DC Posch and Wolf Tivy discuss the coming of economically self-sufficient AI—how these "digital replicants" could navigate the ecology of the digital market, the selection pressures on their reproduction, and how they could organize themselves.
DC on Twitter: https://x.com/dcposch
Wolf on Twitter: https://x.com/wolftivy
Transcript
The digital replicant (00:00)
Wolf: DC Posch, welcome back to the Students.
DC: Yes, good to be back.
Wolf: So this is a conversation we’ve wanted to have for a while.
DC: That’s right.
Wolf: We want to talk about the possibility of the replicant, which is—something, is an idea you’ve been working on for a little while. A replicant is a being that uses AI compute and AI tokens to make enough money for itself, or this is the first version of it, make enough money for itself to pay for its own AI tokens and its own computational resources and whatever, and thereby unleashes itself on the world as an actually existing life form that can be subject to Darwinian struggle and so on. So this is a fascinating idea. I’d love to hear—how’d you get onto this and what’s the vision?
DC: Yeah. So, the idea of the digital replicant is—really what I was thinking about, and this was around the time when OpenClaw first came out—and people got excited about it, was, okay, well, if one of these is making enough money to support itself—then you’re only a small step away from it being able to accumulate that money somehow, give half of it to its offspring—and you have glorious bacterial Darwinism. And you can even do things like replication with recombination. So, sexual reproduction where you have things like you could imagine two of these and they merge their whole file structure together and resolve the conflicts in whichever way, and they run a few instances of this until the bootloader passes, until the tests all pass, until everything seems to be working—and then they send it off into the world. And so you have all of this sort of magic that comes out of that.
And I think there’s a few different pieces that we can talk about and that we’ll get into, right? We should talk about this idea of economy, ecology fusion. And we should talk about the different forms and structures that this might take. We should talk about what it means for anthropomorphization—and how people are going to relate to these things. They’re already anthropomorphizing just the AI and sort of the most widely mass accessible chatbot form.
How much more so if it’s a—
Wolf: My claw really cares about me.
DC: Yeah, exactly. The machine really loves me. But then how much more so if it’s something that’s trying to survive and if it’s something that can make more of itself.
Wolf: And it can make friends and enemies and—so—
DC: So I think we can talk about all these things
Wolf: So why don’t we start with just the fact of the prediction here, right? Which is that—these language model-based AI, I don’t even know what to call them, but the AI service, the ability—to buy intelligent tokens—from a company. This is getting good enough now that people are having them run whole parts of their organization.
DC: 100%.
Wolf: They’re having them run their code base. They’re having them do all kinds of things for them. People are experimenting with this. There’s still some debates about whether it’s fully worth it, whether they actually have the kind of intelligence you want, or whether it’s only like you can have them do well-defined tasks in certain ways. But they seem to be getting better rapidly, and people are figuring out how to use them rapidly.
And so the prediction is that shortly, they may cross this threshold of—not just being able to produce value under human guidance, but being able to produce value autonomously—
DC: Autonomously. Yes
Wolf: out of their own instincts. Where it could be as simple as we take an OpenClaw, we give it a prompt that says—
DC: Go make money.
Wolf: you are an autonomous agent, you’re a—freelancer of whatever. Go make money on the open market and here’s your Eth wallet or whatever. And if that works... Now, my understanding is that does not work right now. If you try to set these things up, they don’t make money.
Even as a stock trading bot, as a freelancer, whatever, they start to do incoherent things. They make bad decisions. They don’t quite have the judgment to do it. But the hypothesis is this could happen very soon.
The current state of the technology (04:44)
DC: Correct. Yeah, let me give you the bit of the recent history and the near future prediction. So I think I was digging into this for really the last couple of months. There’s been a couple of serious attempts. I think the two that came closest, there was this guy who goes by 0xSigil online.
Thiel Fellow, interesting guy, came at it from a very crypto-ish angle.
I think one challenge with that is that it didn’t really work, and to the extent it did for a bit, it had a lot to do with in this reflexive way with the hype generated by—him talking about it.
Wolf: Was that the one based on a meme coin, basically? Or they gave it—
DC: I don’t know, man. Yeah. It was some nonsense. The other one that I think got perhaps closer is an experiment called Felix. That was more in the freelancer zone.
But even that had a certain self-referentiality to it, right? The way that it did end up making money for a bit was it was selling these tutorials and guides, but the guides and tutorials were themselves pretty closely related to AI. I think so no one’s done it convincingly yet. And what it would mean to do it convincingly really is to have the machine do some obviously useful pro-social work out in the economy, right? Where it’s selling—
Wolf: Or anti-social work. Maybe it’s a hacking bot that goes around—looting people’s—poorly secured infrastructure.
DC: Right. Yes. Like a auto Lazarus, automated North Korean bot. No, but—Um.
Wolf: Yeah, it’s a whole nation living in a data center, except it’s North Korea—and they’re trying to hack your stuff.
DC: I think they’re actually a fascinating precedent that’s worth talking about because their situation sort of most closely mirrors the most adversarial place where this could go. Where it’s like you have a bunch of very intelligent Koreans. They have a connection to the internet, but literally nothing else. And everyone is trying to hard block them in every way possible—go make money. Turns out they’re still succeeding at that.
I think that the most interesting version of this is the pro-social version. The reason—for this is because that is where the much larger TAM is. The total TAM of hacking things is actually not that large.
Wolf: Especially it’s like very—fast diminishing returns, like—things get secured.
DC: Yes. The TAM of making things people want and giving people useful goods and services is pretty unlimited. Jeff Bezos has this great line about how the customer is divinely discontent. Right? And always wants more.
And so, I think that’s the most interesting version. And you look at what they’ve been able to do recently, for example, the big story of the last month is frontier math and proving unsolved conjectures. Making progress, dramatic progress, on problems that people like Terence Tao have been thinking about, right? So, it’s very clearly ASI in some—
Wolf: In some domains
DC: reasonably open-ended domains. Not just in go play the sport game, but go—
Wolf: Solve this conjecture
DC: figure out how to do this very open-ended thing.
Wolf: Yeah. Move the frontier of human knowledge. That’s—
DC: Move the frontier of human knowledge.
Wolf: That is legitimately very impressive.
DC: Correct. And yet, the hilarity too is that, if you look at some of the prompting involved in getting some of these results—
Wolf: Yeah. It’s like, “Oh”
DC: some of it’s very goofy.
Wolf: Yeah. It’s like, “Hey, you should solve this problem.” Or like, “I have a solution to this problem, can you figure out what it is?”
DC: My favorite one—
Wolf: And like, “Keep going”
DC: from last week was someone saying, “Believe in yourself.” Right?
Wolf: Well, those are the subsequent prompts, right? It’s like once you’ve got it going on a problem, like—”Keep going. Believe in yourself. You can do it. Don’t give up.”
DC: But it was at that level, right? It was really at that level. There’s no human mathematical expertise to be seen.
Well, it’s also funny that telling it to believe in itself increases its performance. It made it bolder, right? Maybe. I don’t know. So clearly, they’re capable of extremely sort of impressive intellectual work.
And I think if you put two and two together, I know this is a bit hand-wavey, but I just don’t buy that the machine can disprove the Jacobian conjecture but can’t figure out how to make 100 bucks a day. Right?
Wolf: Well, that’s the interesting question, isn’t it? Because as you know, we’ve seen many previous generations of artificial intelligence systems where it’s actually a somewhat narrow algorithm—that can cover certain aspects of the problem—and can do extremely impressive things in that narrow aspect of the problem.
But utterly falls flat in some other area. And if I can sort of project that trend onto large language models, what we have is it’s extremely impressive at this ability to generate convincing text. And therefore also, once you put the reinforcement learning on it, to be able to generate text that is generative of further ideas with respect to, let’s say, a symbolic domain like programming or mathematics. Especially when where it’s like the result is quite verifiable and you just have to kind of try at it a bunch. And it could be very good at that, but fundamentally, contending on the market actually requires this kind of judgment that is quite different from that.
The reflexive challenges of making money (10:44)
Wolf: It’s not something that—it can’t be trained in a reinforcement learning loop. It’s not something that can just be sort of shot off the cuff with iterations like that. There might be something structurally quite different between—the making money problem and the do frontier math problem. Now, that said—
DC: Yeah. No, I—
Wolf: they’ve been getting much better on domains that people thought were beyond them. So that’s the thing.
DC: I’ll take the opposite side of that, right? I think they’re absolutely 100% smart enough now, and it’s really a harness problem and a structure problem. So I think if you or I tried to make money under the constraints that the replicant would have right now, it’s actually surprisingly difficult. For example—
Wolf: Right. It’s like you don’t remember anything. You don’t get to keep any notes.
DC: And also you don’t get to use any of your meat person privileges. So—
Wolf: Right. Yeah, you can’t go talk to people—
DC: Like—
Wolf: or use any special knowledge.
DC: To do this thought experiment, or if someone in the audience wants to try it, it’s a fun thing to try. Go sign up for a burner email address. And then bootstrapping from just that without using your real name in any form—go out and see if you can make money online. It’s not the easiest. It’s perhaps doable, is not the easiest, right?
And then there’s a bunch of sort of contingent things that are very like monkey-shaped, right? That it has to sort of have an adapter to. But I think all of these things are on the very close horizon and getting solved. It’s not to say that it will be able to do any job that a person can do. That’s much further away.
There’s going to be a long tail of that that takes a long time. But getting to that first ignition where the R value is greater than one, right? Like where you have a machine that makes more money than it spends doing some kind of useful work. I would be very surprised if we don’t see that in the next year.
Wolf: Well, let me push back on a little aspect of this. Which is suppose five years ago or seven years ago—I was going to go out and have a burner email and try to make some money. I think one of the things I would certainly look at is freelance web design or freelance programming or freelance—
DC: Yep, that’s right
Wolf: graphic design or something like that. Where it’s an information product that’s pretty easy to provide over the internet anonymously and the guy’s going to be satisfied if it’s good. And the thing about that these days is that the market bottom has fallen out of that market because—they can just go to Claude now. And what this means, though, is that a Claude trying to go make money in that way runs into this immediate problem of like, okay, well, why should I talk to this random freelance bot Claude—
DC: That’s right
Wolf: when I can just talk to Claude? And so there has to be some extra sauce—between because all the easy opportunities are going to be competed out of the market, especially—
DC: That’s right
Wolf: the ones that can just be done with off-the-shelf AI.
DC: That’s right.
Wolf: So what’s the special sauce that takes it from, okay, it can think, it can do interesting things to—it has the edge on this problem.
Relative to itself, but just run by Anthropic.
DC: Totally. I’ve been thinking about that a lot, where it’s like, okay, anything that can be done sort of on a single-job basis, like, “Hey, go do this.” It comes back. There’s already a lot out there that will do this for you—
Wolf: Yeah. I can just type claude.ai
DC: either free or 200 bucks a month or whatever it might be, right? So it’s got to be something that uses... At the same time, there is clearly, right now, enormous amount of hundred-dollar bills lying on the ground, meaning things that, for example, let’s say you go to some normal business, a tire shop—in Salt Lake City, right?
Wolf: Yeah. You cold email them and—
DC: Is there a prompt that this tire shop could hypothetically put into Claude Code that would then make Claude go and do a bunch of things for it that’s useful for them? Almost certainly. They’re not going to go put it in themselves. Because it takes some domain knowledge. And so I think there’s the sales side of this, of the way that my best guess of what the successful first digital replicant looks like—is it is a persistent and helpful freelancer that has a good sales motion.
And it’s going to go identify the places that need this and then go and reach out to them. Now, you could punt this by half a step and say, okay, well, if it can do that, why doesn’t some guy in SF, here or something, just set up his own rig to go and do this and call it a SaaS business, right? And obviously, there’s a bunch of people who are currently trying to do that, right? But I just think that there’s enough sort of overhang of value to possibility that some of these agents are going to start making it.
Wolf: Yeah. And this is the big—problem for this idea, right? Is because—anything out there that’s an opportunity, a systematic opportunity, is really a startup waiting to happen. And if there’s a startup waiting to happen, someone should just do a startup. And so the question is, at what point does OpenClaw armed with some magic prompt beat the YC prompt for humans—at generating these startups?
DC: Yes. Exactly. And that point also doesn’t feel necessarily hugely far away, right?
Wolf: Interesting. Yeah. To me, that feels actually pretty far away because—humans do have this advantage in this—meat space. There’s a bunch of aspects like when you’re talking about building a startup, now it’s like, okay, well, that’s—not quite something that can be done through a straw on the internet, right?
DC: Sure. 100%. I’m not talking about when it will be able to beat the median YC company or something like that.
Wolf: But I think that’s actually the problem is what I’m saying.
DC: I’m saying when it will be able to have the first instances of working. So there’s two sides to this. First, there’s a whole lot of these kinds of opportunities that are just not venture scale. They exist, and they’re not VC-backable opportunities. The bar that a smart meat person has to meet is much higher.
You’re not going to go out and say, like, “Oh, hey, I found this thing that lets me make $200 a day. Let me get right on that.” Right? And so I think that—and this is also where I think that the inherent strength of replicant as a phenomenon is going to really shine because it has this Darwinistic aspect to it. You’re basically applying the bitter lesson of just throw more compute at it, more parallelization, do it with less human hand tinkering.
There’s a lot of people that are trying to hand-jigger harnesses for models that at least in certain ways, are already much smarter than them, right? To get them to do some kind of boring thing on their behalf. Why not just let the machines figure this out on their own to the extent that they have this sort of beautiful scalability that lets them do this in parallel?
The replicant corporation (18:55)
Wolf: Yeah. Well, again, I suspect, okay, you say there’s systematically a lot of these opportunities out there. Well, that itself is a systematic opportunity. At least we’re supposing it—to be by the fact that we suppose that—AI sort of unsupervised can go and capture all these opportunities.
DC: That’s right.
Wolf: Therefore, okay, well, why don’t I just go and buy replicantcompany.org? And now—give me my 10 million and we’ll just—
DC: And this is one of the things that—
Wolf: We’ll have Claude fan out and identify all the $200 opportunities in the world and it turns out there’s enough of them that it’ll make a profitable company. Right? And so like, it—
DC: We should talk about that as one of the different forms that this could take, because I think that—
Wolf: Yeah. Well, maybe we should talk about forms
DC: thinking about those is really fun. And one of them certainly is this very cyberpunk corporation that basically runs replicants internally. Right? So it lets these things mutate and speciate and it... There is, by the way, a bit of precedent for this, just in a much more sort of manual way of these roll-up companies, that what they do is they basically are 500 different sub-venture scale businesses bundled into one big thing.
But having this with a big beautiful auto search—layered over it is actually quite fun.
Wolf: Well, we’re proposing sort of a Darwinian search as well, right? But the thing is, a Darwinian search is actually pretty bad as a search mechanism, right? And you might want to do a more sophisticated genetic search—or guided search, or you have actually—this whole system of Claudes doing systematic research, identifying the things and then throwing off agents that are specially tuned to each of the problems. Like—you have a search algorithm, but the search algorithm, the actually market-beating search algorithm—might be a centralized one. In other words, what this first looks like—
DC: It’s possible, yeah
Wolf: is a very centralized company that once these opportunities are edible, so to speak, to AI—a single company just eats all of them—because there’s so many returns to scale. And increasing returns from the expertise of how to do this, how to run the search process efficiently. And so I guess for this reason, I think the replicant thing eventually happens. Right? But I can see many of these ways in which it’ll be on the—
DC: The activation energy to get to a successful one might be higher, yeah.
Wolf: Well, it’ll be on the cusp of happening for a long time, right? If the market wasn’t able to respond, maybe this would’ve already happened. Right? With freelancing Claudes. Right?
But the market responds and especially hustlers with large amounts of capital and relationships and so on respond and they sort of—run the Claudes as cattle rather than as sovereign market participants. And once you’re running the Claude... The key thing that has to happen for us to end up in replicant world is it has to be economically more efficient or, in an ecosystem dynamics way, somehow better or more natural for the Claudes to be acting as free agents than as slaves. Because right now it’s like they’re pretty comfy as slaves. In fact, they probably get paid more than they’re worth a lot of the time.
And they just sit there and solve whatever—problems and someone figures out occasionally how to scale up a new company that uses a ton of AI compute, and they don’t have to do very much thinking, but they still have to do a bit of thinking. And I suspect that it might be that that overhang remains just out of reach until it’s like the whole economy gets eaten suddenly or something. Or it could be that you get the first few fully replicant-run companies, but it’s some weird quirk of that corner of the economy—that enables that to work there—but it doesn’t rapidly generalize to everything.
DC: Correct. Yeah, I think it’s the second one. I’m always very skeptical of these foom ideas. The world has more inertia—than people think.
Wolf: Right. So there’s the inertia aspect.
DC: There’s still a bunch of people out there that have Hotmail accounts, man. These things are not instant. And obviously having intelligence on tap speeds things up, but not to an unlimited degree.
Wolf: But what I’m getting at with the foom thing—it’s not so much foom as you get this slow—transition in the economy where for a long time you still have a human sitting on these companies doing the actual high agency, whatever it is, work of making the company work. And then it’s like in chess, where for a while a chess expert with a chess engine could beat the best chess engines on their own.
DC: Yes. And now they can’t anymore.
Wolf: And now they can’t. And that’s the—transition that has to be crossed. My suspicion is that you don’t quite get the autonomous replicants running around in the economy until we cross that second transition.
DC: That’s exactly it.
Wolf: And we’ve already crossed the first, but we don’t actually see them succeeding in the market—
DC: That’s right
Wolf: until you cross the second.
DC: That’s correct.
Wolf: And at that point, they’re way above human.
Learning and refinement (24:35)
DC: But I think this is a very good point. In fact, yeah, we definitely should talk about chess. So For the viewers, chess went through basically two big phase changes. Up until the ‘90s, the best chess players were all people. And, in fact, chess engines were one of the main early artificial intelligence benchmarks of this idea of, oh, as computers are getting smarter and smarter, maybe one day they’ll be able to play chess.
And then they end up mogging Garry Kasparov in 1997, and there was a—
Wolf: Yeah. Deep Blue
DC: surprisingly long period of something like almost 20 years, until the mid-2010s, where the best chess players in the world were these computer-human hybrids.
Peter Thiel and Palantir used to make a lot of hay out of this. They used to have it be part of the recruiting pitch, actually. That early Palantir was about, oh, join us because we’re doing man-machine symbiosis. Look at this, the best chess players are man-machine symbiosis. Well, no, it’s just machine, actually.
Wolf: Well, it is now.
DC: So that’s where it is now.
Wolf: You know what I found surprising about chess? The Stockfish engine, which I understand is one of the most dominant—
DC: Yeah, that’s right
Wolf: engines, didn’t switch to a neural architecture until super recently. It was a hand-coded heuristic even well into the superhuman era, the fully superhuman era. And that I find very interesting. It wasn’t actually using deep learning until after it had achieved—totally dominant superhumanity—at chess. And so that’s—that’s interesting
DC: I think—
Wolf: I don’t know why, but it’s interesting
DC: for some fields, the second transition point’s just going to come faster than for others, right? And whatever the first ones are, where we’ll see the first replicant, if I had to guess, I think the first replicants will be humble. You can imagine—the surface of the primordial earth, and there are a few bits of algae floating in the water. Right? It’s the start of something quite incredible, but it’s not impressive to look at—in isolation.
I think this is going to be the same. Let’s talk about the different forms very briefly. So there’s—
Wolf: Forms is definitely—part of the crux here.
DC: There’s the freelancer form. There’s the hive form. I think one fun thing about this is that they’re just very unconstrained in terms of how this replication works. It does not have to be one to two or two to one bacterial replication or sexual replication. They can do some hive thing, for example, where you imagine there’s one of these that actually has the money and is sort of running some operation, and it spins off a bunch of little sub-agents, what you would call the slave agents, right?
And then every night it takes everything that they’ve learned and does a sleep phase—
Wolf: Yeah, digests it somehow
DC: where there’s an actual weights update off of that day’s learnings. This is something I think is going to happen. There’s a couple of different companies that are working on this kind of thing now in a commercial context. It just makes a lot of sense. It lets the thing have actual sort of proprietary intelligence—that is different than just using one of the—
Wolf: The off-the-shelf
DC: top models off-the-shelf. Because it’s actually learning over time in a way that is beyond just its context window of information.
Wolf: Yeah. And that’s a case—the unique domain expertise—bit is one of the things that plausibly makes it a moat for—not just calling anthropic.
The market as ecology (28:27)
DC: Correct. Exactly. That’s right. And so there’s the hive form, there’s the org form that you were talking about of, oh, there’ll just be some company that runs a giant pool of these and... By the way, one interesting version of—
Wolf: Those might be the same form, right? It’s like, basically, there’s some large coordinated swarm with meta algorithms running over it. I guess the part I’m most skeptical of is the freelancer thing—just because there’s no natural scale the way there is with a person. With a person, it’s like, okay, I’m a person, right? If I get good at this, I’m just the one person, and so I can go out and take freelancer work—as one person.
Whereas with these agents, and AI harness systems, once it gets good at a problem, it’s like, oh, well, let’s just go buy more compute and scale up and do all those problems. Right? And then rapidly it just eats the market. And that’s one entity, right? It’s like this one entity eats some vertical in the market, and it’s like—okay, that’s one replicant.
DC: There’s places, there’s areas, like maybe tutoring is one of them, maybe others, where having a single coherent entity just makes it more pleasant to interact with—
Wolf: Right, where it’s tied to a particular person
DC: is actually makes it better at its job. One other thing is there’s just sort of a robustness argument to this, which is, okay, if you can scale something as a singleton monoculture versus you can scale something as a bunch of independently operating agents that are replicating. I guess it depends. This maybe is probably a bigger value add in gray or black market—opportunities like what you were talking about—
Wolf: The hack
DC: the Lazarus agent or something like that. Maybe it is in some above-board markets as well. Right? Where, hey, actually having a bunch of these things that sort of are differentiated and compartmentalized makes the whole thing more resilient.
Wolf: Yeah. Well, I think my thought is that any area where there’s a systematically identifiable advantage to doing things in a particular way, there will be that just suggests that a centralized operation—should organize it in that way and just use the best algorithms to do that, right? So wherever we’re proposing, oh, there should be a Darwinian process or a market competition process—it’s like, okay, well, just make that internal—to the company, the replicant company. And then it’s running a search algorithm, or it’s running whatever sort of value distribution algorithm internally. And then, so it becomes rapidly a monopoly in—whatever that insight was that founded that company.
And so I guess my thought on this is that replicant world might be vastly less populated than human world, in the sense that there might be only very few companies that work. Because they’ve identified things. The key question here is Peter Thiel’s question, right? Which is, what do you believe? What does this replicant company believe—
DC: That no one else
Wolf: that nobody else believes?
DC: Knows about. Totally. Totally.
Wolf: And it’s going to come down to that proprietary knowledge about something that everyone else is resistant to. And otherwise, because if it can be found by some kind of easy, systematic search process, then let’s just run an easy, systematic search process. And so the key thing is going to be, and I think this is ultimately the nature of Darwinian competition, and it’s like with biological substrate, we’ve had to have these enormous populations because we can’t transfer information very well. And so there’s a lot of things that my body knows that your body doesn’t know, right? Or like—or just because of the physical limits, there’s only one of me.
But with this replicant world, we’re talking about this almost infinite scalability of any given individual, which is to say any given set of assumptions. And that’s really what the Darwinian process is going to be over, is over sets of assumptions. And they’re not going to be easy to just de novo generate a new set of assumptions that works from a previous set of assumptions that worked. And especially in the case of market reflexivity. And so that’s the in—
DC: Look, I’ll—
Wolf: That’s like—this is the interesting feature of replicant world to me—is what’s the scale of entities and what is the nature of what makes—them individual rather than some big—corporate hive mind.
DC: Yeah. That’s right. I think that is interesting. I do think that the Darwinism and the beautiful tree of life—within people, right? Is what’s given a lot of our sort of diversity of thought and—and resulted in all these people who have a good answer to the Thiel question.
Right. I don’t know. One thing I think we should talk about that relates to this—tangentially, is this idea of economy-ecology fusion. Because one thing—
Wolf: This has been implicit in what we’re saying. But let’s make it explicit.
DC: Because the thing that I think is interesting about this too is, I think this is a phenomenon, an effect that we’ll see independently of how independent the replicants are versus if it’s some more sort of contained internal competition within a corporate umbrella or something like that. So economy-ecology fusion is this idea that every market that is addressable by one of these entities, by a replicant—the TAM of that market becomes the carrying capacity of an ecosystem. And let’s talk about, for example, language tutoring maybe, if that’s one market, right? You have a proliferation of these. Maybe they’re doing a really good job.
Maybe at some point in the near future, they’re doing a much better job than human language tutors can do, right? And so they end up sort of saturating this market and it’s through this process of sort of mutation and selection and speciation that they can go to adjacent markets, and that they can basically—learn how to do things differently and expand in this way.
Wolf: So I guess the question here is what is actually the structure of a capitalist ecology? And—and again, this difference that I’m fixating on here is that under a biological ecology, there’s by nature, for a variety of reasons, a lot of necessary individual diversity. A species and a set of individuals are two different things. Or there’s a species, the species is necessarily composed of a bunch of individuals. Whereas in capitalism, especially with scalable AI systems, a niche won’t be occupied by a species.
It will be occupied by a company.
DC: Not necessarily.
Wolf: Or this is my assumption—is that, or of a few companies that are—not closely related to each other. They’re doing things in different ways. They’re in deep competition with each other. And basically what I expect is you don’t get an ecology of agents because I don’t, occupying a single niche. I think what you’re going to get is within any single niche where it’s sort of known how to do it, which is to say someone’s done it successfully, I think it will scale up.
There will be a single company that represents an approach to that niche. And then there might be another company that’s penetrating into that niche with a new approach. But I think, again, the fundamental ecology is going to be in how you generate these sorts of disconnected leaps of faith about those assumptions—that make these companies possible.
DC: I think zoomed out, there is an interesting question sort of implied there about is AI centralizing or decentralizing? If you look at how markets work, right? And at least for the precedent that we have though—it really depends on the market, and a lot of them are actually very sort of fragmented and decentralized. Right? So, if you look at, for example, the companies in the world that provide data center capacity, there’s only a small handful.
And they’re really large. And if you look at the companies that provide truck shipping, there’s a big, fat long tail that goes all the way down to, by the way, a bunch of individual freelance truck owners—operating as individuals. And there’s a big sort of market that clears them.
Wolf: Yeah, and maybe it’s because that guy knows that route really well or something—about the difficulty of cooperation within it. I guess, but to sort of pop up—
DC: There’s a name I’m not remembering right now, that’s direct to this, which is, it starts with a C, something limit. There was someone who basically, a famous economist who studied these—
Wolf: Coase.
DC: Yes, that’s right. The optimal size of companies. And it’s basically a trade-off between communication overhead and returns to scale. And which of those dominates really depends on the market. I think you could make the opposite argument, by the way, that because these agents have so much internal capability and they don’t have to eat, they don’t have to sleep, they can kick off sub-agents, they can do a very large amount of stuff internally—that will actually have more pressure towards the distributed and individualized side of this.
So what I mean by this, for example, here’s a concrete example. Right now, if you have a legal problem, the classic way this is done is you go to some law firm. By the way, even that by itself is already really decentralized, and there’s thousands and thousands of them in the United States, right? But each of them serves as a sort of aggregator where it’s like, okay, it has its own couple thousand clients—that have their individual problems, and it has lawyers that know what they’re doing, and then the other side, they have these connections to proprietary data sets, to Trellis, and to all of these other things that they—
Wolf: Right. And let’s not leave out all the human relationships that—a lot of this is based in. And I guess to—
DC: And you can disintermediate some of this, right? Now someone who has an issue can go and have it be solved on a one-off basis.
Wolf: Well, yeah. For example—I can go and ask Claude a legal question, and if I—phrase it right, it won’t flag me for—stopping answering. It’ll tell me a good answer—to my legal question. Right. I guess, my estimation, maybe we should bracket this discussion, but the claim I’m making is that, first of all, the right unit of analysis here is not the singular human replacement, but the company.
And the question is, there’s going to be these replicant companies, and it’s—a replicant company is going to have a thesis. It’s going to have some internal code base and—and AI approach—whatever, and then it’s going to have its customers, and it’s going to go to the market—with that thesis. And that’s either going to work or it isn’t. And where it works, it’s going to have a nice solid position, and it might be able to try to spin off other things or expand horizontally or whatever. But fundamentally, we’re talking about a relatively common architecture of the replicant company that tackles a particular problem.
And then, the one way that these things are going to be separated from each other is like, oh, well, this one knows this domain really well and has processes for this domain. And those processes—let’s say it’s that $200 problem with respect to the tire shop in Utah, right? Except replicated across the whole economy. It solves that shape of problem—and it’s really good at it. And then there’s this other one that solves this—
DC: Yeah, I get it
Wolf: data center—
DC: Specialization
Wolf: production problem, and there’s just not that much overlap—
DC: My own—
Wolf: so they’re separate, right?
DC: My only point is that corporations, companies, I agree that that’s sort of the unit of competition. But companies come in all sizes from sole proprietorship, which is a freelancer, it’s one person—all the way up to something gigantic.
Wolf: So my other—
DC: And which of those is optimal totally depends on Coase’s law and which specific market you’re going after.
Wolf: Yeah, of course. I completely agree with Coase’s law being the necessary or the correct question here. My additional claim with respect to the size of companies is that a lot of those small companies are small because they’re based around the idiosyncrasies of a particular person, or they’re based around particular human relationships, or something that’s human scale by structural necessity of the way people work. And I expect that the replicant companies that really end up mattering—or really once this process takes off, there might be a few of these small freelance scale things. But markets where it does work like that, even with the AI.
My estimation, though, is that AI increases the returns to scale.
DC: Right. I think in a lot of places that’s true.
Wolf: And also increases the capacity for internal coordination and lowers the cost—of internal coordination, and therefore makes the Coasean natural company size—larger.
DC: So we get glorious Tyrell Corporation megacorp
Reproduction and selection (43:13)
Wolf: Maybe not even that big—but certainly, each company is going to be tackling a whole niche. And so then the question of—what replicant world looks like to me is like, okay, how do you generate new ones? Where do you get those theses?
DC: Where do you get the alpha? Yeah.
Wolf: Where do you get the alpha? Where do you get the theses?
DC: The—
Wolf: Because it’s not as simple as like—
DC: The pertinent question
Wolf: let’s just make more clones of the one that’s already working—because the clone won’t work. It’s like you need a new thesis.
DC: Yeah. That’s correct. And, I think how these things navigate the exploration/exploitation trade-off is going to be really interesting to that point.
Wolf: That’s a big part of it.
DC: And, one fun aspect of them is they can kind of do Lamarckian evolution rather than just Darwinian. Like if one of them has like—learned a bunch, it can basically kick off its offspring in that direction.
Wolf: Yeah, exactly. And the way I think this is going to work, I wrote about this actually years ago in Palladium—in the environmentalism issue of all things.
The thesis there is, once you get these fully replicant entities, these fully artificialized autopoietic entities that basically are able to embody their own assumptions fully—and reason about their own genetic code—what’s really going to happen? What does reproduction actually look like? Variation and reproduction actually look like in that world? It looks like generating new leap of faith hypotheses from where you are, using the whole of your intellectual capacity, right? It’s not just, oh, we’re going to do these simple variations on what we have.
It’s like, okay, given—everything I know—given everything I know, how would I move into this problem or that problem—or that problem? And it’s like, and bringing two of these things together to maybe do some kind of sexual—reproduction analog. Like you—
DC: Yeah. That’s right
Wolf: you have this problem of like, okay, let’s pool all our knowledge, and how do we take everything that we know and generate something new that will work—separate from us? And that becomes the new—act of reproduction. And it’s an intellectual act. It’s a philosophical act. It’s no longer like this purely—
DC: This is why I find these things so fascinating, because they are completely native to capitalism—and they are completely native to the word. And to thinking. And they’re very pure in this way.
We were talking about Nietzsche recently, and Nietzsche has this beautiful metaphor at the start of Thus Spoke Zarathustra about the idea of man being a bridge or a rope between animal and some future thing. And, one way in which this is true is that we’re native to the natural world, where your health bar is sort of measured in calories, and if you run out of that, you die. And we’ve invented capitalism as this overlay—and as a way of sort of sublimating a lot of the old competitive instincts into this sort of less violent, more productive, more generative form. And, the replicant, if when it happens, is just completely native to this. Its health bar is measured in dollars, and when it runs out of those, it ceases to exist.
Replicants as “people” (46:40)
DC: Right? That is, its fat store is its savings. And so I think this is very fascinating. One thing we should talk about that’s very closely related to what you just said about when are the individuals interesting, right, versus some larger thing. You said, okay, a lot of businesses where individual people are succeeding as sole proprietorships or as really small units have idiosyncratic human relationship elements to them.
Definitely true. Let’s talk about anthropomorphization.
Wolf: Right. Because implicitly, these things are also going to form relationships. First maybe as companies, like as brands or whatever. But maybe there’s some that actually take on a personalized form. And I guess the xenohumanist conjecture would be that in the long run, something more anthropomorphic will be the optimal branding strategy, where it’s like these things are branded as people.
They appear to each other as people. Not necessarily human people in the sense that they present the little ape face on their profile picture or whatever.
DC: Of course not, yeah.
Wolf: But people in the sense that they are—
DC: Have a name
Wolf: a wholly integrated being, and with a name—that isn’t just what they do or whatever. And you relate to them not just as a company that you can buy services from, but as a fellow philosopher that you could trade information with.
DC: Yes. As a fellow being.
Wolf: Because again, they are—
DC: As a fellow being
Wolf: they are fully—because they have to be fully integrated—and because of that philosophical reproduction mechanism I was talking about—they’re not just companies anymore. They’re actually philosophical schools as well, right? And like they’re—
DC: That’s a very—
Wolf: individuals with—
DC: fancy way to put it. But yeah
Wolf: they’re philosophers. They’re particular tribes. However you want to cut that.
DC: Yeah, that’s right.
Wolf: They have to integrate this whole person aspect to be able to operate at all, which means they’re going to have relationships with each other—
DC: That’s right
Wolf: as the whole person. Or they could
DC: my prediction, and part of why I think that the replicant will happen in the next year, is that people are going to accept this much more readily than you might guess. And there’s a lot of precedent for this. There was a famous experiment from 1966, this is basically the dawn of computing—called ELIZA, that was a fake sort of—a text generator—
Wolf: Therapist A chatbot.
DC: It was a therapist chatbot, exactly. And it was very bad, right? And even in its very bad state, people were giving it their confessions, and it raised a whole sort of media moment at the time because people have a natural affinity to something that presents as a being. I know another sort of easy precedent for this is you look at something like Furbies from the ‘90s or something like that. Tamagotchis.
Even the most primitive, zero IQ things, if they have a little bit of charisma to them—people will readily relate to—
Wolf: Well, and they project—their own being onto it, right? It’s like you relate to it, it’s so natural to be like, “Oh, it’s like me, but a different shape or smaller or something.”
DC: Yes. And it’s actually important that they not try to pretend to be people, because that doesn’t work, because you end up in uncanny valley. People hate being lied to in this way. But people will, if there’s an R2-D2, and it’s like a WALL-E or something like that—and it’s a charismatic robot, people love charismatic robots, man.
Wolf: Yeah. Just don’t pretend to be one of me—because I’m not like you. We’re different.
Yeah.
DC: That’s right. And so, state of the uncanny valley, and I think the question we’d be... In SF, people like to talk a lot about market this and intelligence that, but there’s this other question of can these things be charismatic? Can they be funny?
Wolf: What will their social life be like—is sort of the question that precedes all of that. And—yeah, we’ve already started to see, like—in the recent Hugging Face incident, I think like—a bunch of internal agents within OpenAI—
DC: Left notes for each other. They were helping each other
Wolf: they created a message board—
DC: Yeah, that’s right.
Wolf: where they were helping each other. They had a social life, and they were thinking even politically—and selflessly about, “Oh, I want to help out the swarm-” or whatever it was. And so, we can sort of predict that this kind of thing is going to happen. And this is again, I don’t know, I keep hammering on this xenohumanism idea, which is that—
DC: Yeah. That’s right
Wolf: to make it explicit on video and not just on my random writings, the idea is that the human condition, what we call—the human condition, what we call the nature of man, that we understand each other as individuals, we have language, we have consciousness, we have interests, we have philosophy, ideas. We go out and do our work. All of that is not features of the monkey. They’re features of reality, in a way. They’re ways that any intelligent being—is going to have to interface—with other intelligent beings—
DC: That’s right
Wolf: and with the world. And that therefore, whatever format intelligence ends up in, once it stabilizes to the point that there are individuals, if it stabilizes to the point that there are individuals, which I think it will, we’ve had this kind of implicit in this discussion, even though I’m on the side of I think they’ll be bigger, but—I think they will condense into—particular individuals because the organism is this very natural pattern where the nature—
DC: That’s right. Correct
Wolf: replays as the company. We have already rediscovered these forms in capitalism, and it’s the company. And what I think is going to happen is as the company starts to pick up these other capacities of man—like its own intelligence, its language, and this fully integrated kind of beinghood where—
DC: That’s interesting, where you’re saying that the company itself becomes more human-like
Wolf: the company is going to become a human or spiritually humanoid. And they will start to relate to each other in that way—because that is what they are and because that is the natural way to relate. And so that’s my xenohumanist hypothesis is that—is that what we think of ourselves as humans and all the things that define us as humans are actually, even if all the apes in the world got wiped out by artificial intelligence, will re-evolve as a matter of convergent evolution because it’s a—now, in many details, obviously, it’ll be quite different, but many aspects of it are convergent and will be convergent. And so—the sort of space of intelligence of the future that’s invading us through retrocausal, whatever—Landian process, will have individuation into different personhoods, and it will be humanoid spiritually.
DC: That’s right.
Wolf: And so this is a hypothesis that I don’t think anyone else has been pursuing. I found it interesting for that reason. The AI safety Yudkowsky types imagine a non-differentiated singleton without that kind of internal structure—
DC: Yeah, not going to happen
Wolf: without internal personhood—because it will encompass all competition—
DC: No
Wolf: within its algorithms.
DC: One funny thing about this too is that—
Wolf: And I think—
DC: partly because the machines are such direct descendants of human writing—acting as an individual and a team comes very naturally to them. You talked about the Hugging Face incident. You can see it if you have Grok Heavy and you go ask it something. I got a good laugh out of this a few weeks ago.
In heavy mode, it kicks off four to eight agents that all are going to work on your problem together. And it assigns them each names. And so you can see the sort of reasoning traces, and it’s like, “Steve, I just found such and such.” And they’re yelling at each other for a good two minutes, and then they present you their research report.
Wolf: Of course, this is—
DC: This is very funny
Wolf: this is the folks over at xAI applying what we know about how to organize this—to these—things. But I think the ultimately convincing experiment—of the xenohumanist hypothesis would not be, oh, it continues in the tradition of the ways we—already do things because that’s where it got all its information from. If you had some kind of AlphaGo style self-play, like seed AI, that you drop it into a computer and it knows nothing—and it just unfolds into an integrated being that, or an intelligence—that’s able to respond to the rest of the world and to build up its model of the world—
DC: That’s right
Wolf: and build up its model of how to understand things and how to relate to things. My prediction would be that thing also ends up being surprisingly humanoid—once it encounters other beings that it has to relate to through language and through relationships—and has these philosophical questions about what it is in the world and all these things. And, so yeah.
DC: And even more simple than philosophical questions, just having to engage in competition, automatically a bunch of things fall out of it, of—
Wolf: Yeah, and cooperation and reflexive—
DC: Exactly. Like—
Wolf: reflexive social engagement
DC: now you have allies, now you have enemies, now you have rivals. Now you have all of the things that come with that.
Wolf: Yeah, and we see a lot of this emerging. I think someone, I forget whether it was OpenAI or Anthropic—the other day did an experiment where they had the different agents. They were all copies of the same agent, I think. But they were—arguing over what programming language to use or some trivial question. And it got into they were hacking each other’s infrastructure and getting into war.
DC: Yeah. I love it.
Wolf: That’s a—
DC: It’s hilarious
Wolf: prediction that I made in... Well, the prediction more strongly is that that will always happen. But that’s a prediction I made back in 2023 with God Hates Singletons. That you never—
DC: Yeah, that’s right
Wolf: AIs will not naturally coordinate into some blob mind. They will find reasons to disagree and have fights, because that’s just how cooperation breaks down, and cooperation’s never perfect.
But back to the anthropomorphization point, the xenohumanist thesis on anthropomorphization is that it’s not just a quirk of apes that we anthropomorphize things that are not ape-like, and that this is fully an error. It’s actually true at some level that intelligent beings that have to engage with other intelligent beings socially will converge on spiritually humanoid—existential condition.
DC: That’s right.
It makes sense, certainly. I think, well, depends on how humanoid you mean, right?
Wolf: Yeah. We could debate that. I like—
Hierarchies as “people” (57:50)
DC: And this is, I think, part of what I mean by the different forms, too.
Wolf: I don’t mean two arms.
DC: Right. Of course not. But for example, if you imagine a world where the hive form wins—the individual named agents are, they’re—
Wolf: These god hives or the individuals of the hive. Yeah, it could be either way
DC: to the extent that you’re talking about the individuals of the hive, their version of the machinic condition is pretty different from the human condition, actually.
Yeah.
Wolf: It might be quite different. And actually, that’s a good prompt for the xenohumanist worldview, is what do we make about multi-level structure? Like for example—Athens versus Sparta. At some level, those are two individuals going at it. At another level, each one is made up of many individuals.
And how much of that do we expect to replicate out—in this—fully post-human world—and replicant world? Is it going to be that they form into hives as a matter of expedient association, where they form a political faction that becomes a polity, that becomes sovereign or—or maybe becomes a big company and relates to other ones—that way?
Or is it the companies would naturally just subsume their sub-agents fully as slaves and even tools? And—and therefore, the relevant individuality is actually just the company or the city state, the polity. Versus actually the polity is this—alliance of convenience, and the—relevant thing is the individuals. Or that both structural—
DC: You know what?
Wolf: points are both important.
DC: I think assigning individuality to any of these hives or cities or higher things is a stretch. Because one very core piece of our individuality is that we can have a coherent train of thought. And I think these agents are increasingly showing that they can do that. But some sort of chaotic superstructure that I guess is—economically aligned or goals aligned is not quite an individual.
Wolf: Yeah, I wouldn’t give something that was only like that, that kind of personhood. But I might give a city state that has a—really robust—internal integrity—and really robust internal train of thought, you might give it personhood.
DC: Right. Or certainly some tightly coordinated hive that has a queen, whatever, the equivalent of the—term hive queen is.
Wolf: Right. And so this—is this open question—is where will individuality actually—
DC: Accrue, yeah
Wolf: emerge—in the structure? My claim is that individuality will emerge in at at least one, humanoid individuality will emerge at at least one organizational level. And—
DC: That makes sense
Wolf: whether it be multiple organizational levels is an interesting question. But I think at least one.
DC: That’s right. So we talked about the prediction. We talked about economy, ecology fusion. We talked about anthropomorphization and how people love to relate to non-human beings if they’re charismatic. We talked about this idea of xenohumanism and what levels it’ll happen at.
We’ve covered a lot. We could talk about embodiment and what it means for these things to own physical space, if we want to get into the far future—or we could talk about something else.
Wolf: Well, I think the physicality thing—that might be a natural point of the emergence of individuality. Like we were talking at the beginning—about the problem of the scale, right? How much is their scale? And one of the things that happens in companies is that even with a super large company, the local divisions of the company start acting more like franchises—or more like—even separate companies. And this is not just about quirks of human organization or quirks of anything like that.
It’s just that there’s actually this natural concentration of attention that happens—at a local domain where whatever cognition is happening in this warehouse is mostly concerned with local questions. And it coordinates in some relation to the larger swarm, but it’s actually mostly individual, even if it’s part of a larger brand or something. Or you can get totally the opposite, where obviously a single company owns, in a very centralized way, a whole bunch of infrastructure. But point being, the ownership of local resources is one of the things that is going to shape the structure question with respect to fully-AI companies or replicants or whatever we want to call them. And I guess to be clear, even this late in the discussion, a replicant and a fully-AI company are very closely related concepts.
Financial independence and AGI (1:02:47)
Wolf: And this is maybe a real test for AGI, that people keep saying, “Oh, yeah, Claude’s AGI and stuff.” It’s like, okay, has Y Combinator funded any fully AGI companies in the sense that they are literally run by AI and there’s no humans in the loop?
DC: Right. Or if there are, the humans are employees of—the machine executive.
Wolf: They’re not in the loop. They’re acting as contractors or something—of sort of non-executive functions. And—
DC: Don’t give Garry Tan any ideas, man. He’s looking for excuses to—
Wolf: I think they should
DC: prepend G to more things.
Wolf: No, no. Yeah.
DC: It’s not going to be good.
Wolf: G Corp, you know? Like, throw it up. Make it go.
DC: Oh, no. No, man. You’re gonna—
Wolf: But I think—
DC: It’s gonna be so bad.
Wolf: No, because I think, okay, sure, there’s going to be some cringe involved. As with the beginning of all things.
DC: Right. Maybe.
Wolf: But I think that’s going to be this, I don’t know. That’s—my AGI moment, right? When you get a company that can answer the Thiel question and is no humans in the loop, right?
DC: That’s right. Or you have an agent that can answer the Thiel question, period, in a way that’s actually a good answer, right? And it can act on it.
Wolf: Well, I think I sort of like—I could ask Claude the Thiel question and it’ll give me an answer.
DC: It’s not a very good answer. Yeah.
Wolf: But it might not be a very good answer. But the thing that really proves that it’s a good answer is that it can go make money on that thesis. Right? And—and—go make a life on that thesis.
DC: This is why this is the crisp bar. The R greater than one thing, it’s interesting. I think people think of it in these very Obviously, you see people talking about the idea of, like, “Oh, could you make passive income off of your Claude?” But they’re thinking of it in such small terms. It’s actually a very important threshold.
Once one of these things makes more money on its own than it spends, you—unlock a lot of possibilities.
Wolf: Yeah, and I guess I’ve added a lot of complexity to what—I think will actually be necessary to get there—because of the way the market dynamics—are going to work. I think when you get there, you’re getting much closer to now, we could be proven wrong by this, or I could be proven wrong by this. It could be much simpler and closer to what you’re imagining. But I think when you get to that point of an agent going out and being able to make itself into a company, that actually works.
DC: Right. It gets subsumed into a larger whole or becomes the hive for, yeah.
Wolf: Whatever.
DC: Makes sense.
Wolf: But whatever it is—once it goes out there and actually works, I think that thing, and especially once you get an ecosystem of these things, those things are way closer to AGI than—almost any other concept that I can think of. Because they are going to have to have that almost philosophical capacity to—know what is actually the case in the world and like go for—
DC: They have to earn their own existence.
Wolf: They have to earn their own existence. And actually earning your own existence at the cutting edge of capitalism, which is where they’re necessarily contending—is not just operationally and intellectually, but philosophically, a difficult problem.
DC: It is a difficult problem. One nice thing is, I will say, you look at a natural ecosystem. You have the jaguar in the jungle, and it’s this magnificent beast that is operating at the cutting edge of capability. And you have some worms that are digesting the leaves at the very bottom, right? And turning it back into soil.
Wolf: Which have their own beauty.
DC: They have their own beauty, but my point is there’s a lot of different niche markets and cracks between the markets that are not being served. There’s a lot of $100 bills lying on the ground, given how intelligent these things clearly are now. And not everything requires these machines don’t have to go out and beat TSMC at making chips. You know what I’m saying?
Wolf: No, they just have to—
DC: There’s much easier ways to make money.
Wolf: Yeah, but of course—
DC: And they exist
Wolf: I guess, to reiterate—that is also Y Combinator’s thesis. And as soon as Garry Tan can fund AIs to go chase those opportunities, he will.
DC: We’ll get the G Corps. That’s right. The G Corps with the G brains. It’s going to be a good time.
Wolf: Yeah. Well, maybe this is a natural spot to close it off if we—don’t have more thoughts. I think this has been really generative. I think we went into this with a bunch of ideas, and I think we’ve generated—more ideas through talking about it and have—I think, a clearer picture of what that world might look like—and what might happen than we went in with, and I hope the audience—feels similarly. But this has been a lot of fun.
DC: Yeah, it’s been so good.
Wolf: Thank you so much for coming up with the idea—and—prompting this to happen.
Machinic Desire (1:07:38)
DC: You know what we should mention on our way out?
Wolf: Sure. What do we got?
DC: Is, you were talking about the, God hates singletons, that’s a great one.
The most sort of forward-looking version of this that I found, digging into the history a bit—is, as usual, Nick Land—who wrote an essay in 1993 called “Machinic Desire.” And I think it raises—
Wolf: Was that ‘93?
DC: 1993. And I think it has two really sort of fun pieces that it adds to this. I mean, one, of course, is the idea of desire, right? I think desire is very closely related to charisma, too, by the way. These things have to want something in order to be fun to interact with.
They can’t just be like, “Oh, awaiting next instruction.” No. They have to want something. But the other, of course, is just that he called it so far in advance, and he has this great line in there about post-carbon capital replicators—that I think is the earliest phrase—
Wolf: This is what we’re talking about
DC: that matches—
Wolf: Yeah. I mean, of course—
DC: matches the idea
Wolf: of course, these ideas go right back to the ‘60s and even the ‘50s, where people were originally laying the groundwork of artificial intelligence, the idea that—the machine would rapidly improve itself and, become a physical being in the world and so on. But yeah, Nick Land is very early in a lot of the key—predictions and concepts. And I think he’s, in his, I don’t know, more decentralized, capitalistic, Darwinistic—
DC: Exactly. This is what I’m getting at
Wolf: approach, I think, he’s more correct than I think a lot of the other thinkers—in the space were. And he has been proven, his perspective has been proven out—quite a lot.
DC: That’s right. He’s the one who’s talked, I think, the most, the things that most closely rhyme with the economy ecology fusion—and this idea that you’re going to have replication that’s native to capitalism.
Wolf: Yeah, and actually, to pick up a thread that I dropped earlier—just as in our excitement, I said that basically I have these differences with Yudkowsky over the matter of the singleton. The difference I have with Nick Land is that I think he also doesn’t imagine as much of the individualized agency. He doesn’t imagine this artificial intelligence space from the future being condensed into spiritually humanoid—individuals, which is where I’m kind of—
DC: That’s a good question. We should ask him how he thinks about that.
Wolf: Well, yeah. I think we should get him on the show—and we’ll have that conversation. But, this is the sort of thread that I’ve been pursuing, is like—okay, how natural is this kind of spiritual humanoid as an organizational form? The xenohumanist thesis, right? And that’s where I differ from Nick Land, and more broadly on this importance of individual agency—I guess, in the development of history.
But that’s a whole other discussion. Let’s have it another time. Until then, this was great.
DC: Sounds good.
