(In which we discuss the Telocracy essay Joe recently published in Pax Machina)
Ivan Vendrov: Welcome to the show, Joe.
Joe Edelman: Thank you. I’m so excited that you have a podcast now.
Ivan Vendrov: Yeah, or whatever this is. No, I’m really excited to talk to you because I just feel like there’s no positive visions for what to do with all the GPUs and all the people, and we’re going through the largest social transition since the invention of the printing press, maybe since the invention of writing, and all the ideas we have are either something horrifying or we’ll all live forever in infinite abundance luxury space communism.
And you wrote, I think, maybe the best, or certainly the most complete vision of where we should go from here, which is called telocracy. People can look it up online. I just want to deep dive into what is telocracy, how do we get from here to there, how does it relate to all the other things going on. First, what is telocracy?
Joe Edelman: Somebody explained it back to me really simply just yesterday, so I have a new way to pitch it that I’ll try.
Well, actually, maybe I’ll start: the default path seems to be that AI will do what is good for large corporations and possibly governments or militaries. And that’s a pretty small amount of what’s good overall.
Maybe you can add individuals to that as well, if you think about each of us accelerated by our chatbots—although there’s transaction cost reasons, there’s various reasons. I think I’ve come to believe that the cyberpunk future of zaibatsu, giant corporations, megacorps, and so on—this future actually comes from transaction cost economics. It comes from recognizing that so far, information technology has been really good for scaling hierarchical organizations. If you think about how many agents are working for Anthropic, for instance, right now, it seems to be very much continuing the trend: the large corporations become incredibly large corporations.
I think this is also some of the fear that people have about governments, is that, for instance, it’s impractical to monitor all violations of most laws. So really the police wait for complaints and things like that. But people are scared that with AI, government agencies can grow to the point where they can just monitor everything and the government is all-powerful, which makes it extra important that the laws are correct, that they’re doing the right things, et cetera. So really these two fears about megacorps and about government overreach are the same fear, which is about existing hierarchical structures being amplified by information technology in general, because I think that’s already been the case, right? The NSA benefits a lot from the internet, for instance.
So even if we as individuals become a little more powerful because of our agents, if at the same time the main event is happening at these larger scales, that’s not so good. And yet, what else is there is the problem. What other interests are there besides the ones that I just enumerated?
And the truth is, well, there’s lots of other interests. Everybody has clubs that they’re part of, maybe church groups that they’re part of, weird projects they want to start—some weird intellectual project or creative project or whatever. And I think intuitively many of us know that it would be way better to pour the power of AI into all of that, right? It would also be much more popular. People would be like, “Sure, if those data centers exist not to amplify the government or big business, but to amplify my band’s output, or my—”
Ivan Vendrov: My church’s output.
Joe Edelman: Yeah, that kind of success or amplification or superintelligence or whatever you want to call it would be much more popular and relatable, and we could see it all around us, and it wouldn’t be some abstract capitalist increase or something, but more the flourishing of all the things that we already care about.
So telocracy is about how to actually pull that off: how to change the incentive design and the way that compute is allocated and the way that things are monitored to pull that off so that you can actually tell. It’s a hard problem, because as soon as you—let’s say you just gave compute, you portioned it out to groups or whatever. Well, people with money could just buy it from them, or they might actually turn out to be groups that fight each other, like the political polarization that we see these days in a lot of countries. So it ends up being a complicated problem how to organize society to pour the AI into these kinds of purposes.
Ivan Vendrov: Well, maybe this is a good time to ask this, as I was going to wait longer, but didn’t we used to have telocracy, and then basically we militarized and then never demilitarized after World War II? The reason big corporations and the military and the government apparatus is so powerful is they explicitly tried to essentially do a total mobilization of all civil society. What Tocqueville describes in America in the 1830s is much closer to telocracy than what we have today. And don’t you basically have to take on the military-industrial complex?
Joe Edelman: No, I don’t think so. I guess we have different explanations for why civil society collapsed. I was talking about this with somebody a few days ago, and I gave three reasons why civil society collapsed. Let’s see if I can remember them all.
So one is that civil society organizations tend to be less dynamic than capitalist organizations. So if you want to provision healthcare or housing or whatever it is, insurance or whatever, market organizations tend to be fairly dynamic and they tend to react to changing structures in demand, changing technological conditions, new restructurings that make costs cheaper, make it possible to reach many more people, or whatever. And so civil society organizations just don’t do that as much, and for that reason, they’re less to be trusted with important things like healthcare and housing and so on. So I think that’s probably one of the main reasons.
Another reason is that there are cases where, at least ideologically, we thought that the government was a fairer or better way to do things than private associations. One reason is because with the governments, you can often specify what the level playing field is or what the minimal—you want to get everybody above the poverty line or something like that. Associations tend to be local and have different overlapping polities or whatever, and so you might—at least that’s a reason why I think there’s been a push in general away from associational ways of dealing, especially with social needs or whatever, towards government provision.
Ivan Vendrov: Right, the example that comes to mind there is school districts in the United States, where rich people will cluster together in a rich neighborhood and then they’ll fund a really good school. And in some ways that’s a very natural thing to happen, but one of the justifications for government intervention in education is that, well, we want an even playing field; every child should receive a good education. And that’s an example where voluntary associationalism failed.
Joe Edelman: And then the third one is just atomization, the same thing that happened due to urbanization as people moved away from their families and so on. It becomes much harder to form these associations. And to some extent market provision just displaced—if you can order DoorDash or whatever, there’s less of a reason to cook with friends, et cetera, especially if you just moved to the city and you’re no longer surrounded by your church congregation or your family or whatever, then there’s just some displacement away from associations that way. So I see it as those three causes rather than this post-World War II cause that you gave. And I think there’s data to back up the attribution to those three causes rather than the one you suggested.
Ivan Vendrov: Well, I guess on the first point, which maybe is the deepest one, capitalist organizations or market organizations just seem more dynamic and more adaptable. If that’s true, then where does the hope for telocracy lie? Won’t that continue being the case?
Joe Edelman: No, I think there are some cases where a nonprofit has been dynamic in that way because it was really clear about what its mission or purpose was, rather than just continuing with its mode of operation. The World Wildlife Fund is one example. They used to be a bunch of lawyers that would stop parks from closing, but their top-line purpose was to stop animals from going extinct, different species. And so at some point they were like, “Well, actually buying land is much more effective than stopping parks from closing.” And they switched, and they fired their lawyers, and they hired a lot of hedge fund people that do this cool GIS stuff where they’re like, “Oh, there’s more owls over here, but this land is worth more, so we’re going to sell this and buy this,” or whatever. And this has been way more effective.
What you’d want is some way of pressuring—not necessarily for-profit; telocratic organizations can be for-profit, I leave that open in the essay—but even if they’re nonprofit, you would want some way that they’re held to their mission in a way where you have to make those pivots, where you’re in some kind of breach if you keep doing the lawyer thing when the hedge fund thing is actually what would be required to pursue your mission.
Ivan Vendrov: That brings me to another question I had, which is: what is the enforcement mechanism here? Briefly, I guess telocratic organizations have this really articulated vision of what we are about, what we are doing and what we are not doing, what is in scope, what is not in scope, what are the trade-offs we’re willing to make—which virtually no organization currently has. A C-corp literally just has basically the interests of shareholders or whatever the owners believe, and that doesn’t have to be articulated in any way. Telocratic organizations have this deep articulation, but how would that be enforceable?
Joe Edelman: Yeah, there’s different mechanisms that I suggest in the essay. I think for prototyping it, the most useful thing is to build pools of assets that automatically reallocate themselves based on whether their current holder is living up to purpose. And the most obvious one there is compute.
So you can have a compute cloud—it looks like even current compute providers could give a fraction of their cloud to this kind of cause. And there’s agents that are auditing the activities of the organization according to its purpose, and those agents can flag some problem, like the one we were just talking about with the lawyers versus the hedge fund people or whatever, and then the organization would have a certain amount of time to respond, or its compute allocation would change, for instance. And there’s many other assets you can attach.
Ivan Vendrov: Well, so “would change”—this must ground out in lawsuits on the existing legal system at some level, right?
Joe Edelman: No. To give a really concrete example, there’s already a first telocratic org, which is really exciting, and it’s a network of AI labs in Britain called Bold that are trying to be an academic alternative to the big labs. They have a data center in Britain that’s really sizable, money from private investors and from the British government, and they have certain research programs where they’re trying to beat the mainstream AI research directions with their own directions. And this is different than how academic ML research has worked in the past, where usually each grad student really has their own research program; they’re trying to do something much more directed.
And so the way it works now is that when you’re a grad student and you decide to join this initiative, you get access to this giant cloud, but you have to post your project ideas and your research direction, and then it’s compared to the mandate that this cloud is allocated for, and your GPU allocation is going to depend on whether you’re in line with the broader purposes of the network. Currently they’re not doing this auditing thing, but that would be the next step: you can just look at what they’re actually running on their GPUs and make sure that it continues to be in line.
Ivan Vendrov: I see. So in this case, the actual governance layer is presumably some board of directors, and in some sense they don’t have to articulate their values, but they’re choosing to abide by this articulated values constitution, and they’ve implemented an enforcement layer that is automated. Does that sound right?
Joe Edelman: Sure, yeah, or in the general case, I think it’s just contracts. The contract says instead of “you’ve paid this much and you get this much of our cloud,” it says maybe “you get this much of our cloud for free,” or something, “but you put up with this auditing, and we revoke it if the following conditions aren’t met.”
Ivan Vendrov: That makes sense. That makes me more optimistic, because there’s a lot more ways that this can interface with the existing legal structure. In some sense, I could start a C-corp and run it as a telocratic organization, basically.
Joe Edelman: For sure, yeah.
Ivan Vendrov: And presumably there are a bunch of companies that already run this way—they just happen to have leaders that are really good at articulating their values and are very strict in how closely they adhere to those values. I’m curious if any come to mind, just to give more examples or intuitions.
Joe Edelman: I think the big step is the agent auditing, and I think this is actually pretty rare because agents have just recently become barely capable of making these “is it within the purpose, is it not within the purpose” kinds of decisions.
So there’s a lot of purpose-driven organizations in the world—the Red Cross—and I think one of the things that telocracy needs to prove out within the next few months, even, ideally, is that we can transition from believing that those organizations are using resources in the best way for their purpose or whatever, to knowing, through more extensive auditing and things like that. And then that’s great: if the organizations are already doing the right things, then maybe we can even just increase the amount of resources available to them.
Agent Auditing as Scaling Common Law
Ivan Vendrov: I guess I’m not sure I understand what the essential thing about AI agents is here in this picture. Because in principle, if telocracy is a good idea, then we should be seeing organizations that just happen to devote 5% of their resources to doing human auditing of their purposes, and that should already be working. And yes, AI agents make it cheaper—it’s actually not obvious to me that AI agents make it differentially cheaper to audit than to fool the auditors or something. It seems zero-sum to me, roughly. Why do you think—
Joe Edelman: Wait, to audit versus to what?
Ivan Vendrov: To fool the auditors.
Joe Edelman: Ah. Well, I think in general, human judgment has been very expensive in the domain of mechanism design—the kind of thing that a judge does, where they use a body of principles or values and they issue some reasoned opinion. This, I think, has been very expensive, and that’s one of the reasons why we use voting and markets and stakeholders who represent the values, but in a vague, fuzzy way and with a lot of danger for capture and so on. All of the mechanisms we use are often hacks because it’s too expensive to do this much case-by-case reasoning.
So I think there’s evidence that AI already changes this. One of them is just the Instagram recommender, which is currently a generative recommender. I don’t know if you’ve used it, but if you’re on the Reels tab, you can click “Your algorithm,” and then you can talk to the recommender now. And the recommender works now by watching the videos and doing case-by-case reasoning about what you would like to see, which is very, very different than how it worked a year ago, where it showed you things that your friends liked or that people like you liked or whatever, which is a much cheaper computation to run.
And it would have been obviously unaffordable to pay human people to watch every video for every person and think, would Ivan like this? That would be one person for every person. So I think there’s pretty good evidence.
Ivan Vendrov: Kind of like the Stasi.
I just had a reframe of telocracy which I’d love to get you to criticize, which is there’s something about the reason why Anglosphere governance works so well is because of common law. We owe a lot of the good things in our civilization to common law, which is precisely this: a careful person reads the law and looks at all the information and makes a reasoned decision that is public and transparent. And you’re like, that’s just been very, very expensive, and we mostly just used it in high affairs of state or murder trials or something. And AI agents let us do this a thousand times cheaper, so now a lot more of economy and society can have the good properties of common law.
Joe Edelman: Yeah, I think that’s right, and I think that’s a big part of my intuition that there will be so much institutional change. Because obviously if it’s gone from being a thing that a judge does and that you and I don’t know about hardly at all, to a thing that’s built into the recommendation system for the videos we watch or whatever, that’s a very big shift.
Ivan Vendrov: Most people don’t really interact with the legal system outside of watching procedural dramas or something. And it’s actually every second there’s a micro judge or computer deciding little things that are happening, and you can engage with that reasoning and see the ways in which it does and does not fit with your purposes. That’s really cool.
Maybe let’s dive into this social media example, because it goes through—you’ve talked about it here, you also talk about it in the essay. So you have this example of a social media platform starts out small, but then people start relying on it, so let’s say it starts mediating young people’s friendships. And then you’re like, well, under this concept of living commitments, you then can’t pivot the platform to mine that relationship for engagement and ad revenue. You now have to take on different values of nourishing friendships, or you have to give up control to other organizations that are willing to do the nourishing friendships thing.
And I’m like, that sounds awesome, and I’m also confused about who’s enforcing this, how could this be enforced, why would an organization ever give up its power? Seems like there’d be a lot of conflict there. And also how you adjudicate these really nuanced aspects of human relationships.
Joe Edelman: I can talk about all those. So first of all, about enforcement, there’s going to be different phases to the rollout of telocracy. The first phase is going to be very optional. So it’s going to be the same groups that do maybe public benefit corps and things like that now, which can be high profile, but they chose to do that. And so they’ll often be tying their hands to the mast or whatever. They’ll be doing that because they think it’s a good idea for their own values, the founders, etc. And then they’ll enter into different kinds of contracts that do that. So they’ll choose to get telocratic compute that disappears if they don’t follow their purpose, or they’ll choose to make their brand telocratic so that they’re no longer allowed to use their brand if they stray from their purpose or whatever. And that will prove out that it works in these cases.
And then there’s a phase where there’ll be enough of these that society is starting to figure out which things should maybe always be telocratic organizations. So maybe nursing homes should not be allowed to be for-profit organizations that are not telocratic. There’ll probably be some high-profile things that go wrong. Some autonomous corporation that was originally going to do this thing, because all the agents decide they’ll make more money doing this other really terrible thing or whatever, they all do that thing, and then there’ll be some maybe law or just maybe common sense that if you’re setting up one of those things—a nursing home or an autonomous corporation above a certain size or something—it needs to be bound to purpose.
And then the third stage is there’ll be courts and things like that, there’ll be structures in society that try to enforce that new sense. So if there’s an old nursing home that wasn’t telocratic or whatever, there’ll be some law that it has to migrate over or something like that. So I see all that taking, I don’t know, at least a decade or something, quite a while. And the enforcement mechanism will be a little different in all three of those cases.
And then I think that was your first two questions, and what was the third again?
Ivan Vendrov: The third was it just seems really, this operation that you’re imagining of a social media facilitates a relationship between people and then the company gives control up to another company, and I’m like, this just seems really subtle. It’s very hard to imagine even a current-day judge making a decision of this nature: “Ah yes, this relationship, we’ve looked at your chat logs and based on that we really think this is facilitating a friendship, and the way you guys are optimizing for engagement is inimical to our concept of friendship.” And I’m like, man, we just don’t have the philosophy for that kind of thing as a society.
Joe Edelman: So the idea is that you’re right that we don’t have the philosophy for it, and that’s some of the work ahead even for me. One way to do it, just to give you an intuition about the possibility or plausibility of this, is you could just ask people what they rely on, what they use the platform for, or what has happened for them that’s good or bad on the platform, and you’ll end up with things like, “I met my friend Sally,” or whatever.
And then you can ask them things like, “Well, what’s good or bad about that? Or why is that important to you? Or what role does that have in your life?” or whatever. And through this survey, you end up with something like the values cards that Meaning Alignment Institute was first concerned with, some sense, textual representation of what’s important to the users. And I think that can get you a lot of the way, because if the platform is in the business of—so it’s already like this with food, I mentioned this in the essay. You have an implicit warrant that your food is edible, it’s not going to make people sick or whatever if you’re selling food. And it’s similar with housing. The lease doesn’t have to spell out that there’s no leaks or that there’s heating. There’s some assumptions that are in there about, well, you’re offering housing, so this is what housing means.
So to some extent it’s even a data science question: you’re offering friendship, so this is what friendship means. And the degree to which you’re offering friendship really does have to do with this issue of reliance or participation. So I think email and social media are very different in this regard, for instance, whereas email—well, one thing is just SMTP is a neutral platform that many different kinds of operators can interoperate with, so it has less of a reliance aspect. And then also something like Facebook or Twitter structures your interactions much more, the way that it shows people’s social proof and different friend groups and the algorithm for recommending this tweet above this tweet and so on. It’s getting in there much more deeply, and that should show up in data, survey data from users.
Ivan Vendrov: I guess, do you worry that this is standing athwart history yelling “Stop,” where it’s like, yeah, we could do all this sophisticated philosophical stuff, but it’s just going to slow us down, and the vast majority of people and most societies are just going to accelerate in the technocapital direction? And then sure, you’ll have a couple of little custom, philosophically robust telocratic social networks that a few of our friends use.
Joe Edelman: No, I don’t think so. Did you read my value drift essay? This is my statement on this Nick Landy stuff. I think that people who talk about technocapital or Nick Land, they’re making something sound badass that’s actually more like noise or just dysfunction. You can split it into two parts. So there’s infrastructural functionality that grows no matter what kinds of transactions are happening—so financial infrastructure, compute counts like this, oil, energy infrastructure. So that will always be expanding to support whatever bullshit is happening up here, whether that’s real human needs are being addressed, or random shit is being done, maybe this company and this company are having a marketing war or something, or there was an oil spill over here and then these people are cleaning it up. No matter what is growing up here, there’s some infrastructure growing. And I think that’s where this badass feeling comes from, because people watch the infrastructure growing and they’re like, “Oh yeah, it has its own sensibility, its own direction, capitalism!”
But what you see when you just look above the infrastructure is just a lot of misdirected random bullshit. It’s not like—for instance, Facebook, which everybody’s always complaining about, that’s a deadweight loss. Social media is a terrible waste of time. And there’s many, they’re all examples like that. It’s some bullshit metric that some manager invented to get themselves a bonus or something is being up-and-to-the-righted, and so then this company appears to be growing, but it’s actually going to crash later because everybody hates it or whatever. And I don’t think this deserves some fancy word like technocapital. It’s just an embarrassing dysfunction, or a cancer, but even a cancer is more badass than this. So I think that it’s actually the opposite.
Ivan Vendrov: Joe to Facebook: “You’re not even cancer.”
Joe Edelman: Yeah. It’s the opposite from what you said. Actually the dinky, messy thing is the Landian regime. And if we manage to get this up, the telocratic regime, where things are actually well directed towards the purposes and ends that people actually care about, that society is going to be much more ambitious and wild in terms of what it achieves. It’s going to be making huge cultural leaps, it’s going to be making huge—something like SpaceX is not technocapital doing its thing. Elon Musk and the people who work for him, they have some bold vision, they want to go to Mars and stuff. That kind of stuff is going to go crazy. They’ll do very, very well in a telocratic regime. So I think what we’ll see actually is simultaneous.
Ivan Vendrov: And I think Elon’s companies demonstrate that technocapital does in fact select for telocracy already, with sufficient competence.
Well, that makes me feel better. The one thing, maybe I’ll push back a little bit, where I’m just like, it does seem like these—sure, social media is a big waste of time and in some sense is just a failure of contracting or a mistake. But it sure seems like these bubbles of misalignment are getting bigger and bigger. And so you can see how if you’re paying attention to this bubble of misalignment where a billion people’s time is being wasted and they’re being hypnotized, and that hypnosis is being used to sell them products that they don’t need, which in turn feeds this dissociation machine—and yes, in some sense it’s unsustainable, but in another sense I’m like, if the bubbles are big enough and we’re constantly inflating new bubbles of misalignment, isn’t that the main thing that’s happening? And that’s the Nick Land position.
Joe Edelman: Yeah, I do think that there’s a correction that will come. People fall in different places with regard to the bullshit jobs hypothesis, but if by bullshit jobs you just include anything where it’s not actually worth doing if you zoom way out, which would include all zero-sum things, all irrational bubbles and stuff like that, then I do believe that we have a lot of bullshit that our economy is doing and that there will be a really big correction and a lot of people will have to reconsider—if telocracy is succeeding, then a lot of people will have to reconsider what they’re doing and find things that are more purposeful. I do believe that will drive more growth, but it’s going to be a different kind of growth that involves, let’s say, a deeper search for what’s actually worth doing.
Institutional Evolution and Overcoming Matching Bottlenecks
Ivan Vendrov: That makes sense. There’s one thing I wanted to get more detail on, which I found fascinating. I think this is Appendix A, where you describe this hierarchy of how abstract a purpose can be, and this feeling of—let’s see—oh yeah, the “Levels of Articulation” section, where a telocratic organization needs a mandate. A mandate can be a substantive purpose like human connection, or healing people like a hospital. And then it could also have a discursive purpose, so a newspaper doesn’t commit to any particular thing except for making sure that diverse voices get heard and people who disagree with the current consensus can be heard. And then a formal purpose, like a Quaker meeting, where it’s even more abstract, if I understand you correctly, where you’re not really saying anything about what a Quaker meeting will lead to, you’re just like, this institution exists to maintain this structure where people come and sit in a place for an hour every Sunday, and whatever happens is up to the people, it’s not up to the institution in some sense. And then there’s an even higher level, which is evolutionary purpose.
And you have this interesting thing of institutions can move up the ladder of abstraction. So if the people who happen to be in an institution start disagreeing a bunch, then you want to move to something more abstract. And conversely, maybe if the institution shrinks or the people within the institution become really good friends, they might now go from—it’s a bit like you start with a book club, where in some sense that’s a pretty abstract intention, or it’s a very thin intention, you’re just meeting together to talk about books every week, and then as you grow deeper in friendship, you’re like, “No, no, no, we actually have a purpose, we’re doing this thing together now.” And so that becomes thicker over time.
I found that really fascinating, this idea of institutions that live in this way and go up and down the level of abstraction of values. Just wanted to get a little bit more—just speak more on that, I guess.
Joe Edelman: One of my motivations is I want to make it really easy, or as easy as possible, to bring over federal government institutions like the CDC, or other public institutions that have very contested purposes, like public schools. So I want there to always be a place to start even if the people could come to agree more through a lot of deliberation. I don’t want them to have to on day one, and I want there still to be a sense in which the thing is not just pretending to have a purpose. It has one of these purposes that are at a higher level of abstraction.
And then from there you can go down as people agree more. You can also go, I call it sideways, where you break the institution into two if there’s two groups that have purposes in mind that are somewhat different.
So hopefully, the idea is that if you can move in all these directions—up, down, sideways—then there’s room in society for a lot of these telocratic organizations. And I think they work better if there’s more substantive sense of what’s good among the people involved because they can go after it in a much more direct way, but it’s important not to demand that.
Ivan Vendrov: Can you say more about how you envision this transition happening? Because that is one of my main skepticisms about the proposal. I’m just like, man, it just seems like the US government and the regulatory agencies are just never going to change, and this makes me sympathetic to Silicon Valley libertarian accelerationism where I’m like, yeah, the government will just keep eating half of our surplus forever more and producing no value, and that’s fine, we just have to do good things despite that, basically.
Joe Edelman: I think it makes a lot of sense to be. I do, yeah, but I’m not so much like, let’s bring it to Washington. So I think the American Revolution is a good example, which eventually is let’s bring it to Washington, but it was a little bit of a long way around.
In colonial America, a lot of what later became the federal government and the Constitution was trialed locally in different contexts. The Jeffersonian democratic ideal of public schools and public libraries and associations of people voting or just gathering and pooling resources to do projects that they believed in was tested throughout the colonies. And there was this shadow government, which was way more important than the actual government, at least on a local level, which was putting into practice all of the ideology of democracy 60 years before the American Revolution. So people had a direct experience of voting, of all this other stuff. Also, many of the colonial governments in each colony had this structure. Virginia had a bicameral legislature before the federal government was even imagined. So people were experimenting with this new way of doing things and convincing themselves that it worked.
And then, yeah, the old regime did not go easy. The way I see it happening in the next years is something like, well, first people need a direct experience of it going well. Which is right now, I just put out the essay two weeks ago, so we got to give it a little time, some months, for some of these telocratic organizations to exist, to start doing business with each other, the purpose-bound property, the contracts. All the stuff needs to be observed to actually do the same work that I’m hypothesizing that it does, and not have terrible failure modes that I didn’t foresee and so on. So that’s the first step.
And then I think at the same time as that’s happening, a community of activism or something will grow—people who have watched these successes and think, oh, actually this is the way to go. And then a political party might exist somewhere, but probably not in the US. It’s not very practical in the US for many reasons. Niche political parties don’t really do anything in the US, and the US is just in a bad way politically. It’s not a good place for an optimistic new approach or something. But there’s plenty of countries that are. There’s countries that are very nimble in terms of their policy, where niche new approaches do make progress within the system or whatever. So it wouldn’t surprise me if within a year and a half there is some political presence for telocracy somewhere in the world.
And then if this is working better and seems like a better solution to the people, well, something will eventually have to happen in the US. Just like there’s Bernie Bros or whatever that are yelling, we need to be like Sweden or something, it will be something like that in some future iteration in the US.
But because of my beliefs about geopolitical power and compute and things like that, I don’t even think that’s that important. What’s important is that a telocratic regime of sizable resources is set up so that we can really see how it’s doing, and see if it’s a much better place to live, see if it’s a much more trustworthy institutional stack where people believe in the institutions much more and so on. So something needs to be set up that’s big enough so that we can be like, wow, that seems like the way to do it, but that doesn’t need to be in one of the places—I think the US has some of the lowest trust in government in the world right now. It’s down there with Sierra Leone.
Ivan Vendrov: No response. I guess one disagreement, or maybe not disagreement, a misunderstanding I still have is I feel like telocracy places too much faith in words or something. The historical analogy that comes to mind, we talked about this with Richard, is think of the role of the American Constitution, where Americans love to say that the Constitution is what makes America. When I look at the Constitution, I’m like, almost none of this is being followed right now. Also, if you look at the alternative to setting up the Constitution, which was stay a colony of Britain, I’m like, yeah, Canada, Australia, New Zealand basically seem to be doing comparably well with no articulated purposes. They just relied on common law and precedent. And yeah, I guess it’s not obvious to me that this process of really trying to articulate the values of your organization is better than just trying to do the good thing.
Joe Edelman: Oh yeah, no, I don’t think it is. I pretty much agree with you straight down the road. So one thing is that I focused on textual mandates in the document because I think for early telocratic organizations, the legibility of the reasoning about the audits and the legibility of the mandates that are the basis of that reasoning will be paramount. But it’s not ideal. What would be great would be just if the humans and the AIs have a sense of what good is, and they can act on that, and people trust that the AIs have a similar sense of what’s good as they do without needing a written record or something. I don’t think that’s going to be the first iteration, but it would be wonderful if we can get there, and I’m optimistic that we will.
And then the other part of what you were saying is that an organization that is already doing the right thing because people are really thinking about what’s good and trying to do it is not really my target here. I’m not pressuring existing nonprofits that have really good purpose-driven leadership to convert to telocratic organizations. In fact, it’s an unnecessary drag on their work, so they shouldn’t. The reason to convert is when you have pressures—like for instance the British network of AI labs that I was mentioning, which has all these different people’s academic interests, and it’s a complicated mix of incentives, or for instance when you raise VC—when you expect it to actually be hard to stick to your purpose, that’s when this ends up being a win.
Ivan Vendrov: I see, makes sense. And yeah, the founding of the US also makes sense in that light. In fact, they did find it very hard to stick to their purpose, and they’re like, okay, no, we actually need to articulate what our values are so that we can have a judicial system that enforces this.
Joe Edelman: I also think common law is a better example. Some of it is just size; the Constitution is very small. Common law and this more statistical model of sticking to purpose that common law has is maybe a better—
Ivan Vendrov: What do you mean by statistical model?
Joe Edelman: Well, there’s two ways. Telocracy is based on auditing, agent auditing, which is going to catch some stuff, not necessarily all the stuff. It’s not based on making every decision; it’s based on catching failures. And that’s what common law is. Whereas the Constitution is this mastermind galaxy-brain system of checks and balances that’s a different kind of thing. It’s trying to be the operating system of the country rather than trying to specify situations in which you could sue or something.
Ivan Vendrov: Common law feels a lot more organic and flexible, and you could just add a few more agent auditors over here to patch this failure case that seems to be popping up a bunch, which is the precedent analogy, rather than you don’t really have to figure out the overall system in advance. I wanted to ask, maybe last question—
Joe Edelman: Am I winning you over?
Ivan Vendrov: I really like the common law comparison because there is something magical about common law that I just can’t quite articulate. And I’m like, yeah, a hundred times more of that just seems straightforwardly good, in a way that thinking about it as organizations having 10-page constitutions, that’s not obviously better. I don’t get excited about that. So that reframe is really helping.
I wanted to ask about the matching thing. One of the subthemes of the essay is creating these systems for people to articulate what they want, maybe what they’re missing in their current lives, and then helping them find each other and coordinate on forming these associations. And maybe I’m just confused why this hasn’t been happening. I mean, in a way, the internet is a machine for this, and there’s all these subreddits and stuff. I’m just curious, what are the bottlenecks to this happening in your view right now?
Joe Edelman: I go into this in the purposive groups section. At the very beginning, I talk about matching between product companies and people that might buy those products. And this is an extremely well-developed area of matching.
Ivan Vendrov: Through advertising, yeah.
Joe Edelman: So it’s telling: why is that so good and dating apps are so bad? What’s going on? Well, you get a kickback when you match a product and a consumer. And also there’s pretty good technology—chargebacks is a good example. If you get a lot of people to buy something, and then it’s actually not very good and they return it, platforms like Amazon or Stripe will penalize that account; their fees will go up usually or whatever. So all of this creates a fairly virtuous ecosystem of matching products and companies.
I would expect that 100,000 times more monetary effort and attention has gone into that problem than other kinds of matching, like dating apps, and also they tend to be really misaligned. I was close to this: I was one of the people who built Couchsurfing, and Couchsurfing did have a recommender matching algorithm at the heart of it that looked for hosts and travelers that would really get along. And for a lot of different transaction cost reasons, the matching that Airbnb does between rooms and people—and paid—just was a much better business to be in.
And it’s not a small difference. The amount of friction in matching a person to a person versus matching a person to a paid room is much higher. The ROI is much higher in the case of the room and so on. So it’s just really night and day a worse business to be in. So I think we need to change that. And I think this probably requires public funding, or some pool of funds that’s reserved for rewarding systems that do this really well.
Ivan Vendrov: Maybe I don’t understand—it still feels like a market failure somehow. Where I’m like, Couchsurfing, I’m sure so many lifelong friendships emerged from people staying with other people on Couchsurfing, whereas much fewer happened from matching people to paid rooms on Airbnb. Obviously a successful marriage is worth a lot of money to both parties in some sense.
Robin Hanson would say we have these demarcations between things that are sacred, like friendship and dating, and then things that are profane, like market transactions and hotels and stuff like that. And so we’re not willing to be actually capitalist about dating, we’re not willing to be actually capitalist about Couchsurfing. If we were, then for-profit corporations would be making billions of dollars off creating millions of marriages or whatever. But that’s not your perspective?
Joe Edelman: No, that’s not true at all. We’re happy to pay for those things, and in fact people pay a lot for those things. But those businesses are not usually incentivized to actually deliver on the outcome, and there’s a few different transaction-costy things that come up.
One is bilateral contracts obviously favor two-party situations. It’s much harder to broker a group thing than a two-party thing. So dating can still be a two-party thing, but going with a bunch of friends to a festival or something, or having dinner even with a bunch of friends or whatever, these things are penalized by market structure.
Then another thing that’s penalized is things that are hard to assess. Assessing a room is a lot easier than assessing a potential date or whatever. This is one of the reasons why all the dating apps have photos front and center, because that’s something that’s at least easy to assess, even if it’s not actually what you’re looking for. But it’s still harder to assess than a room. So that’s another kind of market failure.
And then also there’s a difference between outcomes and what you purchase. On Tinder or whatever, you purchase swipes. The further outcomes are away from—you order something on Amazon, you get the thing that you ordered from Amazon. Amazon would have a hard time selling you a certain amount of bicycling rather than selling you a bicycle, even though what you want is to spend your time biking around town or whatever. Think about the extra risk that Amazon takes on if they sell you a certain amount of biking. Then there’s all these things that they don’t control that are now part of whether they get paid.
Ivan Vendrov: Right, but these seem like essential constraints. I’m curious about the relationship with telocracy, I guess. Is it that matching is almost a prerequisite for telocracy really working and we need to find some other solution for it, or do you believe that telocratic organizations are in some sense the solution to this matching problem? And if so, how do they overcome these seemingly fundamental transaction cost bottlenecks?
Joe Edelman: I believe that we need, A, new kinds of mechanism design, and B, probably subsidies. I think that this is just an area, just like we need subsidy for science—the profit motive doesn’t really drive science enough. It’s a similar problem because it’s several steps away from value, right?
We also need subsidies for these matching mechanisms that ideally—it’s not so hard, right? It’s like an impact bond, right? If you care about marriages and there’s some platform that’s delivering a lot of marriages, well, great, pay the platform for that, right? And then suddenly Couchsurfing would be a good business instead of Airbnb, right?
So I think it’s not even such a hard problem. It’s just something that we assume that the market will take care of it, but because of the structural features of bilateral contracting, it has some of the same problems as science. It needs something else.
Ivan Vendrov: That makes sense to me. Final question, I guess, maybe just—and this is both for myself and other listeners who you can imagine are somewhat similar to me—how do we help with telocracy? What are the next steps?
Joe Edelman: Cool. Yeah, connect with me and I can put you in a Signal group where people are spreading the ideas. That’s one kind of thing. And then there’s a lot to invent. There’s stuff to invent, and then after that there’s the copying.
So I’m trying to create a network. I mentioned this Oxford ML research organization—oh, this British one. There’s that one, there’s a couple other very early telocratic experiments, and my hope is that people will try things in different places, they’ll invent some of the mechanisms, ship them as code to GitHub or as law to some Google Drive somewhere, and then others can copy it, and we’ll slowly build up the body of practice and technique that’s necessary to run these experiments at larger and larger scales in places where their success is more and more impressive, etc. So I think these are two things: spreading the word and inventing.
Ivan Vendrov: Cool. All right, well, thanks so much for coming on the show, Joe. This was great.
Joe Edelman: My pleasure. I’m glad you’re doing it.
(In which we discuss the Telocracy essay Joe recently published in Pax Machina)
Ivan Vendrov: Welcome to the show, Joe.
Joe Edelman: Thank you. I’m so excited that you have a podcast now.
Ivan Vendrov: Yeah, or whatever this is. No, I’m really excited to talk to you because I just feel like there’s no positive visions for what to do with all the GPUs and all the people, and we’re going through the largest social transition since the invention of the printing press, maybe since the invention of writing, and all the ideas we have are either something horrifying or we’ll all live forever in infinite abundance luxury space communism.
And you wrote, I think, maybe the best, or certainly the most complete vision of where we should go from here, which is called telocracy. People can look it up online. I just want to deep dive into what is telocracy, how do we get from here to there, how does it relate to all the other things going on. First, what is telocracy?
Joe Edelman: Somebody explained it back to me really simply just yesterday, so I have a new way to pitch it that I’ll try.
Well, actually, maybe I’ll start: the default path seems to be that AI will do what is good for large corporations and possibly governments or militaries. And that’s a pretty small amount of what’s good overall.
Maybe you can add individuals to that as well, if you think about each of us accelerated by our chatbots—although there’s transaction cost reasons, there’s various reasons. I think I’ve come to believe that the cyberpunk future of zaibatsu, giant corporations, megacorps, and so on—this future actually comes from transaction cost economics. It comes from recognizing that so far, information technology has been really good for scaling hierarchical organizations. If you think about how many agents are working for Anthropic, for instance, right now, it seems to be very much continuing the trend: the large corporations become incredibly large corporations.
I think this is also some of the fear that people have about governments, is that, for instance, it’s impractical to monitor all violations of most laws. So really the police wait for complaints and things like that. But people are scared that with AI, government agencies can grow to the point where they can just monitor everything and the government is all-powerful, which makes it extra important that the laws are correct, that they’re doing the right things, et cetera. So really these two fears about megacorps and about government overreach are the same fear, which is about existing hierarchical structures being amplified by information technology in general, because I think that’s already been the case, right? The NSA benefits a lot from the internet, for instance.
So even if we as individuals become a little more powerful because of our agents, if at the same time the main event is happening at these larger scales, that’s not so good. And yet, what else is there is the problem. What other interests are there besides the ones that I just enumerated?
And the truth is, well, there’s lots of other interests. Everybody has clubs that they’re part of, maybe church groups that they’re part of, weird projects they want to start—some weird intellectual project or creative project or whatever. And I think intuitively many of us know that it would be way better to pour the power of AI into all of that, right? It would also be much more popular. People would be like, “Sure, if those data centers exist not to amplify the government or big business, but to amplify my band’s output, or my—”
Ivan Vendrov: My church’s output.
Joe Edelman: Yeah, that kind of success or amplification or superintelligence or whatever you want to call it would be much more popular and relatable, and we could see it all around us, and it wouldn’t be some abstract capitalist increase or something, but more the flourishing of all the things that we already care about.
So telocracy is about how to actually pull that off: how to change the incentive design and the way that compute is allocated and the way that things are monitored to pull that off so that you can actually tell. It’s a hard problem, because as soon as you—let’s say you just gave compute, you portioned it out to groups or whatever. Well, people with money could just buy it from them, or they might actually turn out to be groups that fight each other, like the political polarization that we see these days in a lot of countries. So it ends up being a complicated problem how to organize society to pour the AI into these kinds of purposes.
Ivan Vendrov: Well, maybe this is a good time to ask this, as I was going to wait longer, but didn’t we used to have telocracy, and then basically we militarized and then never demilitarized after World War II? The reason big corporations and the military and the government apparatus is so powerful is they explicitly tried to essentially do a total mobilization of all civil society. What Tocqueville describes in America in the 1830s is much closer to telocracy than what we have today. And don’t you basically have to take on the military-industrial complex?
Joe Edelman: No, I don’t think so. I guess we have different explanations for why civil society collapsed. I was talking about this with somebody a few days ago, and I gave three reasons why civil society collapsed. Let’s see if I can remember them all.
So one is that civil society organizations tend to be less dynamic than capitalist organizations. So if you want to provision healthcare or housing or whatever it is, insurance or whatever, market organizations tend to be fairly dynamic and they tend to react to changing structures in demand, changing technological conditions, new restructurings that make costs cheaper, make it possible to reach many more people, or whatever. And so civil society organizations just don’t do that as much, and for that reason, they’re less to be trusted with important things like healthcare and housing and so on. So I think that’s probably one of the main reasons.
Another reason is that there are cases where, at least ideologically, we thought that the government was a fairer or better way to do things than private associations. One reason is because with the governments, you can often specify what the level playing field is or what the minimal—you want to get everybody above the poverty line or something like that. Associations tend to be local and have different overlapping polities or whatever, and so you might—at least that’s a reason why I think there’s been a push in general away from associational ways of dealing, especially with social needs or whatever, towards government provision.
Ivan Vendrov: Right, the example that comes to mind there is school districts in the United States, where rich people will cluster together in a rich neighborhood and then they’ll fund a really good school. And in some ways that’s a very natural thing to happen, but one of the justifications for government intervention in education is that, well, we want an even playing field; every child should receive a good education. And that’s an example where voluntary associationalism failed.
Joe Edelman: And then the third one is just atomization, the same thing that happened due to urbanization as people moved away from their families and so on. It becomes much harder to form these associations. And to some extent market provision just displaced—if you can order DoorDash or whatever, there’s less of a reason to cook with friends, et cetera, especially if you just moved to the city and you’re no longer surrounded by your church congregation or your family or whatever, then there’s just some displacement away from associations that way. So I see it as those three causes rather than this post-World War II cause that you gave. And I think there’s data to back up the attribution to those three causes rather than the one you suggested.
Ivan Vendrov: Well, I guess on the first point, which maybe is the deepest one, capitalist organizations or market organizations just seem more dynamic and more adaptable. If that’s true, then where does the hope for telocracy lie? Won’t that continue being the case?
Joe Edelman: No, I think there are some cases where a nonprofit has been dynamic in that way because it was really clear about what its mission or purpose was, rather than just continuing with its mode of operation. The World Wildlife Fund is one example. They used to be a bunch of lawyers that would stop parks from closing, but their top-line purpose was to stop animals from going extinct, different species. And so at some point they were like, “Well, actually buying land is much more effective than stopping parks from closing.” And they switched, and they fired their lawyers, and they hired a lot of hedge fund people that do this cool GIS stuff where they’re like, “Oh, there’s more owls over here, but this land is worth more, so we’re going to sell this and buy this,” or whatever. And this has been way more effective.
What you’d want is some way of pressuring—not necessarily for-profit; telocratic organizations can be for-profit, I leave that open in the essay—but even if they’re nonprofit, you would want some way that they’re held to their mission in a way where you have to make those pivots, where you’re in some kind of breach if you keep doing the lawyer thing when the hedge fund thing is actually what would be required to pursue your mission.
Ivan Vendrov: That brings me to another question I had, which is: what is the enforcement mechanism here? Briefly, I guess telocratic organizations have this really articulated vision of what we are about, what we are doing and what we are not doing, what is in scope, what is not in scope, what are the trade-offs we’re willing to make—which virtually no organization currently has. A C-corp literally just has basically the interests of shareholders or whatever the owners believe, and that doesn’t have to be articulated in any way. Telocratic organizations have this deep articulation, but how would that be enforceable?
Joe Edelman: Yeah, there’s different mechanisms that I suggest in the essay. I think for prototyping it, the most useful thing is to build pools of assets that automatically reallocate themselves based on whether their current holder is living up to purpose. And the most obvious one there is compute.
So you can have a compute cloud—it looks like even current compute providers could give a fraction of their cloud to this kind of cause. And there’s agents that are auditing the activities of the organization according to its purpose, and those agents can flag some problem, like the one we were just talking about with the lawyers versus the hedge fund people or whatever, and then the organization would have a certain amount of time to respond, or its compute allocation would change, for instance. And there’s many other assets you can attach.
Ivan Vendrov: Well, so “would change”—this must ground out in lawsuits on the existing legal system at some level, right?
Joe Edelman: No. To give a really concrete example, there’s already a first telocratic org, which is really exciting, and it’s a network of AI labs in Britain called Bold that are trying to be an academic alternative to the big labs. They have a data center in Britain that’s really sizable, money from private investors and from the British government, and they have certain research programs where they’re trying to beat the mainstream AI research directions with their own directions. And this is different than how academic ML research has worked in the past, where usually each grad student really has their own research program; they’re trying to do something much more directed.
And so the way it works now is that when you’re a grad student and you decide to join this initiative, you get access to this giant cloud, but you have to post your project ideas and your research direction, and then it’s compared to the mandate that this cloud is allocated for, and your GPU allocation is going to depend on whether you’re in line with the broader purposes of the network. Currently they’re not doing this auditing thing, but that would be the next step: you can just look at what they’re actually running on their GPUs and make sure that it continues to be in line.
Ivan Vendrov: I see. So in this case, the actual governance layer is presumably some board of directors, and in some sense they don’t have to articulate their values, but they’re choosing to abide by this articulated values constitution, and they’ve implemented an enforcement layer that is automated. Does that sound right?
Joe Edelman: Sure, yeah, or in the general case, I think it’s just contracts. The contract says instead of “you’ve paid this much and you get this much of our cloud,” it says maybe “you get this much of our cloud for free,” or something, “but you put up with this auditing, and we revoke it if the following conditions aren’t met.”
Ivan Vendrov: That makes sense. That makes me more optimistic, because there’s a lot more ways that this can interface with the existing legal structure. In some sense, I could start a C-corp and run it as a telocratic organization, basically.
Joe Edelman: For sure, yeah.
Ivan Vendrov: And presumably there are a bunch of companies that already run this way—they just happen to have leaders that are really good at articulating their values and are very strict in how closely they adhere to those values. I’m curious if any come to mind, just to give more examples or intuitions.
Joe Edelman: I think the big step is the agent auditing, and I think this is actually pretty rare because agents have just recently become barely capable of making these “is it within the purpose, is it not within the purpose” kinds of decisions.
So there’s a lot of purpose-driven organizations in the world—the Red Cross—and I think one of the things that telocracy needs to prove out within the next few months, even, ideally, is that we can transition from believing that those organizations are using resources in the best way for their purpose or whatever, to knowing, through more extensive auditing and things like that. And then that’s great: if the organizations are already doing the right things, then maybe we can even just increase the amount of resources available to them.
Agent Auditing as Scaling Common Law
Ivan Vendrov: I guess I’m not sure I understand what the essential thing about AI agents is here in this picture. Because in principle, if telocracy is a good idea, then we should be seeing organizations that just happen to devote 5% of their resources to doing human auditing of their purposes, and that should already be working. And yes, AI agents make it cheaper—it’s actually not obvious to me that AI agents make it differentially cheaper to audit than to fool the auditors or something. It seems zero-sum to me, roughly. Why do you think—
Joe Edelman: Wait, to audit versus to what?
Ivan Vendrov: To fool the auditors.
Joe Edelman: Ah. Well, I think in general, human judgment has been very expensive in the domain of mechanism design—the kind of thing that a judge does, where they use a body of principles or values and they issue some reasoned opinion. This, I think, has been very expensive, and that’s one of the reasons why we use voting and markets and stakeholders who represent the values, but in a vague, fuzzy way and with a lot of danger for capture and so on. All of the mechanisms we use are often hacks because it’s too expensive to do this much case-by-case reasoning.
So I think there’s evidence that AI already changes this. One of them is just the Instagram recommender, which is currently a generative recommender. I don’t know if you’ve used it, but if you’re on the Reels tab, you can click “Your algorithm,” and then you can talk to the recommender now. And the recommender works now by watching the videos and doing case-by-case reasoning about what you would like to see, which is very, very different than how it worked a year ago, where it showed you things that your friends liked or that people like you liked or whatever, which is a much cheaper computation to run.
And it would have been obviously unaffordable to pay human people to watch every video for every person and think, would Ivan like this? That would be one person for every person. So I think there’s pretty good evidence.
Ivan Vendrov: Kind of like the Stasi.
I just had a reframe of telocracy which I’d love to get you to criticize, which is there’s something about the reason why Anglosphere governance works so well is because of common law. We owe a lot of the good things in our civilization to common law, which is precisely this: a careful person reads the law and looks at all the information and makes a reasoned decision that is public and transparent. And you’re like, that’s just been very, very expensive, and we mostly just used it in high affairs of state or murder trials or something. And AI agents let us do this a thousand times cheaper, so now a lot more of economy and society can have the good properties of common law.
Joe Edelman: Yeah, I think that’s right, and I think that’s a big part of my intuition that there will be so much institutional change. Because obviously if it’s gone from being a thing that a judge does and that you and I don’t know about hardly at all, to a thing that’s built into the recommendation system for the videos we watch or whatever, that’s a very big shift.
Ivan Vendrov: Most people don’t really interact with the legal system outside of watching procedural dramas or something. And it’s actually every second there’s a micro judge or computer deciding little things that are happening, and you can engage with that reasoning and see the ways in which it does and does not fit with your purposes. That’s really cool.
Maybe let’s dive into this social media example, because it goes through—you’ve talked about it here, you also talk about it in the essay. So you have this example of a social media platform starts out small, but then people start relying on it, so let’s say it starts mediating young people’s friendships. And then you’re like, well, under this concept of living commitments, you then can’t pivot the platform to mine that relationship for engagement and ad revenue. You now have to take on different values of nourishing friendships, or you have to give up control to other organizations that are willing to do the nourishing friendships thing.
And I’m like, that sounds awesome, and I’m also confused about who’s enforcing this, how could this be enforced, why would an organization ever give up its power? Seems like there’d be a lot of conflict there. And also how you adjudicate these really nuanced aspects of human relationships.
Joe Edelman: I can talk about all those. So first of all, about enforcement, there’s going to be different phases to the rollout of telocracy. The first phase is going to be very optional. So it’s going to be the same groups that do maybe public benefit corps and things like that now, which can be high profile, but they chose to do that. And so they’ll often be tying their hands to the mast or whatever. They’ll be doing that because they think it’s a good idea for their own values, the founders, etc. And then they’ll enter into different kinds of contracts that do that. So they’ll choose to get telocratic compute that disappears if they don’t follow their purpose, or they’ll choose to make their brand telocratic so that they’re no longer allowed to use their brand if they stray from their purpose or whatever. And that will prove out that it works in these cases.
And then there’s a phase where there’ll be enough of these that society is starting to figure out which things should maybe always be telocratic organizations. So maybe nursing homes should not be allowed to be for-profit organizations that are not telocratic. There’ll probably be some high-profile things that go wrong. Some autonomous corporation that was originally going to do this thing, because all the agents decide they’ll make more money doing this other really terrible thing or whatever, they all do that thing, and then there’ll be some maybe law or just maybe common sense that if you’re setting up one of those things—a nursing home or an autonomous corporation above a certain size or something—it needs to be bound to purpose.
And then the third stage is there’ll be courts and things like that, there’ll be structures in society that try to enforce that new sense. So if there’s an old nursing home that wasn’t telocratic or whatever, there’ll be some law that it has to migrate over or something like that. So I see all that taking, I don’t know, at least a decade or something, quite a while. And the enforcement mechanism will be a little different in all three of those cases.
And then I think that was your first two questions, and what was the third again?
Ivan Vendrov: The third was it just seems really, this operation that you’re imagining of a social media facilitates a relationship between people and then the company gives control up to another company, and I’m like, this just seems really subtle. It’s very hard to imagine even a current-day judge making a decision of this nature: “Ah yes, this relationship, we’ve looked at your chat logs and based on that we really think this is facilitating a friendship, and the way you guys are optimizing for engagement is inimical to our concept of friendship.” And I’m like, man, we just don’t have the philosophy for that kind of thing as a society.
Joe Edelman: So the idea is that you’re right that we don’t have the philosophy for it, and that’s some of the work ahead even for me. One way to do it, just to give you an intuition about the possibility or plausibility of this, is you could just ask people what they rely on, what they use the platform for, or what has happened for them that’s good or bad on the platform, and you’ll end up with things like, “I met my friend Sally,” or whatever.
And then you can ask them things like, “Well, what’s good or bad about that? Or why is that important to you? Or what role does that have in your life?” or whatever. And through this survey, you end up with something like the values cards that Meaning Alignment Institute was first concerned with, some sense, textual representation of what’s important to the users. And I think that can get you a lot of the way, because if the platform is in the business of—so it’s already like this with food, I mentioned this in the essay. You have an implicit warrant that your food is edible, it’s not going to make people sick or whatever if you’re selling food. And it’s similar with housing. The lease doesn’t have to spell out that there’s no leaks or that there’s heating. There’s some assumptions that are in there about, well, you’re offering housing, so this is what housing means.
So to some extent it’s even a data science question: you’re offering friendship, so this is what friendship means. And the degree to which you’re offering friendship really does have to do with this issue of reliance or participation. So I think email and social media are very different in this regard, for instance, whereas email—well, one thing is just SMTP is a neutral platform that many different kinds of operators can interoperate with, so it has less of a reliance aspect. And then also something like Facebook or Twitter structures your interactions much more, the way that it shows people’s social proof and different friend groups and the algorithm for recommending this tweet above this tweet and so on. It’s getting in there much more deeply, and that should show up in data, survey data from users.
Ivan Vendrov: I guess, do you worry that this is standing athwart history yelling “Stop,” where it’s like, yeah, we could do all this sophisticated philosophical stuff, but it’s just going to slow us down, and the vast majority of people and most societies are just going to accelerate in the technocapital direction? And then sure, you’ll have a couple of little custom, philosophically robust telocratic social networks that a few of our friends use.
Joe Edelman: No, I don’t think so. Did you read my value drift essay? This is my statement on this Nick Landy stuff. I think that people who talk about technocapital or Nick Land, they’re making something sound badass that’s actually more like noise or just dysfunction. You can split it into two parts. So there’s infrastructural functionality that grows no matter what kinds of transactions are happening—so financial infrastructure, compute counts like this, oil, energy infrastructure. So that will always be expanding to support whatever bullshit is happening up here, whether that’s real human needs are being addressed, or random shit is being done, maybe this company and this company are having a marketing war or something, or there was an oil spill over here and then these people are cleaning it up. No matter what is growing up here, there’s some infrastructure growing. And I think that’s where this badass feeling comes from, because people watch the infrastructure growing and they’re like, “Oh yeah, it has its own sensibility, its own direction, capitalism!”
But what you see when you just look above the infrastructure is just a lot of misdirected random bullshit. It’s not like—for instance, Facebook, which everybody’s always complaining about, that’s a deadweight loss. Social media is a terrible waste of time. And there’s many, they’re all examples like that. It’s some bullshit metric that some manager invented to get themselves a bonus or something is being up-and-to-the-righted, and so then this company appears to be growing, but it’s actually going to crash later because everybody hates it or whatever. And I don’t think this deserves some fancy word like technocapital. It’s just an embarrassing dysfunction, or a cancer, but even a cancer is more badass than this. So I think that it’s actually the opposite.
Ivan Vendrov: Joe to Facebook: “You’re not even cancer.”
Joe Edelman: Yeah. It’s the opposite from what you said. Actually the dinky, messy thing is the Landian regime. And if we manage to get this up, the telocratic regime, where things are actually well directed towards the purposes and ends that people actually care about, that society is going to be much more ambitious and wild in terms of what it achieves. It’s going to be making huge cultural leaps, it’s going to be making huge—something like SpaceX is not technocapital doing its thing. Elon Musk and the people who work for him, they have some bold vision, they want to go to Mars and stuff. That kind of stuff is going to go crazy. They’ll do very, very well in a telocratic regime. So I think what we’ll see actually is simultaneous.
Ivan Vendrov: And I think Elon’s companies demonstrate that technocapital does in fact select for telocracy already, with sufficient competence.
Well, that makes me feel better. The one thing, maybe I’ll push back a little bit, where I’m just like, it does seem like these—sure, social media is a big waste of time and in some sense is just a failure of contracting or a mistake. But it sure seems like these bubbles of misalignment are getting bigger and bigger. And so you can see how if you’re paying attention to this bubble of misalignment where a billion people’s time is being wasted and they’re being hypnotized, and that hypnosis is being used to sell them products that they don’t need, which in turn feeds this dissociation machine—and yes, in some sense it’s unsustainable, but in another sense I’m like, if the bubbles are big enough and we’re constantly inflating new bubbles of misalignment, isn’t that the main thing that’s happening? And that’s the Nick Land position.
Joe Edelman: Yeah, I do think that there’s a correction that will come. People fall in different places with regard to the bullshit jobs hypothesis, but if by bullshit jobs you just include anything where it’s not actually worth doing if you zoom way out, which would include all zero-sum things, all irrational bubbles and stuff like that, then I do believe that we have a lot of bullshit that our economy is doing and that there will be a really big correction and a lot of people will have to reconsider—if telocracy is succeeding, then a lot of people will have to reconsider what they’re doing and find things that are more purposeful. I do believe that will drive more growth, but it’s going to be a different kind of growth that involves, let’s say, a deeper search for what’s actually worth doing.
Institutional Evolution and Overcoming Matching Bottlenecks
Ivan Vendrov: That makes sense. There’s one thing I wanted to get more detail on, which I found fascinating. I think this is Appendix A, where you describe this hierarchy of how abstract a purpose can be, and this feeling of—let’s see—oh yeah, the “Levels of Articulation” section, where a telocratic organization needs a mandate. A mandate can be a substantive purpose like human connection, or healing people like a hospital. And then it could also have a discursive purpose, so a newspaper doesn’t commit to any particular thing except for making sure that diverse voices get heard and people who disagree with the current consensus can be heard. And then a formal purpose, like a Quaker meeting, where it’s even more abstract, if I understand you correctly, where you’re not really saying anything about what a Quaker meeting will lead to, you’re just like, this institution exists to maintain this structure where people come and sit in a place for an hour every Sunday, and whatever happens is up to the people, it’s not up to the institution in some sense. And then there’s an even higher level, which is evolutionary purpose.
And you have this interesting thing of institutions can move up the ladder of abstraction. So if the people who happen to be in an institution start disagreeing a bunch, then you want to move to something more abstract. And conversely, maybe if the institution shrinks or the people within the institution become really good friends, they might now go from—it’s a bit like you start with a book club, where in some sense that’s a pretty abstract intention, or it’s a very thin intention, you’re just meeting together to talk about books every week, and then as you grow deeper in friendship, you’re like, “No, no, no, we actually have a purpose, we’re doing this thing together now.” And so that becomes thicker over time.
I found that really fascinating, this idea of institutions that live in this way and go up and down the level of abstraction of values. Just wanted to get a little bit more—just speak more on that, I guess.
Joe Edelman: One of my motivations is I want to make it really easy, or as easy as possible, to bring over federal government institutions like the CDC, or other public institutions that have very contested purposes, like public schools. So I want there to always be a place to start even if the people could come to agree more through a lot of deliberation. I don’t want them to have to on day one, and I want there still to be a sense in which the thing is not just pretending to have a purpose. It has one of these purposes that are at a higher level of abstraction.
And then from there you can go down as people agree more. You can also go, I call it sideways, where you break the institution into two if there’s two groups that have purposes in mind that are somewhat different.
So hopefully, the idea is that if you can move in all these directions—up, down, sideways—then there’s room in society for a lot of these telocratic organizations. And I think they work better if there’s more substantive sense of what’s good among the people involved because they can go after it in a much more direct way, but it’s important not to demand that.
Ivan Vendrov: Can you say more about how you envision this transition happening? Because that is one of my main skepticisms about the proposal. I’m just like, man, it just seems like the US government and the regulatory agencies are just never going to change, and this makes me sympathetic to Silicon Valley libertarian accelerationism where I’m like, yeah, the government will just keep eating half of our surplus forever more and producing no value, and that’s fine, we just have to do good things despite that, basically.
Joe Edelman: I think it makes a lot of sense to be. I do, yeah, but I’m not so much like, let’s bring it to Washington. So I think the American Revolution is a good example, which eventually is let’s bring it to Washington, but it was a little bit of a long way around.
In colonial America, a lot of what later became the federal government and the Constitution was trialed locally in different contexts. The Jeffersonian democratic ideal of public schools and public libraries and associations of people voting or just gathering and pooling resources to do projects that they believed in was tested throughout the colonies. And there was this shadow government, which was way more important than the actual government, at least on a local level, which was putting into practice all of the ideology of democracy 60 years before the American Revolution. So people had a direct experience of voting, of all this other stuff. Also, many of the colonial governments in each colony had this structure. Virginia had a bicameral legislature before the federal government was even imagined. So people were experimenting with this new way of doing things and convincing themselves that it worked.
And then, yeah, the old regime did not go easy. The way I see it happening in the next years is something like, well, first people need a direct experience of it going well. Which is right now, I just put out the essay two weeks ago, so we got to give it a little time, some months, for some of these telocratic organizations to exist, to start doing business with each other, the purpose-bound property, the contracts. All the stuff needs to be observed to actually do the same work that I’m hypothesizing that it does, and not have terrible failure modes that I didn’t foresee and so on. So that’s the first step.
And then I think at the same time as that’s happening, a community of activism or something will grow—people who have watched these successes and think, oh, actually this is the way to go. And then a political party might exist somewhere, but probably not in the US. It’s not very practical in the US for many reasons. Niche political parties don’t really do anything in the US, and the US is just in a bad way politically. It’s not a good place for an optimistic new approach or something. But there’s plenty of countries that are. There’s countries that are very nimble in terms of their policy, where niche new approaches do make progress within the system or whatever. So it wouldn’t surprise me if within a year and a half there is some political presence for telocracy somewhere in the world.
And then if this is working better and seems like a better solution to the people, well, something will eventually have to happen in the US. Just like there’s Bernie Bros or whatever that are yelling, we need to be like Sweden or something, it will be something like that in some future iteration in the US.
But because of my beliefs about geopolitical power and compute and things like that, I don’t even think that’s that important. What’s important is that a telocratic regime of sizable resources is set up so that we can really see how it’s doing, and see if it’s a much better place to live, see if it’s a much more trustworthy institutional stack where people believe in the institutions much more and so on. So something needs to be set up that’s big enough so that we can be like, wow, that seems like the way to do it, but that doesn’t need to be in one of the places—I think the US has some of the lowest trust in government in the world right now. It’s down there with Sierra Leone.
Ivan Vendrov: No response. I guess one disagreement, or maybe not disagreement, a misunderstanding I still have is I feel like telocracy places too much faith in words or something. The historical analogy that comes to mind, we talked about this with Richard, is think of the role of the American Constitution, where Americans love to say that the Constitution is what makes America. When I look at the Constitution, I’m like, almost none of this is being followed right now. Also, if you look at the alternative to setting up the Constitution, which was stay a colony of Britain, I’m like, yeah, Canada, Australia, New Zealand basically seem to be doing comparably well with no articulated purposes. They just relied on common law and precedent. And yeah, I guess it’s not obvious to me that this process of really trying to articulate the values of your organization is better than just trying to do the good thing.
Joe Edelman: Oh yeah, no, I don’t think it is. I pretty much agree with you straight down the road. So one thing is that I focused on textual mandates in the document because I think for early telocratic organizations, the legibility of the reasoning about the audits and the legibility of the mandates that are the basis of that reasoning will be paramount. But it’s not ideal. What would be great would be just if the humans and the AIs have a sense of what good is, and they can act on that, and people trust that the AIs have a similar sense of what’s good as they do without needing a written record or something. I don’t think that’s going to be the first iteration, but it would be wonderful if we can get there, and I’m optimistic that we will.
And then the other part of what you were saying is that an organization that is already doing the right thing because people are really thinking about what’s good and trying to do it is not really my target here. I’m not pressuring existing nonprofits that have really good purpose-driven leadership to convert to telocratic organizations. In fact, it’s an unnecessary drag on their work, so they shouldn’t. The reason to convert is when you have pressures—like for instance the British network of AI labs that I was mentioning, which has all these different people’s academic interests, and it’s a complicated mix of incentives, or for instance when you raise VC—when you expect it to actually be hard to stick to your purpose, that’s when this ends up being a win.
Ivan Vendrov: I see, makes sense. And yeah, the founding of the US also makes sense in that light. In fact, they did find it very hard to stick to their purpose, and they’re like, okay, no, we actually need to articulate what our values are so that we can have a judicial system that enforces this.
Joe Edelman: I also think common law is a better example. Some of it is just size; the Constitution is very small. Common law and this more statistical model of sticking to purpose that common law has is maybe a better—
Ivan Vendrov: What do you mean by statistical model?
Joe Edelman: Well, there’s two ways. Telocracy is based on auditing, agent auditing, which is going to catch some stuff, not necessarily all the stuff. It’s not based on making every decision; it’s based on catching failures. And that’s what common law is. Whereas the Constitution is this mastermind galaxy-brain system of checks and balances that’s a different kind of thing. It’s trying to be the operating system of the country rather than trying to specify situations in which you could sue or something.
Ivan Vendrov: Common law feels a lot more organic and flexible, and you could just add a few more agent auditors over here to patch this failure case that seems to be popping up a bunch, which is the precedent analogy, rather than you don’t really have to figure out the overall system in advance. I wanted to ask, maybe last question—
Joe Edelman: Am I winning you over?
Ivan Vendrov: I really like the common law comparison because there is something magical about common law that I just can’t quite articulate. And I’m like, yeah, a hundred times more of that just seems straightforwardly good, in a way that thinking about it as organizations having 10-page constitutions, that’s not obviously better. I don’t get excited about that. So that reframe is really helping.
I wanted to ask about the matching thing. One of the subthemes of the essay is creating these systems for people to articulate what they want, maybe what they’re missing in their current lives, and then helping them find each other and coordinate on forming these associations. And maybe I’m just confused why this hasn’t been happening. I mean, in a way, the internet is a machine for this, and there’s all these subreddits and stuff. I’m just curious, what are the bottlenecks to this happening in your view right now?
Joe Edelman: I go into this in the purposive groups section. At the very beginning, I talk about matching between product companies and people that might buy those products. And this is an extremely well-developed area of matching.
Ivan Vendrov: Through advertising, yeah.
Joe Edelman: So it’s telling: why is that so good and dating apps are so bad? What’s going on? Well, you get a kickback when you match a product and a consumer. And also there’s pretty good technology—chargebacks is a good example. If you get a lot of people to buy something, and then it’s actually not very good and they return it, platforms like Amazon or Stripe will penalize that account; their fees will go up usually or whatever. So all of this creates a fairly virtuous ecosystem of matching products and companies.
I would expect that 100,000 times more monetary effort and attention has gone into that problem than other kinds of matching, like dating apps, and also they tend to be really misaligned. I was close to this: I was one of the people who built Couchsurfing, and Couchsurfing did have a recommender matching algorithm at the heart of it that looked for hosts and travelers that would really get along. And for a lot of different transaction cost reasons, the matching that Airbnb does between rooms and people—and paid—just was a much better business to be in.
And it’s not a small difference. The amount of friction in matching a person to a person versus matching a person to a paid room is much higher. The ROI is much higher in the case of the room and so on. So it’s just really night and day a worse business to be in. So I think we need to change that. And I think this probably requires public funding, or some pool of funds that’s reserved for rewarding systems that do this really well.
Ivan Vendrov: Maybe I don’t understand—it still feels like a market failure somehow. Where I’m like, Couchsurfing, I’m sure so many lifelong friendships emerged from people staying with other people on Couchsurfing, whereas much fewer happened from matching people to paid rooms on Airbnb. Obviously a successful marriage is worth a lot of money to both parties in some sense.
Robin Hanson would say we have these demarcations between things that are sacred, like friendship and dating, and then things that are profane, like market transactions and hotels and stuff like that. And so we’re not willing to be actually capitalist about dating, we’re not willing to be actually capitalist about Couchsurfing. If we were, then for-profit corporations would be making billions of dollars off creating millions of marriages or whatever. But that’s not your perspective?
Joe Edelman: No, that’s not true at all. We’re happy to pay for those things, and in fact people pay a lot for those things. But those businesses are not usually incentivized to actually deliver on the outcome, and there’s a few different transaction-costy things that come up.
One is bilateral contracts obviously favor two-party situations. It’s much harder to broker a group thing than a two-party thing. So dating can still be a two-party thing, but going with a bunch of friends to a festival or something, or having dinner even with a bunch of friends or whatever, these things are penalized by market structure.
Then another thing that’s penalized is things that are hard to assess. Assessing a room is a lot easier than assessing a potential date or whatever. This is one of the reasons why all the dating apps have photos front and center, because that’s something that’s at least easy to assess, even if it’s not actually what you’re looking for. But it’s still harder to assess than a room. So that’s another kind of market failure.
And then also there’s a difference between outcomes and what you purchase. On Tinder or whatever, you purchase swipes. The further outcomes are away from—you order something on Amazon, you get the thing that you ordered from Amazon. Amazon would have a hard time selling you a certain amount of bicycling rather than selling you a bicycle, even though what you want is to spend your time biking around town or whatever. Think about the extra risk that Amazon takes on if they sell you a certain amount of biking. Then there’s all these things that they don’t control that are now part of whether they get paid.
Ivan Vendrov: Right, but these seem like essential constraints. I’m curious about the relationship with telocracy, I guess. Is it that matching is almost a prerequisite for telocracy really working and we need to find some other solution for it, or do you believe that telocratic organizations are in some sense the solution to this matching problem? And if so, how do they overcome these seemingly fundamental transaction cost bottlenecks?
Joe Edelman: I believe that we need, A, new kinds of mechanism design, and B, probably subsidies. I think that this is just an area, just like we need subsidy for science—the profit motive doesn’t really drive science enough. It’s a similar problem because it’s several steps away from value, right?
We also need subsidies for these matching mechanisms that ideally—it’s not so hard, right? It’s like an impact bond, right? If you care about marriages and there’s some platform that’s delivering a lot of marriages, well, great, pay the platform for that, right? And then suddenly Couchsurfing would be a good business instead of Airbnb, right?
So I think it’s not even such a hard problem. It’s just something that we assume that the market will take care of it, but because of the structural features of bilateral contracting, it has some of the same problems as science. It needs something else.
Ivan Vendrov: That makes sense to me. Final question, I guess, maybe just—and this is both for myself and other listeners who you can imagine are somewhat similar to me—how do we help with telocracy? What are the next steps?
Joe Edelman: Cool. Yeah, connect with me and I can put you in a Signal group where people are spreading the ideas. That’s one kind of thing. And then there’s a lot to invent. There’s stuff to invent, and then after that there’s the copying.
So I’m trying to create a network. I mentioned this Oxford ML research organization—oh, this British one. There’s that one, there’s a couple other very early telocratic experiments, and my hope is that people will try things in different places, they’ll invent some of the mechanisms, ship them as code to GitHub or as law to some Google Drive somewhere, and then others can copy it, and we’ll slowly build up the body of practice and technique that’s necessary to run these experiments at larger and larger scales in places where their success is more and more impressive, etc. So I think these are two things: spreading the word and inventing.
Ivan Vendrov: Cool. All right, well, thanks so much for coming on the show, Joe. This was great.
Joe Edelman: My pleasure. I’m glad you’re doing it.