for me an important, an important source of evidence here is, is sort of variance in the quality of human experiences. So if you think about your kind of peak experiences, um, they're often... It's, it's a really big deal. You're kind of, you're kind of sitting there going, "Wow, this is radically... This is serious." Um, and, uh, kind of feeling in touch or, or feeling that this is, this is, uh, in some sense a, a r- um...... something you would trade m- much, much sort of mundane experience for the sake of. Um, and I think it's important... So the thing that I think we need to do is sort of extrapolate from there. So you sort of look at the trajectory that your mind moved along as you, as you moved into some experience or some broader, non-experiential. Like, your community got a lot better, your relationships got abou- a lot better. Look at that trajectory, and then sort of stare down, you know, where is that going. Um, and I do think that requires a kind of, I don't wanna call it faith. I think it requires a kind of, um, extrapolation into a sort of zone that is in some sense beyond your experience, but that is sort of deeply worthy and important. And I think that's, um, something that, uh, is often associated with, with spirituality, um, and religion. And, and I think, uh, I think that's okay. Um, but I, I actually think there's a, there are a number of really important differences between utopia and something like Heaven. Um, so, uh, you know, centrally, utopia will be a sort of concrete limited situation. There w- you know, there are gonna be frictions. There are gonna be resource constraints. Uh, it's going to be finite. Um, th- there is, there is a bunch of... It, it's still going to be in the real world. Whereas, I think, um, uh, many, you know, most religious visions have, don't ha- (laughs) don't have those constraints, and that's an important, a important feature of their, um, uh, uh, yeah, of their, their situation.

5. 26:01 – 28:24

   Robin Hanson’s EMs

   1. JCJoseph (Joe) Carlsmith26:01

   2. DPDwarkesh Patel26:01

      Yeah. Speaking of constrain- uh, constraints, this reminds me of Robin Hanson's theory that, uh, uh, you know, eventually the universal economy will just be made up of, um, these digital people M's. And that because of competition their wages will be driven down to subsistence levels, uh, which, um, may- maybe that's compatible with some engineering in their ability to experience such that, you know, it, it's still blissful for them to work at subsistence levels of compute or whatever. Um, but, uh, yeah, so th- it seems like this sort of, like, uh, first order of economic thinking implies that there will be no, there will be no utopia. In fact, things will get, um, things will get worse for on average, but may- maybe better o- uh, overall if you just add up all the experience, but worse on average. Uh, yeah, so, so I don't know. Uh, m- this vision seems incompatible with yours of a utopia. What, what do you think?

   3. JCJoseph (Joe) Carlsmith26:54

      Yeah. I would not call, uh, Robin's world a utopia. (laughs) Uh, and so, you know, a, a thing I haven't been talking about is, what should our overall probability distribution be with respect to different quality of futures? Um, and what, um, you know, exactly how possible is it, uh, and how likely is it that we, we build something that is sort of profoundly good as opposed to, uh, mediocre or much worse? Um, and, uh, I would class Robin's scenario (laughs) in the, uh, mediocre or, uh, or much worse zone.

   4. DPDwarkesh Patel27:28

      But, but, but so do you have a, a criticism of the logic he uses to derive that?

   5. JCJoseph (Joe) Carlsmith27:33

      To some extent, I think my main, my main criticism or the first thing that, that would come to mind is that I think we will very likely, um, uh... Like, I think competitive pressures are, uh, are a source of kind of, kind of pushing, pushing, uh, the world in, in bad directions. But I also think there are ways in which, um, kind of wise forms of coordination and kind of preemptive action can, uh, can s- stave off the sort of bad effects of, of competitive pressures. And I... And so that's, that's a sort of, um, that's the way I imagine avoiding, uh, stuff in the vicinity of, of what Robin was talking about. The, you know, there, there are a lot of complexities there.

   6. DPDwarkesh Patel28:13

      Yeah. Yeah, yeah. Um, the last few years have not reinforced my, uh, uh, my, my belief in the possibility of wise coordination. But, uh, yeah, yeah. Uh, anyways, so, um, yeah. One thing I wanted to talk to you about is

6. 28:24 – 35:04

   Human Computational Capacity

   1. DPDwarkesh Patel28:24

      you have a, a paper on what, what it would take to match, uh, humans' brains computational capacity. Um, uh, and then also shared with that you have, uh, v- you know, a, a very good summary on open philanthropy. Um, yeah. So do, do, do you wanna talk about, uh, the approach you took to estimate this, and then wh- why this is a important metric to try to figure out?

   2. JCJoseph (Joe) Carlsmith28:46

      Yeah. So, um, the approach I took was to look at the evidence from neuroscience and the literature on, uh, the kind of computational capacity of the human brain and to talk to a bunch of neuroscientists, and to try to, you know, see, see what we know right now about, uh, the, uh, the number of floating point operations per second, uh, that would be sufficient to kind of reproduce the task-relevant, uh, aspects of human cognition in a computer. Um, and that's important... I, I mean, it's actually not, uh... You know, it's not clear to me exactly how important this parameter is to our overall picture. Um, I think the way in which it, it's, uh, relevant to thinking that I've been doing and that Open Phil has been doing is, um, as an input into an overall methodology for estimating when we might see, uh, kind of human level AI systems that proceeds by first trying to estimate roughly the, the kind of computational capacity of the brain, or the, or the sort of, um, uh, th- the sort of size of, size of, uh, uh, kind of AI system. And it's, it's kind of overall parameter count, uh, and, uh, kind of compute capacity. And that would be sort of analogous to humans. And then you extrapolate from that to the training cost, the cost to kind of create a system, um, of that kind using, uh, current methods in machine learning and kind of current scaling, scaling laws. Uh, and, um, that methodology though brings in a number of additional assumptions that I think aren't, um, aren't like just transparent that that's... Oh yeah, of course that's how we would do it or that... And so, um, I think you have to sort of be a little bit more in the weeds to see exactly how it, um, how it feeds in.

   3. DPDwarkesh Patel30:22

      I see. And then, yeah. So what, what, what... I, I think you said it was 10 to the 15 flops, uh, for, um, a, for a human brain. But like, what, what... Did you have an estimate for how many flops it would take to train, uh, to train something like the human brain? I know GPT-3 is like, um, on- only 175 billion parameters or something, which is... Can fit into a, you know, like a micro SD card even. Um, but, uh, but yeah. It was like, oh, $20 million to train. So, um, yeah. So w- what... Do you have... Were you able to come up with some sort of estimate for how you... what it would cost to train something like this?

   4. JCJoseph (Joe) Carlsmith30:55

      Yeah. So my focus in that report was not on the training extrapolation. That was work, uh, that Ajeya Cotra at Open Philanthropy did, um, using my reports estimate as an input. And, uh, that ... Her methodology involves assigning different probabilities to different kind of ways of using that, that input, uh, to, to, to drive an overall training estimate. Um, and in particular, an important source of uncertainty there is, uh, the kind of amount of compute required or the sort of number of times you need to run a system per data point that it gets. So in the case of something like GPT-3, you get a meaningful data point and a gradient update as to how well you're performing, um, with each token that you output as you're doing GPT-3 style training. So you're, you know, you're predicting text from the internet. You know, you, you suggest the next token and then your training process says like, "Nope, do better next time," or something like that. Whereas if you're, uh, say, learning to play Go and you have to play, uh ... I mean, this isn't exactly how or this isn't how a Go system would work, but it's, it's an example. If you have to play the full game out and that's sort of hundreds of moves, um, then before you get a, an update as to whether, uh, you know, you're playing well or poorly, then, uh, that's a big multiplier on, on the compute requirement. And so that's, that's one of the central pieces. That's called, uh, what Ajeya calls the horizon length of, of training. And, um, that's a sort of very important, uh, source of uncertainty in getting to your overall, overall, uh, training estimate, I think. But ultimately, you know, she ends up with this big spread out distribution from something like ... I think GPT-3 was like, um, 10 to the 24. Yeah, four times 10 to the 23 or something like that, and you know, she's, she spreads out all the way up to the, the evolution anchor, I think is something like 10 to the 41, and, uh, I think her distribution is centered somewhere in the low 30s.

   5. DPDwarkesh Patel32:43

      Okay. That's, that's still quite a bit, I guess. Um, how much does this rely on, you know, the scaling hypothesis? If, if one thought that the current efforts and the current approach were not, um, not likely to lead in a, or at least not likely in a sample efficient way towards, uh, towards human intelligence, you know? It might be analogous to somebody saying we have, um, enough deuterium on Earth to power civilization for millions of years. Um, but if we haven't figured out fusion, then it, it may be irrelevant, uh, statistic.

   6. JCJoseph (Joe) Carlsmith33:12

      Yeah. So I think the approach does assume that you can train a human level or sort of transformative AI system with a sort of non-astronomical amount of compute and data using current ... You know, without, without major conceptual or algorithmic breakthroughs relative to, to what's currently available. Um, now, the, the actual methodology Ajeya uses allows you to assign probabilities to that assumption too, so you can, if you want, you know, say, "I'm only 20% on that." Um, and then, uh, you have ... Then there are sort of other ... Uh, there are a few other options, so you can also kind of rerun evolution, which is not, uh ... And, and, and so that's, that's an anchor that she provides to sort of, uh ... And this is often what people will say as a sort of upper bound on how hard it is to create, um, to create human level systems is, is to do something, something analogous to, to, um, to simulating evolution, um, though that, you know, there are a lot of open questions as to how, how hard that is. Um, but I do think this methodology, uh, is a lot more compelling and interesting if you, um, are compelled by the, uh, the kind of available techniques in deep learning and, and by, and by kind of scaling hypothesis-like views, at least in, as an upper bound. I think it's important. So, you know, there's different ways of, of kind of being interested in algorithmic breakthroughs. One is because you think deep learning isn't enough. Another is because you think they will sort of provide a lot of efficiency relative to deep learning, such that an estimate like Ajeya's is an overestimate, because actually, you know, we won't have to do that. We'll make some sort of breakthrough and it'll happen a lot earlier. Um, and, uh, uh, and I put, I put weight on that view as well.

   7. DPDwarkesh Patel34:52

      Yeah. That's really interesting. So yeah, that implies that like e- even if you think the current techniques are not, uh, not optimal, may- maybe that, maybe that should update you in favor of thinking it could happen sooner. That's, that's really interesting. Um, uh,

7. 35:04 – 41:04

   FLOPS to emulate human cognition?

   1. DPDwarkesh Patel35:04

      um, yeah. So yeah, then how, how did you go about estimating, uh, like, uh, the amount of flops it would take to emulate, uh, the interactions that happen in the brain? Uh, obviously, uh, it would be unreasonable to say that you have to emulate every single atomic, uh, atomic interaction. Um, but then what is, what is your proxy that you think it would be sufficient to emulate?

   2. JCJoseph (Joe) Carlsmith35:28

      So I used a few different methodologies and tried to kind of synthesize them. So one was looking at the kind of mechanisms of the brain and what we know about, uh, the kind of complexity of what they're doing and how hard it is to capture the kind of task relevant or our best, our best guess about the task relevant dimensions of the, the signaling happening in the brain. Um, and then I also tried to bring in comparisons with, uh, existing AI systems that are replicating kind of chunks of functionality, um, that humans, uh, that the human brain has, and in particular, in the context of vision. Um, so sort of, uh, how do our, how do our current, um, vision systems compare with, uh, the parts of the brain that, uh, are kind of plausibly doing analogous processing, though they're often, they're often doing other things as well. Um, and then I used a third method, which is, has to do with physical limits on the kind of energy consumption per unit computation, um, that the brain is plausibly doing. And then a fourth method I sort of gestured at, which tries to extrapolate from, uh, the communication capacity of the brain, uh, to its computational capacity using comparisons with, uh, with current computers. So it's sort of a triangulation of like a ... You look at a bunch of different sources of evidence, all of which, in my opinion, are pretty weak. I think we are, um, uh, we are quite ... Well, the, the physical limit stuff is, is maybe more complicated, but it's sort of, uh, uh, upper bound. Um, I think we are significantly uncertain about all of this, and, and I ... my distribution is, is pretty spread out. Um, but, uh, the hope is that by looking at a bunch of things at once, you can at least get, um, a sort of educated guess.

   3. DPDwarkesh Patel36:57

      And then, yeah, so I'm very curious, um, uh, is there a consensus in neuroscience or other relevant fields that we understand the signaling mechanisms well enough that we can say like, "Basically this is what is involved. Um, this is what the system is reducible to, um, and yeah, so this is how many bits you need to represent, uh, out of all the synaptic connections here." Or, I, is there a variance of opinion about like just how complicated the enterprise is?

   4. JCJoseph (Joe) Carlsmith37:24

      Uh, there's definitely disagreement. And, um, it was, you know, interesting and in some sense disheartening to talk with neuroscientists about just how, uh, (laughs) you know, how difficult neuroscience is, you know? It's sort of, I think it's easy... A consistent message, and I have a section on this in the report, um, was kind of how far we are from really understanding, uh, what's going on in the brain, um, especially at a kind of algorithmic level. Um, that said, so in some sense the report is somewhat opinionated in that, um, you know, there are experts that I found more compelling than others. Uh, there are experts who are much more in a sort of agnosticism mode of, like, "We just don't know." Um, you know, "The brain is really, really complicated," who sort of err on the side of, uh, very large compute estimates, a lot of emphasis on biophysical detail, a lot of emphasis on sort of mysterious things that could be happening that aren't happening. And then there are other neuroscientists who are more, uh, uh, you know, more willing to say stuff like, "Well, we kind of basically know what's going on at a mechanistic level," which isn't the same as knowing kind of the algorithm, the sort of algorithmic organization overall and how to replicate it. I sort of lean towards the latter view, though I give weight to both and, and try to, um, yeah, try to synthesize the kind of opinions of people I, I, I saw overall.

   5. DPDwarkesh Patel38:41

      Just looking at the post itself, I haven't really looked deeper into the actual, um, the, the, uh, paper from which it's derived. Uh, but it seems like you were to estimate the flops mechanistically, you were adding up the different systems at play here. Um, should we expect it to be additive in that way or maybe it's like multiplicative or there's more complicated interaction, like the flops grow super linearly to the inputs? Uh, I know that probably sounds really naive having studied it, but just like from a, uh, first glance kind of way, that's a question I had.

   6. JCJoseph (Joe) Carlsmith39:15

      Yeah, so the way I was understanding, um, and breaking down the forms of processing that you, you would need to replicate in the brain, um, made, uh, made them seem not multiplicative in this way. So, uh, you know, an example would be if you think about sort of simple examples. Suppose we have some neurons and they're, uh, they're signaling centrally by a spike through synapses or something like that. And then we have, uh, glial cells as well, which are signaling via like slower calcium waves and it's a sort of separate, uh, separate network. Um, you know, you could think that if it were something like, you know, the rate of calcium signaling is, um, uh, dependent on the rate of spikes through synapses or something like that, then that's an important interaction. Uh, but, you know, overall if you sort of imagine this kind of network processing, um, uh, these are kind of ind- you can just, you can estimate them independently and then add it up. They're not actually multiplicative processes on that conception. Um, I do think there are kind of correlations between the estimates for, for, um, the different parts. But I, uh, it's sort of additive at a fundamental level.

   7. DPDwarkesh Patel40:25

      I see, okay. And then, yeah, how much credence do you put in, um, the sort of almost woo-woo hypotheses that, I don't know, Roger Penrose has that thing about there's something like, uh, something quantum mechanical happening in the brain and that's very important in, uh, for understanding cognition. Um, yeah, to what extent, uh, do you put credence in those kinds of hypotheses?

   8. JCJoseph (Joe) Carlsmith40:47

      I put very little credence in those hypotheses.

   9. DPDwarkesh Patel40:50

      (laughs)

   10. JCJoseph (Joe) Carlsmith40:50

       Um, uh, yeah, I don't see a lot of reason to think that. Um, I see a good amount of reason not to think it. Um, but it wasn't something I dug in on a ton.

   11. DPDwarkesh Patel41:03

       Okay, gotcha.

8. 41:04 – 1:01:40

   Infinite Ethics

   1. DPDwarkesh Patel41:04

      All right, so you have this really interesting blog post about infinite ethics. Um, do you want to talk about why this is an important topic, why it's important to integrate into our worldview and so on?

   2. JCJoseph (Joe) Carlsmith41:14

      Sure. So infinite ethics is ethics that tries to grapple with how we should, uh, act with respect to kind of infinite worlds. Um, and how should we, you know, how should we rank them? Um, how should they enter into our, uh, our expected utility calculations or our attitudes towards risk? Um, and I think this is important for both kind of theoretical and practical reasons. So I think at a theoretical level when you, when you try to do this with a lot of common, um, ethical theories and constraints and principles, um, they just break on, uh, on infinite worlds. Um, and, uh, I think that's an important clue as to their viability because I think infinite worlds are at the very least possible. Um, even if our world is finite, um, even if our causal influence is finite or our influence overall is finite, um, it's possible, uh, to have infinite worlds and we have opinions about them. You know, like an infinite heaven is better than an infinite hell. And, you know, uh, so I think, um, often in ethics we have, we expect our ethical principles to extend to, um, kind of ranking scenarios or sort of acting in hypothetical scenarios or overall kind of to, um, all possible situations rather than just our actual situation. I think, um, uh, infinities come in there. But then I think maybe more importantly, um, I think it's, uh, it's an issue with practical relevance, um, and a way to see that is that, you know, I think we should have nonzero credence that we live in an infinite world, um, and, uh, you know, it's a very live, uh, physical hypothesis that the universe is infinite even if I think the mainstream view is that our causal influence on that universe, um, is finite in virtue of things like entropy and light speed and stuff like that. Um, but the universe itself may well be infinite in, uh, in, um, uh, you know, uh, and possibly infinite in a number of different ways. Uh, uh, if-... that sort of Max Tegmark has some work on, all the different kind of like large, (laughs) you know, ways the universe can be really very large. There's a number of ways that, yeah, I think it's just y- we should have non-zero credence that, that we, we can have, um, infinite influence wi- in our actions now. Um, so, uh, you know, our kind of, uh, the causal influence, our, the, the limitations there could be wrong. It may be that there are ways, you know, in the future we'll be able to do infinite things. Um, and then I also think somewhat more, uh, uh, exotically that, um, it's, there, there are sort of ways of having a causal influence, um, on a, on an infinite universe, even if you are, uh, limited in your causal influence. And that comes from some additional work I've done on decision theory. Um, and so if you try to incorporate that, if you're a sort of expected value reasoner, um, it just very quickly starts to dominate or at least break your expected value calculations. So, you know, you mentioned longtermism earlier, uh, and, you know, a natural f- reason, a natural argument for, for getting an interest in longtermism is, "Oh, you know, in the future, there could be all these people. Their lives are incredibly important." So if you do the EV calculation, sort of your effect on them is what dominates. Um, but actually, if you have even a tiny credence that you can do an infinite thing, uh, you know, either that dominates or it breaks. And then if you have tiny credences on doing different types of infinite things and you need to compare them, um, you need to know how to do it. Uh, and so I just think this is actually, you know, it's actually a part of our, of our epistemology now, um, though it's, I think we often don't, uh, don't treat it that way because we're often not doing EV reasoning or really thinking, thinking about that. Um, that, uh, that these are questions that, that just apply to us.

   3. DPDwarkesh Patel44:28

      Yeah. Yeah. The, the, so that's, that's so- super fascinating. Um, I, I, I... If it is the case that we can only have an impact on a finite amount of stuff, then maybe it is true that like there's infinite suffering or happiness in the universe at large, but, uh, the delta between, uh, the best case scenario for what we do and the best- worst case scenario is finite. Um, but yeah, it, it, I don't know. That still seems less compelling if the, the, the hell or heaven we're surrounded by is overall not, uh, doesn't change. Um, uh, can you talk a bit more, I think you mentioned, uh, in your other work on having impact, uh, i- having infinite impact be- beyond, uh, the scope of what light speed and entropy would allow us. C- c- can you talk a bit more about how that might be possible?

   4. JCJoseph (Joe) Carlsmith45:12

      Sure. So, um, you know, uh, a common decision theory, um, though it's not, I think, the mainstream decision theory, it's a contender in the literature, is evidential decision theory, where you should act, um, such that, uh, you would be, you know, roughly speaking, happiest to learn that you had acted that way for that reason. (laughs) Um, and, uh, so the reason this allows you kind of a causal influence, uh, so, you know, a way of thinking about it is suppose that you are a, um, a deterministic simulation, um, and there's a copy of you being run, uh, sort of too far away for, uh, for you to ever, uh, causally interact with it, right? Um, but you know that it's a sort of, um, it, you know, it's, uh, it's a deterministic copy, and so it'll do exactly what you do absent some sort of commu- computer malfunction. Um, and now, uh, you're deciding whether to give, uh, you know... You have two options. You can send a million dollars to that. Well, that's a little complicated because he's too far away. But, um, uh, you know, just in general, like, if I raise my hand or if I wanna write stuff on, on my whiteboard, right?Um, or if I'm going to, uh, you know, there's, let's say I have to make some ethical decision, like whether I should take an expensive vacation, um, or I should donate that money to save someone's life. Because that, the, the other guy, uh, is going to act just like I do, um, even though I can't cause him to do that, in some sense, when I, when I make my choice, um, after doing so, I should think that he made the same choice. And so evidential decision theory treats his action as, in some sense, under my control. Um, and so, uh, if you imagine an infinite universe where there are an infinite number of copies of you, or even not copies, people whose actions are correlated with you, such that when you act a certain way that gives you evidence about what they do, um, in some sense, their actions are under your control. And so if there are an infinite number of them, uh, on evidential decision theory and a few other decision theories, uh, then, uh, in some sense, you're having influent- influence on the universe.

   5. DPDwarkesh Patel47:04

      Yeah, this sounds really similar to, um, the thought experiment in quantum mechanics called EPR pair, uh, which, which you might have heard of. But the basic idea is if you have two entangled bits and you take them very far away from each other, and then you measure one of them, and you, like, before they're brought apart, you come up to some rule that like, "Hey, if, if it's plus, we do this. If it's minus, we do the other thing." It seems, uh, at first glance that measuring something yourself, uh, has an impact on what the other person does, even though, um, it shouldn't be allowed, uh, by light speed. Uh, it gets resolved if you take the many worlds view. But, um, yeah, yeah. So that, that, that's very interesting. Is this just a thought experiment or is this something that we should anticipate for, uh, some cosmological reason to actually be a way we could have influence on the world?

   6. JCJoseph (Joe) Carlsmith47:53

      So I haven't dug into the cosmology a lot, but my understanding is that it's, at the very least, a very live hypothesis that the universe is, um, infinite in the sense that there are... Um, you know, it's sort of infinite in extent, and there are, uh, you know, suitably far away, um, there are copies of us having just this conversation. And then, you know, even further away, there are copies of us having this conversation, but wearing raccoons for hats, um, and, you know, and, and all the rest. Um, which, you know, is itself something to wonder about and sit with. But I, you know, my understanding is this is, this is just a live hypothesis, and, and that more broadly, um, kind of infinities play... You know, infinite universes are just sort of a part of, of, uh, of mainstream cosmology at this point. Um, and so, uh, yeah, I think it... I don't think it's just a thought experiment. I think infinite universes are, are live. And then I think, um, uh, you know, these sort of non-casual decision theories are actually my sort of best guess decision theories. Um, though that's not a mainstream view. Uh, so, uh, it's fairly, um... I think it comes in fairly directly and substantively if you have that combination of views. But then I also think it comes in, uh, I think everyone should have non-zero credence in all sorts of different infinity involving hypotheses. And so infinite ethics gets a grip regardless.

   7. DPDwarkesh Patel49:06

      I see. Um, and then, uh, so y- you're taking that example. Um, if it l- if you're having an impact on every identical copy of yourself i- i- in the infinite universe, it seems that for any such copy, there's infinite amount of other copies that a- are slightly different. So, it's not even clear if you're increasing, m- maybe it makes no sense to talk about proportions in an infinite universe. But, you know, if- if- if there is a- a- another infinite set of copies that just scribble the exact opposite thing on the whiteboard, then it's, um, uh, i- i- it's not clear that you had any impact on the total amount of good or bad stuff that happened? I- I- I don't know. My brain breaks here, but, uh, m- maybe you can help me understand this.

   8. JCJoseph (Joe) Carlsmith49:47

      Yeah. So, I mean, there's a general... I think there's a couple of dimensions here, there. So one is, um, trying to understand actually what sort of difference does it make if you're in this sort of infinite situation and you're thinking about a causal influen- influence. Um, what even did you change, um, at a, at a sort of empirical level, before you talk about how to value that? Um, and I think that's a pretty gnarly question. Um, even if we settled that question though, in terms of, like, the empirical, uh, acausal impact, uh, there's a further question of how do you, how do you rank that? Or how do you deal with, um, you know, the sort of, the normative dimension here? And there, th- you know, so that's the sort of ethical question, and there, things get really gnarly, very fast. Um, and, you know, so, uh, and in fact there are kind of, um, impossibility results that show that even very basic constraints that you really would have thought that we could get, um, at the same time in our ethical theories, uh, you can't get them at the same time, um, when- when you come, when it comes to infinite universes. Um, and, uh, so we know that something is gonna have to go and change if we're going to extend our ethics to- to infinities.

   9. DPDwarkesh Patel50:58

      I see. But then, um, uh, so and then what- what- is there some reason you've settled on... I- I guess you mentioned you're not a utilitarian, but on some version of EA or long-termism as your attentive moral hypothesis, despite the fact that this seems unresolved? And then, like, how- how- how- how do you sit with that tension while tentatively, uh, r- remaining an EA?

   10. JCJoseph (Joe) Carlsmith51:21

       Yeah. So, I think there's a, there's two dimensions there. One is that I think it's, um, I think it's good practice to not totally upend your life and do, and- and, you know, if you encounter some destabilizing philosophical idea, especially one that's sort of difficult and, you know, you don't totally have a grip on it, to- to then come back to it.

   11. DPDwarkesh Patel51:38

       I- I- but isn't, isn't that what long-termism is?

   12. JCJoseph (Joe) Carlsmith51:41

       Yeah. So, I think there's a real tension there in that I think, um, many, you know, how seriously should we take these ideas? At what point should you be making what sorts of changes to your life on the basis of different, different things that you're, um, uh, you're thinking and believing? You know, it's a real art, right? And I think some people go, you know, they grab the first idea they see, um, and they start doing crazy stuff, and, uh, in an unwise way. And some people are too, um, just kind of sluggish, and they, they're not willing to take ideas seriously and not willing to reorient their- their life on the basis of- of, uh, changes in what- in what seems true. Um, but I think, you know, nevertheless, I think... Especially things that involve, like, "Ah, turns out it's fine to, you know, do terrible things," or, you know, "There's no reason to eat your lunch," or whatever. Like, things that, you know, sort of really- really holistically breaking of your ethics views, I think, I think one should- should tread very cautiously with. Um, so that's one aspect. At a philosophical level, um, the way I resolve it is, I think for many of these issues, uh, the right path forward, or at least a path that looks pretty good, is to, um, survive long enough for our civilization to become much wiser. Um, and if, as, and- and- and then to use that position of wisdom and empowerment, uh, to act better with respect to these issues. Um, and so, and that's what I say in the end of the infinite ethics post is that, um, you know, I think future civilization, if all goes well, will be much better equipped to deal with this, um, and, you know, we are at, we are at square one in kind of really understanding how- how- how these issues play out and how to respond. And so, uh, I think both at an empirical level and at a, at a kind of philosophical level. Um, and so it- it looks convergently pretty good to me to survive, become wiser, keep your options open, and then act from there. Um, and that looks... That ends up pretty similar to a lot of long-termism and existential risk. It's just that it's focused less on, "And the main event will be what happens to future people," and it's- it's more about getting to the point where we are wise enough to understand and reorient, um, in a better way.

   13. DPDwarkesh Patel53:42

       Okay. Um, yeah. So, what- what I find really interesting about this is that you can, um... Yeah, you, uh, so, uh, different people tend to have, like, different thresholds for, um, e- epistemic learned helplessness, where they basically say, "This is too weird. I'm not gonna think about this. Uh, let's just stick with my current, uh, current, uh, moral theories." Um, so for- for some- somebody else, it might be before they became a long-termist, where it's just like, "Yeah, trillions of future people. What- what are we talking about here? Let's... Uh, I'm not changing my mind on this stuff." And then, yeah, for- for- for you, maybe it's before the infinite ethics stuff. Um, uh, is- is- is there some principled reason for thinking that this is where that stop should be? Or b- is it just a mat- matter of, like, temperament and openness?

   14. JCJoseph (Joe) Carlsmith54:25

       So, I don't think there's a principled reason. And- and I should say I don't think of my attitude towards infinite ethics as solely, "Oh, this has gotten too far down the crazy, the crazy path. I'm out." Um, it is, this thing about the wisdom in the future is pretty important to me, um, as a- as, um, a- a reason, uh, uh, as an, as a mode of orientation. A first pass cut that I use is when do you feel like it's real? Um, if you feel like a thing is real, uh, as opposed to a kind of abstract, fun argument, um, then that's important, or that's- that's a real signal. And I think, I... So, um, uh, and I generally encourage people, if- if the sort of mode that I- I- I- I- I don't know, I'm drawn to is something like, if there's an idea that seems compelling intellectually, that's a reason to investigate it...... a, a lot and think about it and really grapple with, uh, you know, if, if this doesn't seem right to you or if it seems too crazy, why, um, and really kind of processing, you know, it's a reason to pay a lot of attention. But if you've paid a lot of attention, at the end of the day, you're like, "Well, I guess at an abstract level that sort of makes sense, but it just doesn't feel to me like the real world. It just doesn't feel to me like, um, wisdom or, uh, you know, like a healthy way of living or whatever," then I'm like, "Well, maybe you shouldn't do it," right? I mean, and, and, uh, and I think some people will do that wrong, and they will end up bouncing off of ideas that are in fact good. Um, but, uh, you know, I think overall, uh, e- e- e- these, these are sort of sufficiently intense and difficult issues that, um, uh, kind of being actually persuaded and not just sort of chopping off the rest of your epistemology for the sake of some, like, version of the abstraction, uh, is, uh, I- it seems to me important, and it's, and it's a sort of a healthier way to relate.

   15. DPDwarkesh Patel56:12

       Yeah, so another example of this, um, is that you have this really interesting blog post on ants, uh, and (laughs) take your, uh, the, your, your, your thoughts after, uh, uh, sterilizing a colony of them. So, um, (laughs) I, uh, yeah, so this is another example of a thing where, uh, uh, almost everybody other than, I don't know, maybe a Jane who wears a face mask to prevent bugs from going into his mouth would agre- uh, say like, "Okay, at this point, if we're talking about how many hedons are in a hectare of forest from, from all the millions of insects there, um, uh, then, uh, you've lost me." Um, but then, you know, somebody else might say, "Okay, well, there, there's not a strong reason for thinking they have no, absolutely no capacity to feel suffering." Um, uh, yeah, so, uh, I, I, I wonder how you think about such questions because you can't, like, stop living and not kill... Uh, you're, you're not even gonna stop going on road trips where you're probably killing hundreds of insects by just driving. Um, but yeah, so how, how, how, wh- what do you think about such conundrums?

   16. JCJoseph (Joe) Carlsmith57:11

       The... I have significant uncertainty about, uh, you know, e- exactly, and I think this is the appropriate position about exactly how much, uh, kind of consciousness or suffering or other forms of moral, you know, other ways, other kind of properties that we associate with moral patienthood, how much those apply to different, um, different types of insects. Um, I think it's a strange view t- to be, you know, extremely confident that, uh, what happens with insects is, uh, totally morally neutral. And I think it actually doesn't fit with our common sense. So let's say you see, if you see a child, like, frying ants, uh, with, uh, with a magnifying glass. I think we, you know, there is some, uh, y- you know, one... Y- you could say, "Ah, well, that just indicates that they're gonna be cruel to other things that matter." Um, but, uh, I don't think so. I think, you know, when you see the ants, like, you know, and they're, they're twitching around and, and, um, I... So I think we aren't... Um, and, you know, as in many cases with an- with animal ethics, I think we're a bit, like, kind of schizophrenic about, about what cases we, we view as sort of morally relevant and, and which, which not. Um, you know, we have ant... We have, you know, pet treatment laws and then we have factory farms and, and stuff like that. Um, so I, I don't see it as a radical position that ants matter somewhat. Um, I think there's a further question of what your overall p- practical response should, to that should be. And I think, um, the main... Uh, and I do think that kind of costs... As, as in a lot of ethical life, there are trade-offs and you have to make, um, you have to make a call about what, what sort of constraints you're gonna put on yourself at the cost of other goals. And, um, in, you know, in the case of insects, it's not my, my current moral focus and I don't pay a lot of costs to kind of, um, uh, to lower my impact on animals and, and I don't, you know, (laughs) I don't sw- I don't sweep the sidewalk or, or any, or sorry, on, on, on ants in particular. Um, uh, and so I think it's, I think, and I think that's, you know, that's my best guess response and that, and that has to do with other ethical priorities to my life. Um, but I think, you know, there's, there's the middle ground between, um, I shall ignore this completely and I shall, you know, be a Jane, um, which is recognizing that this is a, this is real trade-off, there's uncertainty here, and, um, uh, and taking responsibility for how you're responding to that.

   17. DPDwarkesh Patel59:23

       Yeah, this seems, um, kind of similar to the infinite ethics example where if you put any sort of credence that, um, uh, that they have any ability to suffer, then at least if you're, uh, not going to say that, "Oh, m- uh, it doesn't matter because, like, the far future, uh, trillions and trillions of ants." Um, (laughs) uh, it, it seems like this should be, uh, a compelling, uh, uh, a com- a com- a compelling thing to think about. But then the result is, um, yeah, it's, it's not even, like, become a vegan where it's like you change your diet. Um, uh, and, and then so, you know, uh, uh, as you might know, this is used as a reductive at ab- ad absurdum of veganism where, you know, if you're gonna start caring about other, uh, non-human animals, why not also care about insects? And even if they were worth, like, a millionth of a cow, then, you know, you're probably still killing, like, a million of them on any given day from all your activities, uh, indirectly maybe. Uh, like, I, I don't know. Like, the food you're eating, all the pesticides that are used to create that food. Uh, I don't know how you go about resolving that kind of stuff.

   18. JCJoseph (Joe) Carlsmith1:00:24

       I mean, I guess I'd want to really hear the empirical case. I think, um, uh, I think it's true. You know, there are a lot of insects. Uh, and, but, y- you know, I think it's easy... Uh, you know, I think if you want to say, like, "Ah," taking seriously sort of the, the idea that, um, there's some reason to, to not, like, squash, squash a bug (laughs) , um, if you see it leads immediately to kind of Jane-like behavior, uh, absent long-termism or something like that. I, I really, I, I feel like I want to hear the empirical case about, like, exactly what impact you're having and, and how, um... And, and I'm not at all persuaded that that's the practical upshot. Um, uh, and if it is, if that's a really strong case, then I think that's an interesting, um, an interesting, uh-... uh, you know, that's an interesting kind of implication of, of this view. Um, and, uh, and, you know, worth, worth concern. But I wouldn't jump... It, it feels to me like it's easy to jump to that almost out of a desire to, to get to the reductio, um, without kind of... I would try to move slower and, and really see. It's like, "Wait. Is that right? There are a lot of trade-offs here. What's the source of my hesitation about that?" Um, and kind of, uh, yeah, and not, not jump too quickly to something that's sufficiently absurd that I can be like, "Ah. Therefore, I get to reject this whole mode of thinking even though I don't know why."

   19. DPDwarkesh Patel1:01:38

       I see. Yeah. Um,

9. 1:01:40 – 1:18:42

   SIA vs SSA

   1. DPDwarkesh Patel1:01:40

      okay. So, let's talk about, uh, the two different ways of thinking about observer effects and their implications. So, do, do, do you wanna, uh, ex- explain, um... You have a four-part series on this, but do you wanna explain, uh, the self-indication assumption and the self-sampling assumption? Uh, (laughs) I, I know it's a big topic but, uh, yeah. A- as much as possible.

   2. JCJoseph (Joe) Carlsmith1:02:03

      Sure. So, I think one way to start to get into this debate is by thinking about the following case. So, um, you wake up in a white room and there's a message written on the wall, and let's say you're gonna believe this message. And the message says, "I, God..." It's from God. "I, God, created... I flipped a coin, um, and if it was heads, I created one person in a white room, and if it was tails, I created a million people, all in white rooms." And now you are asked to assign probabilities, uh, to, uh, the coin having come up heads versus tails. Um, and, uh, so one approach to this question, um, uh, which is the approach I favor or at least think is better than the other, uh, is the self-indication assumption. These names are terrible, um, and (laughs) but, you know, so it goes. Um, so SIA, uh, says that, uh, your probability that the coin came up heads should be approximately one in a million. Um, and that's because SIA thinks it's more likely that you exist in worlds where, uh, there are more people in your epistemic situation or more people who have your evidence, which in this case is just waking up in this white room. Uh, and that's... And so that's, that can be a weird conclusion and go to weird places. Um, but I think, uh, it's, uh, a better conclusion than the alternative. SSA, uh, which is the main alternative I consider in that post, which is the self-sampling assumption, says that you should... You think it more likely that you exist in worlds where people with your evidence are a larger fraction of, uh, something called your reference class. Um, uh, where it's quite opaque what, what a reference class is supposed to be. But broadly speaking, a reference class is the sort of set of people you could have been, or that's kind of how it functions in SSA's discourse. So, um, uh, in this case, in both cases, everyone has your evidence, um, and so the fraction is the same. Um, uh, and so you, you stick with the one-half prior. Um, but that's not true... So SSA, in other contexts, um, not everyone has your evidence, and so it updates towards worlds, um, where it's a larger fraction. So famously, uh, SSA leads to what's known as the doomsday argument, um, where you imagine that, uh, there are two possibilities. Either humanity will go extinct very soon, um, or we won't go extinct very soon and there will be tons of people in the future. Um, and in the former case... Uh, and then you imagine, um, everyone as sort of ranked in terms of when they were born. Um, uh, in the former case, people born, you know, at roughly this time, um, are a much larger percentage of all the people who ever lived. Um, and so if you imagine, you know, God first creates the world and then he inserts you randomly into, like, some group. It's much more likely, uh, that you would find yourself in the 21st century, um, if humanity goes extinct soon than if it's, if there are tons of people in the future. If God randomly inserted you into these tons of people in the future, then it's like really... It's a tiny fraction of them are in the 21st century. Um, so SSA in other contexts actually im- you know, it has these important implications, namely that in this case you update very, very hard towards the future being short. Um, and that matters a lot for long-termism because long-termism is all about the future being big in expectation.

   3. DPDwarkesh Patel1:05:25

      Okay. So what is the SIA take on this?

   4. JCJoseph (Joe) Carlsmith1:05:29

      Yeah. So, I think a way to think about SIA's kind of story... So, I gave this story about SS- SSA, which is... It's sort of like this. It's like first God creates a world. This is SSA. First, he creates a world, and then he takes... And he's dead set on, on putting you into this world. So, you... He's got your soul, right? And he really wants... And your soul is going in there no matter what, right? Um, but the way he's going to insert your soul into the world is by throwing you randomly into some set of people, um, the reference class. Uh, and so if you wake... So, you should expect, um, to end up in the world where, uh, the kind of person you end up as, uh, is, is sort of, um, more like... A more likely result of that throwing process. It's a sort of larger fraction of, of the total people you could have been. What SSA... Or what SIA thinks is different, the way... The story that I'll use for SIA, though it isn't... This isn't the only gloss, is God de- decides he's gonna create a world, and then he... And say there's, like, a big line of souls in heaven, and he goes and grabs them kind of randomly out of heaven and puts them into the world, right? And so in that case, if there are more people in the world, then you've got more shots at be-... And you're one of these souls. You're sort of sitting in heaven, hoping to get created. Um, uh, on SIA, God has more chances to grab you out of, out of heaven and put you into the world if there are more people, uh, who... Uh, more people like you in that world. Um, and so you should expect to be in a world where there was sort of... There, there are more such people, and that's kind of SIA's vibe.

   5. DPDwarkesh Patel1:07:07

      Doesn't this also imply that you should be in the future, assuming there will be more people in the future?

   6. JCJoseph (Joe) Carlsmith1:07:12

      Tell me more about wh- why, why we would imply that.

   7. DPDwarkesh Patel1:07:14

      Okay. In an analogous scenario, maybe like, uh, uh, go back to the God tossing the coin scenario, where if- you just, uh, substitute for people in white rooms, you substitute-

   8. JCJoseph (Joe) Carlsmith1:07:24

      Mm-hmm.

   9. DPDwarkesh Patel1:07:25

      ... um, being a thing, uh, a conscious entity, and if there's gonna be more conscious entities in the future, like, you would really expect to, just like in that example of being in that scenario where there's a lot more rooms, just as maybe you should expect U- to be in that scenario where there's a lot more conscious beings, which presumably is the future. So then it's still odd that you're in the present, uh, under SIA?

   10. JCJoseph (Joe) Carlsmith1:07:46

       Yes. So in, in a specific sense. So, um, it's true that on SIA, uh, say that, um, say that we don't know what room you're in first, right? So, so, um, you, you wake up in the white room and you're wondering, uh, am I in room one or am I in rooms two through a million (laughs) , right? Um, and on SIA, what you did first, so you woke up and you don't know what room you're in, but there's a lot more people in the world with lots of rooms. And so you become very, very confident that you're in that world, right? So you're very, very confident on tails. And then you're right that, uh, conditional on tails, you think it's much more like... you sort of split your credence evenly between all these rooms. So you are, uh, very confident that you're in one of the, the sort of two through a million rooms and not, not room one. Um, but that was before you've seen your room number. Um, once you see your room number, it's true that you should be quite surprised about your room number. Um, uh, but the, uh, once you get the room number, you're back, you're back to 50-50 on, uh, heads versus tails because you had sort of equal credence in being in room one, uh, conditional on tails. Um, or sorry, uh, you had equal credence in being in tails in room one, uh, and, uh, heads in room one. And so when you get rid of all of the other tails and rooms two through a million, you're left with 50-50 overall on heads versus tails. Um, and so, uh, the, the sense in which SIA leaves you back at normality with the doomsday argument is once you update on being in the 21st century, which admittedly should be surprising. Like if you didn't know what, that you were in the 21st century and then you learned that you were, you should be like, "Wow, that's really unexpected." And fair- so, and that's, that's true. But I think once you do that, you're back at, um, uh, you know, whatever your prior was about, about extinction.

   11. DPDwarkesh Patel1:09:42

       Maybe I'm still not sure on why the fact that you were surprised should not itself be the doomsday argument.

   12. JCJoseph (Joe) Carlsmith1:09:49

       Yeah, I think there's an intuition there, um, which is sort of like, yeah, is SIA making a bad prediction? So you, you could, you could kind of update against SIA because SIA-

   13. DPDwarkesh Patel1:09:59

       (laughs)

   14. JCJoseph (Joe) Carlsmith1:09:59

       ... would have predicted that you're in the future. Um, I think there's something there, and I think there's a few other analogs. Um, like for example, I think SIA naively predicts that, um, y- you know, you should find yourself in a situation where there are just tons of people that, you know, a situation obsessed with creating people with your evidence. Um, and that, you know, this is one of the, one of the, the problems with SIA. So you should expect to find, you know, in every nook and cranny a simulation of you. As soon as you like, you know, you open the door, it's actually this giant bank of simulations of you in like-

   15. DPDwarkesh Patel1:10:28

       (laughs)

   16. JCJoseph (Joe) Carlsmith1:10:28

       ... your previous epistemic state. Um, and so, you know, I think there are... (laughs) and, and then you don't see that, you might be like, "Well, I should update against the anthropic theory that predicted, uh, that I would see that." And I, I, I think there are arguments in that vein.

   17. DPDwarkesh Patel1:10:40

       Yeah. So maybe let's back up to go to the original example, uh, that w- that was used to distinguish these two theories. Yeah. So I, I, how, I, uh, c- can you help me resolve my intuitions here? Where my, my intuition is very much SSA because, um, yeah, it, it seems to me that, uh, you knew you were gonna wake up, right? You knew you were gonna wake up in a white room. Before you actually did wake up, I mean, your prior should have been like one half of heads or tails. So it's not clear to me why, having learned nothing new, your, uh, your posterior probability on either of those scenarios should change.

   18. JCJoseph (Joe) Carlsmith1:11:13

       So I think the, the SIA response to that would be, or at least I think the, a way of making it intuitive would be to say that you didn't know that you were going to wake up, right? So in the, um... if we go back to that Justo story where God is grabbing you out of the, um, out of heaven, uh, you know, i- it's, uh, it's not at all... it's actually incredibly unlikely that he grabs you. There are so many, so many people... I mean, there's a different thing where SIA is, in general, very surprised you exist. And in fact, that's, uh, the, um... so you could make this same argument as like SIA says you shouldn't exist. Isn't that weird that you exist? Um, and I actually think that's a, that's a good argument. Um, or is it... (laughs) uh, so, um, but, uh, once you're in that headspace, then I think the way, the way to think about it is that it's not a guarantee that you, you were... God is not dead set on creating you. You are a particular contingent arrangement of the world. Um, and so that, that you should expect that arrangement to, to come about more often if there are more arrangements of that type, um, rather than sort of assuming that, that no matter what, existence will include you.

   19. DPDwarkesh Patel1:12:17

       Yeah. So c- can you talk more about the problems with SSA, like scenarios where you think it breaks down, uh, like why you prefer SIA?

   20. JCJoseph (Joe) Carlsmith1:12:25

       Yeah. So, um, an easy, an easy problem, uh, or sort of one of the most dramatic problems is, is that SIA... SSA, um, predicts that, uh, it's possible to have a kind of telekinetic influence on the world. So imagine that there's a, um, there's a puppy. You, you're in an... you wake up and you're in an empty universe, except for this puppy and, and you, and this boulder that's rolling towards the puppy, right? And the boulder is inexorably going to kill the puppy. Um, it's a very large boulder. It's basically guaranteed that the puppy is dead meat. But you have the power to make binding pre-commitments, um, that you will in fact execute. And, and you have also to your right a button that would allow you to create tons of people, like zillions and zillions and zillions of people, all of whom, um, are wearing different clothes from you. Uh, so they would be in a different epistemic state than you if you, if you created them. Um, now SSA, uh, so you, so you make the following resolution. You say, um...If, uh, this boulder does not jump out of the way of this puppy, um, like the boulder leaps, you, you know, in some very weird, very unlikely way, um, uh, then I will press this button and I will create zillions and zillions of the people, um, uh, all of whom are in a different epistemic state than me. But let's assume they were in my reference class. Um, SSA thinks that it's sufficiently unlikely that you would be in a world with zillions of those people, um, and, uh, but you, you know, you at the very beginning, uh, with, with a different, with different colored clothes because, you know, that was a tiny fraction of the reference class that those people get created. Um, that SSA thinks it's actually more likely once you've made that commitment that the boulder will jump out of the way, um, and, uh, and, you know, (laughs) and so that, and that looks weird, right? You, it, it just seems like that's not gonna work. You can't, you can't just make that commitment and then expect the boulder to jump. Um, and you get... So that, so that's a sort of exotic example. You get, you get similar analogs even in the God's coin toss case where, um, like naively, it doesn't actually matter whether God has tossed the coin yet, right? So suppose, um, yeah, so like let's say, let's say you wake up and learn that you're in room one, right? Um, but God hasn't tossed the coin. It's like He created room one first before He tossed, and then He's gonna toss and that's gonna determine whether or not He creates all the rooms in the future. Um, if you, on SSA, once you wake up and learn, um, learn that you're in room one, you think it's incredibly unlikely that there's going to be these future people. So you, now you say before, uh, it's a fair coin. God's going to toss it in front of you. You're still going to say, "I'm sorry, God, it's, uh, you know, it's a one in a million chance that this, uh, that this coin lands tails." Um, and, uh, or sorry, one in a million, some- something like a very small number. I for e- exact... I forget exactly. And that's, um, and that's very weird. That's a fair coin. It hasn't been tossed. But you, with the power of SSA, have become extremely confident about, about what, how it's gonna land. Um, and that's, uh, so that's, that's an- another argument. There's a, there's a number of other, uh, I think really, really bad problems for SSA.

Episode duration: 1:32:09