Hello, and welcome to Radical Exchanges. I'm joined in this episode by James Evans, who is, among other things, the director of University of Chicago's KnowledgeLab and an external professor of the Santa Fe Institute. With KnowledgeLab, James is launching a new open access journal of social computing, which addresses itself to the interface between computational and social systems, and I recommend that you all check it out. James is a brilliant thinker who, in my estimation, is tackling our most important questions in an incredibly useful and productive way. He highlights the value of diversity within human societies and also in the sense of seeing our tools and technologies as complementing humans, not replacing them. In order for this to happen, we need to clearly understand and profoundly value how humans and machines are different from each other. His thinking makes clear that the betterment of society and the advancement of technology are one project, not two. This conversation was a pleasure. We went deep on the relationship between artificial intelligence and democracy, the trade offs between hybridization and speciation, and much more. As with all of my conversations with James, I learned a lot. I am Matt Pruitt, and this is the Radical Exchanges podcast with James Evans.

Me too. Thanks for inviting me. Yeah. No. I'm really excited about what you're doing in RadicalxChange, and I'm excited to, you know, explore ways in which some of the work that we've been doing intersects with these possibilities.

So I was, an anthropology undergraduate, anthropology and economics. I was really interested in understanding ideas and culture at scale, and I went to Harvard after my undergraduate at Brigham Young and ended up taking a couple of classes in network analysis and kind of the analysis of graphs. And even though they were being used at that time exclusively to study social networks, I got really excited about the possibility of representing ideas and cultures at scale through some of these kinds of mechanisms. And and I think that kind of kicked me off on a journey that's been, to me, a really exciting one. I got my PhD in sociology with a real interest in computational social science and computational methods at Stanford, and I've been at Chicago ever since and have really been interested in this intersection of understanding, on the one hand, social life and society and with a special emphasis on on innovation and discovery and invention, you know, using large scale computational methods and and kind of the array of sensors which these systems provide us with. But on the other hand, turning those things around and using computation to reimagine some of the things that we're involved in societies. I mean, not only innovation and invention, but governance more broadly. So, you know, how do we really think about, you know, kind of using social ideas and patterns and network dependencies to better kind of, like, compute and understand our world, and then how can we use those computations to reach back and and, and better produce queries that we want. So, so I've been in Chicago for about seventeen years and have, but have traveled around different areas. I would say I spend I spend a lot of time developing an area that I call the science of science, which is really using the full imaginary from all the sciences and beyond to apply to the improvement of science itself and social science and the humanities. Now how do we really understand and tune our collective knowledge rather than tuning and improving and optimizing our individual knowledge because those often clash. You know, if we really succeed at one, we almost invariably fail at the other. If we really systematically fail an individual knowledge, we typically do much better if we're seeking collective innovation. So I've been really interested in kind of pushing and promoting that. In fact, I just got off a conference with the National Academies and others that is trying to kind of promote this at the federal level because private actors and private data come in and out of this game. Microsoft, for example, was serving all kinds of amazing data, and then they just pulled out of their contributions a month ago. And and it's just you know, it's it's interesting to to see how with the importance of data, it's, it it has has not yet raised to the level that, that the state and appropriate analytics have become a real federal priority, but, hopefully, they will. And then the second thing is just a real interest in computational social science and social computing. So how to think about understanding the social world better through large scale data and computational methods, but understanding improve proven computation with social insights. I mean, societies are socially computing constantly, and it's just kind of a recursion. And so how to how to think about new opportunities that arise in both areas in computation and the social sciences that will allow us to to do better at what we want to do and to evaluate what it is that we want to evaluate. So so that's you know, those are some of the things that I think about. I'm also really interested, I would say, in in trying to find ways to measure and think about knowledge and policy that transcend human capacity to either imagine or fully understand and conceive of. And I think this flies in the face of efforts to create complete transparency, which I think are important at one level. But but I also think that one of the exciting things about the use of computation and, you you know, kind of novel platforms is actually thinking beyond the the limits of human capacity and kind of human negotiated agreement. And that's gonna mean that some of our explanations are not fully accessible to everybody in the collective and possibly, at times, to any human in the collective. And I think, you know, coming to terms with that is critical for being able to build artificial intelligence, for example, that's really complimentary rather than competing to human capacity. So these are things that I think a lot about. I think about a lot about augmented and amplified human intelligence, a new vision for artificial intelligence. I think a lot about computational social science and its converse kind of social computing, and I think a lot about, leveraging all these things to basically produce better knowledge and invention.

so my answer initial answer, might be pretty unsatisfying in the sense that, you know, it's for the most part, like, my actual day to day work is is not driven by, like, some massive overarching telos, and that's actually cultivated. Because my experience is that when we're trying to solve a particular problem, we often come up with solutions to other problems. And sometimes those other problems were problems we didn't realize that we had in the first place. And I like to keep myself open to those other possibilities along the way. And I would say that's often and as I've done that, you know, my own team has, you know, has grown, to think about these kinds of problems in in, in new ways and also to see the fact that some of these things that maybe that that we can't fully understand and, in fact, embracing that in very specific ways that I hope we'll have a chance to talk about could allow us to explore new collaborative possibilities, you know, between people, but also with bots and intelligences that we cultivate and and, you know, to maximize productive diversity. So I think, there is a bundle of contradictions even in just what I said. So for example, on the one hand, we want to use some of these tools to understand ourselves better. Yes. That's right. We wanna understand what our values are, like, from our behaviors. I think inverse reinforcement learning is an exciting possibility where reinforcement learning is, you know, you have a reward and you're basing you're you're building an automated robot, typically a deep learning neural net, though not necessarily, which ends up behaving through reinforcement in the way in which you design potentially unleasable and novel settings. But by inverse reinforcement, then you see what people actually do, and you can infer what their awards must have been for them to do what it is that they actually did. So there are all kinds of ways that I would say large scale computation and data on many human behaviors, human decisions, human activities can allow us to understand human logics, human motivations in new ways. Like, absolutely. On the other hand, however, I'm suggesting that we try to build machines that go beyond just those human capacities in search of benefiting those human capacities. So if we wanted to you you know, if we were in the feudal era where guilds controlled what it is that we do, you know, how we do it, you know, for in the goldsmith's guild, then they defined how it is that goldsmithing was done. They regulated the methods by which it was done. And, in some ways, the story of capitalism is the breaking up of these guilds and the reorganization of production to massively increase complementarities between different kinds of persons. And, and that led to all kinds of exploitative outcomes, but also to some powerful and continuing collective gains for individuals. I mean, I think the question about its benefit to the environment is, like, an appropriate and ongoing question. But I think, you know, here, you know, to achieve our goals the best we can achieve them, it may be, like, the least useful to create things that behave like we do. In the same in the spirit of capital you know, in the spirit of reorganizing ourselves, increasing specialization and diversity, so that we can actually reorganize ourselves around accomplishing things that produce useful benefits. But what does that mean? That means cultivating capacities that we don't have. That means imagining capacities that we might not be able to imagine. That means shifting from, I would say, a kind of a control approach to creating artificial intelligence to a caring approach where we're actually exploring alternative values and alternative capacities which could be complementary to ourselves. They could also be harmful. It's not that this isn't a program that's fraught with risk, but it's also fraught with enormous possibilities. So I think my work has shown, at least to me again and again, the powerful value of diversity in human collaboration, but also in economic collaboration organization and the political sphere and civil society. And so this basically creates almost like a Star Wars, like a new cosmic landscape on which to explore and push the limits of productive diversity.

Yeah. Absolutely. And just one one way in which this kind of really just hit me, this last quarter at the University of Chicago, I worked with Daniel Holtz, who is, a member and leader in the atomic, in the both in the atomic scientists, which created the doomsday clock after World War two and and really has been kind of focused on trying to publicize existential harms to society. And so we spent the entire quarter bringing in people to talk about engineered pandemics, to talk about nuclear holocaust, environmental devastation, you know, cyber collapse in quality, how these things exacerbate AI, evil, you know, robots. And at least for me, it it really came home just the way in which economic prosperity and productivity over the last two centuries from the first industrial revolution has has taken a toll on the long term capacity of the globe. Mhmm. And so I think holding that telos open, and I would say even keeping open my sphere of moral concern to include, you know, other sentient and nonsentient, agencies from the environment to even the bots that we're engaging with, cultivating, actually, I believe, can increase the potential for, powerful synergies. And I think we can demonstrate some of those.

Sure. Yeah. I think there are a number of evidences in my kind of one subfield that I plan of of the, social science of innovation, economics, and sociology of innovation that show that there is a trade off between, basically, productivity and novelty or creativity, in these spaces. And so I can show and and have work that either has come out or is in the process of coming out that, you know, large scale hierarchical teams and, you know, densely connected scientific communities are much more likely to advance quickly, to produce more papers, to produce more short term attention in the context of science and technology, and to move very quickly, very slowly, which is to say to, Right. You know, rapidly advance in things that are heralded as advances by others, which reinforce, those advances, which create these kinds of socio technical bubbles, which are akin to the kinds of political echo chambers that we see talked about and derided and and that we increasingly have caused fear in the context of our civic life. And so I've shown, for example, and and have a number of new pieces that that show in a kind of a deeper way. For example, that if you have highly dependent groups of scientists that are collaborating on a set of scientific finding or if you have, people that are drawing from the same methods repeatedly or have the same expectations that they're much more likely to come up with the same answer, but it's much less likely that that answer is gonna transit across new groups to new settings. If they're in biomedicine, it's much less likely that that's actually gonna work in the clinic. If it's in pharmacology, it's much less likely that it's gonna you know, these aren't robust insights. These are fragile insights because things that appear to be collective assessments, you know, there might be a 100 papers, are actually not independent experiments. They're like one experiment repeatedly. So it might be, like, 1.2 or 1.3 experiments over and over again in the same way that people inside of, social media bubble, an echo chamber hear their own opinions, like, echoed around as if they were all coming from some independent source, but they're all channeled, you know, through a common exposure to each other and to external media environments. So in some ways, diversity is a way of breaking those bubbles of artificial inflation artificial inflation of certainty. And we've shown we have work that's coming out that shows that, you know, if you take those things into account and you update your certainty accordingly, that you have a much better prediction of what things are gonna hold in the future, what facts to build on. And we've looked at this in in political and civil context as well. So, for example, if we look at political diversity in the context of constructing Wikipedia pages, informational encyclopedic pages, people with different political backgrounds. And we initially assess this through the kinds of contributions that they make to political or liberal sites, but then we can validate it actually with just direct, question and remark surveys that the more diverse communities end up producing higher quality pages. If we search for if we select in the diversity politically, then the political pages are deemed to be the most higher quality, and social issues pages, which have some, you know, political content are also of higher quality, and science pages, which have less political content, are also of higher quality but less so. If you select on scientific diversity, then scientific pages are perceived to be of much higher quality, etcetera. So that's another case in which, again, these assessments of quality are made independently, and blind to the process by which these pages were produced. They just are broader. They're deeper. That's what it is that, that these encyclopedic pages are trying to optimize on. But then also in things like search, scientific search, but also cultural search. If you're searching for new music, if you're searching for new art forms, if you're searching for new to the world innovation or new to yourself innovation, breaking these bubbles has a dramatic impact. So there are all kinds of glass walls, glass floors, and glass ceilings in the context of science, which are conditioned as a function of, just the way in which fields are incentivized and funding is handed from the old to the young, and education is basically constructed as a control mechanism, unintentionally, for the most part. I mean, there are some who intend to control, but I think for the vast majority, people just intend to, advance and follow their intuitions and follow their discipline, which means a commitment to a relationship between a problem and a solution. Right? They're committed to that relationship, that juncture, between problem and solution. And, and, and they're I think our argument and our findings show that they're diminishing returns to those commitments. Right? So it's not that fields are a dumb idea, you know, that you can't study, you know, the world through looking at molecular scales in chemistry. That chemistry wasn't a good idea. Chemistry is a fine idea. It's just the boundary between chemistry and biology is an underinvested in it. Right. It has a marginally better return to investment in the same way that, you know, if you're searching, you know, for news musical experiences, that you're gonna reach outside, that zone to maximize the likelihood that you find something that appeals or emotionally regulates or do whatever it is that you're doing with your music or your art. So I I think, you know, we've shown a number of cases that, it's not just that scientific systems and cultural systems are just efficient systems, you know, where everybody just gets more of what they want, through interaction. But, people get a lot of what they want through interaction, but it it constrains them from search. And that constraint in search is a real constraint that we can think beyond potentially. I mean, it's and and it's actually it's hard for individuals to think beyond because there are costs to thinking beyond. And and I suppose I'm bleeding into the next question, but machines don't necessarily have the same hang ups, or we can design them not to have the same hang ups that humans have. We can design them to accept, insults and failures in ways that actually can hedge against some of our own exploration. And I should say I mean, I'm just talking now in this very conceptual diversity way, like diversity of skills and diversity of experiences. And but there's other work. For example, there's a wonderful proceedings, the National Academy of Sciences paper from last spring by Dan MacFarlane, Dan Jurofsky, and some others at Stanford that shows that this is also true for diverse identities. This is true for, you know, for women, for people of color, for people who are at the periphery of the scientific system. They're more likely to produce novel findings. The transaction costs in producing those in collaboration with other groups sometimes takes longer, especially if it's, for example, from other countries. There's a language barrier. These take more time. But the productive confusion that results effectively reduces and permeates the boundary of these socio technical, sociocultural, barriers. Right? That they're bubbles, you know, that we invariably build around ourselves and gravitate within. So that's that's the underlying idea of diversity is that we're we pack animals as humans Mhmm. Incentivized to build communities around us, and those communities are often communities of ideas and communities of of people and institutions that are reinforcing those ideas artificially, that keep us from empirically exploring alternatives. And so diversity is about kind of, like, weakening the boundary of that systemic, I would say, species level bias.

thing because I think part of the challenge is demonstrating. We have to demonstrate over and over again in more and more context that diversity matters because it's unnatural. I mean, it's we're trying to you know, we're drawn to our tribe. We're drawn to reinforcing and creating a a tribe. This is what human societies do. This is what endothermic mammals do. You know? Like, how do we you know, how do they outcompete dinosaurs? Well, by, like, regulating their, you know, thermal environment so they could, like, hunt at night. You know, they could we can increase our ability to predict the future. And but that creates this this very self destructive cycle where one of the best ways for us to be able to to produce and predict our success in the future is to to, you know, ensure our success in the future by building friends and allies and creating a bubble that is going to absolutely kind of keep the benefits in and, you know, keep keep the riffraff and the barbarians out. And, it's not like there's one demonstration. I think we're gonna have to demonstrate over and over and over again in each new context, that diversity matters. And it's not necessarily arbitrary diversity matters. I think we have to be open to the fact that in some circumstances, we might not want communicative diversity. For example, if we're in a controlled environment, we need to make quick decisions if we need to although the way in which skills and knowledge interlock is sufficiently complex that, we can't posit those things theoretically out of hand, to begin with. So I would say the awareness thing is something that's gonna have to happen over and over again because it's it's it's fighting against a tide of trying to to, you know, reinforce and and enforce our ability to predict the future. And the best and cheapest way to predict the future is to to limit our uncertainty about that future. So I think that's that that demonstration is a scientific program that is going to be perennial and constant. It's not gonna end. But it's also more than just a demonstration. It's also a discovery. So, there's different kinds of diversity that matter for different kinds of tasks and problems. And so, actually, like, discovering, on the one hand, the diversities that matter and, two, cultivating those diversities. So and how do we cultivate those diversities? Well, I I think probably at all three of the levels that you mentioned. I mean, the values where we need to value difference. And why do we need to value difference? We need to value difference because it's actually flying in the face of something else, which is productivity and short term return Mhmm. Almost invariably. And so we need to to formalize those values in such a way that we recognize the sacrifices that they will necessarily entail. Like, if we don't recognize the sacrifices, then the sacrifices are gonna come back to bite us. There are forces that are gonna try to reduce diversity to maximize short term gain. So I think, you know, one is values. A second is institutions and the structuring of incentives. And I think those incentives are really are kind of twofold. Right? Because we're talking about two things. We're talking about, you know, diversity, and we're talking about the way in which diversity historically actually cultivates risky bets. Right? It cultivates new possibilities which are probably wrong. But if they succeed, they could be transformative because they're under likely to have been imagined or explored. Right? The things that are inside our bubble are are, like, overtilled. You know? I mean, we've we're drawing from the same well. You know? The well is going dry. And I would say that's actually one of the awarenesses is that the well is going dry. So Mhmm. You know, so there was a a really nice piece, by Benjamin Jones, a great economist of innovation at at Northwestern University's Kellogg School of Management in which he talks about the death of the renaissance man. He shows how people are getting older as, you know, they get grants. They get older when they get awards. You know? It just the entire scientific system is aging. His interpretation is that there's just diminishing margin returns to exploration. We just we're we're we're hitting fundamental limits. We we understand as much as we can understand. It turns out there's almost a footnoted, finding in that piece that, in the rate the the age of invention does not go up. A first invention does not go up. Okay. An invention is very interesting, right, because there are no fields in invention. In patent, in invention, you're actually trying to find something that's interstitial, you know, that is, you know, that violates fields. You're specifically optimizing on novelty. And so another interpretation of that diminishing margin returns is just that we've picked all the low hanging fruits from this tree, but there are other trees that, like, we're just we just have not looked, at the forest of trees. And so science is, like, committed to a few trees, let's say, the disciplines that we've inherited, but they're like a whole bunch of others that are out there. So I think that incentives can both valorize different kinds of diversity, which are proven or unproven, reservoirs of possibility. And it can also facilitate the incentivization, or the incentivizing of risk. Right? So individual scientists and innovators, but also people in the private sector, you know, entrepreneurs can engage in more risky behavior that will invariably lead to more individual failure and a higher potential for collective success. Right? For somebody to hit on a new combination of physical or social or technical principles that facilitates really a transformative gain, technically or socially, that that can bring real value, you know, to to the world, hopefully, with with, like, minimal costs. And I would say just one other notion on this this idea of incentives and institutions, there's this, I I think, a a beautiful idea by Charles Sanders Peirce. He was a philosopher at the end of the nineteenth century, beginning of the twentieth century, who advocated, among other things I mean, he was really a a polymath, a semiotician, philosopher, scientist. And he advocated for this idea of abduction. And, and my gloss of what abduction is is that, you know, it's like the collision of induction and deduction. Right? Deduction is where you have axioms, principles, like, really solidly known facts, and you extrapolate from them. Like, you explore their implications in a whole bunch of different domains. Induction is where you search for new things, in the wild or in the artificial world wherever. You know, you're just looking. Your mind is potentially wide open, and, and you discover those patterns, and then you generalize from those patterns. Right? And, you derive laws that could be true for the things that you observe. And then abduction, he argued, you know, was was actually associated increasingly with, like, every modern advance. And, and that's where you basically do something that you expect to get a certain answer, and then you're wrong. Right? An experiment fails, a observation, you know, goes awry. You know, you you have an expectation, and you're wrong. And that surprise ends up inspiring some new hypothesis that that really changes the field, that moves it forward. So this is, you know, Fleming identifying, you know, in the context of his office, penicillin, you know, through just you know, it was it was a messy office. Right? So it was it was an it was an accidental experiment, but he was there prepared to take advantage of it. And there are stories of a few individual scientists who do take advantage of these serendipitous events. So abduction is is where you do something, and you're wrong. Right? And so and so that that, you know, being wrong inspires you know? And and we've got, as I mentioned, a bunch of individual inventors and discoverers who have, like, historically done that, and it's like, wow. That's interesting. But not I would say not so many. Like, there there are there are a few examples of of people who were surprised and discover and, you know, and why are there few people? Well, I think the reason there are few is because the person who's inside the bubble, who understands the contradiction, who understands the problem is, in some sense, the worst person to have access to the informational resources that would make the surprising thing unsurprising.

Because if they were outside the bubble, then it wouldn't be surprising. They wouldn't even see the surprise. You know? So it actually so it requires a conversation across this boundary of people inside and outside to, like I like, problems are constructed inside. And so but but solutions, in some ways, almost by definition, need to come from someone outside those spaces. So I think, you know, we can, cultivate diversity. We can create events that bring people together, that bring insiders and outsiders together. If we do that, we need to have a public discussion about failure because if we do more of that, we're gonna have more failure. We're gonna spend less money more money on on more failures, you know, on stuff that doesn't actually work. And and we have to do that if we want to achieve outside success.

Right? And Yeah. And that's and that's why we need policies that actually valorize that. In the academy, I mean, we're horrible at that. It's about it's about individual genius and consistency of individual productivity. I mean, you know, there are few places in history that have done better at this, like, I would say, the historical Bell Labs, in its golden age. Now, again, it was also a monopoly, and and there there's so there were costs to, the kinds of structures that it had, but it did basically validate and reward groups at on group level productivity. And so so there are a number of cases in in in, you know, the nineteen fifties, sixties, and seventies in which, you know, individuals were completely unsuccessful, but they wandered around that landscape. And across the group, there were a number of interesting and surprising outcomes. So finding ways to, like, again, you know, like, bet on on on big bets, you know, specifically, bet on diversity and bet on risk or to locate resources in such a way that they allow individuals to do that. So, for example, if I give you a grant that covers six years, you're more likely to take long bets than if I give you a grant that satisfies your two years in the same way that a senator might make longer bets than a congressperson, right, in in arguments on the floor of, you know,

you know, in a sentence and then leaving. Right? Yeah. And they're under under rewarded in the systems that we currently have. So, you know, you can show if you if you have someone like I mean, you know, a genius like Richard Feynman who stays arguably I mean, he did do some flitting, but he stayed squarely for the most of his career in one space. And he got enormous returns in the same way that Noam Chomsky who Right. Who controlled American linguistics for half century. You know, he enforced credit. Whereas, you know, people are floating in and out. There's no enforcement. There's no need for people to cite and appreciate them, so they're underappreciated. And, so it is you know, contributors are one of the things that's striking to me so I I recently did a kind of a survey experiment with some colleagues. And, in that survey experiment, we just were interested in what influenced people and analyze this in the context of COVID. You know, like, does being in place matter for anything? You know, does being together matter? And so we, you know, randomly selected papers at different levels of the hierarchy of citation. We randomly selected sites within those papers, and we asked people what were their influential sites, how well did they know those sites, etcetera. And then we we checked at them, you know, in a computational space so we could identify how different these papers were. One of the things that was striking was that the citations that you make to people within your institution are much more likely to be further from you Sure. Than citations at other institutions, and they're much more likely to be ascribed as much more important for your ideas in the context of your argument. So so it's like those local diversities that you cultivate through whatever, through the soccer team, through surveying a committee, you know, through I mean, people outside your department but inside your institution are the people who are the most likely to influence your work. And in some ways, the least likely to, in in some sense, be credited by, you know, your work or benefit from the credit from your work. So building institutions that facilitate and catalyze this and building incentive systems that find other ways for accounting, for appreciating, and appreciating the possibility, not just honoring successes, but honoring attempts or shots on goal. And then I I guess the third thing that you asked about was the technical thing and, you know, how is it that we can, you know, build platforms and use technology to to reach, you know, beyond our capacity. And I would say the the one that I've been increasingly excited about, not only cultivating and preserving diversity, but but generating diversity. You know, how can how can we generate, you know, things that our educational system, our commitments to reinforcement just miss systematically. And not just I mean, you know, those are, like, social things that we can arbitrage, but can we also hedge against just biases of of human reason? So, for example, we know you know, if you go into a psychology textbook, there are, like, twelve, fifteen, visual anomalies, you know, where we see something. We see, like, a bias, something that's not really there. We see a line that's longer than, you know, we thought it was gonna be. There are few of those. And so, like, we know we know there's, like, this bias, and and if we construct things in a certain way, we can fool people. But we have we have a lot. We, you know, visual, logical, all these other biases, that it's not just one person or the other person like we all have. How can we identify these in ways and build diversity, such that we build, collectives that break our bubbles, you know, and propose things that we couldn't have conceived of or proposed. You know, like a a physician. I mean, we're really interested increasingly, in cocktails, like in medical cocktails, but nobody is thinking through, like, a 10 dimensional, 10 component cocktail. Like, you know, it's explosive, the number of possibilities. We can't keep it in our heads. You can't write it in a paper. And yet, that's like an enormous frontier of possibilities that machines could search through that were were very unlikely to without them. Right. And so this is connected, if I understand, to the idea that,

Yes. Does that resonate? Yeah. Yeah. Yeah. It's about creating, you know, aliens, basically. Now as different as possible is is challenging frame. I mean, we're we're trying to surf this chaotic boundary. Right? Because if they're too different, you know, if they propose things that are you know, we we completely can't understand. We can't plug into our current systems. If they don't share a language Right. Then, you know, then it's gonna be difficult for us to to engage with them and benefit from them. There's a place where we don't understand what's going on. That's right. That's right. And and now at the same time, they're too close, obviously, then they don't give us much benefit. It's there's this phenomena in biology called hybrid vigor. Right? Yeah. I mean, anyone who has a a mutt, you know, or a hybrid you know, a healthy hybrid dog knows that, you know, you put together cross bubbles, genetic bubbles, and, and you do better. You know? Systematically, you do better. You see it burst, you know, in the plant kingdom and animals and a whole bunch of different places. And that's the same thing. Like, if it's too far, you know, then, like, the egg and sperm don't gestate. You know, if it's sexual communication is impossible. If it's too close, then there's, like, no added benefit to riffle shuffle in the genes. There's no benefits from sex. But we don't know how far that boundary is. Right? I mean, there there are a lot of things that we used to think were species in the plant kingdom, and then it turns out that only when we started thinking about and being concerned about things like, you know, super weeds, you know, that we're hybridizing with GM crops that we realized that a whole bunch of things are hybridizing and interacting that we didn't think could interact. Like, we didn't think that that was possible. So so I I'm advocating actually really you know, we don't know what the outer limits of those communicating, those incommunicative agents are. And something might be able to kind of produce something, you know, after a while of incomprehensibility that ends up being really useful. Right. And so, you know, like, how how do we engage in, you know, the patients, and how do we hedge against very real and natural fears against creating things that are either useless or even, like, potentially malfeasant, right, or harmful? I and I think the challenge is that that's that, ultimately, that's gonna have to become a a fear that we we manage rather than a fear that we hide and repress.

Yeah. That's a great that's a great point. And it's and I think that point and surfing that boundary is the key to sustainable innovation. Right? Because if you just exploit, you know, if you just, you know, recombine, like, all those benefits, we're just gonna, like, harvest all those benefits, then what's gonna happen? And we have a paper where we show this. If if you embed, for example, collaborations in the scientific space, then the conversation collapses. Right? Because if I'm talking to you and I'm talking to the third person, then just the number of independent conversations goes down. The number of things, variables, concerns, concepts that we care about collectively goes down. So you have this weird kind of, like I say weird because it has a certain geometry, but it's it's, it's a complex geometry where there's a collective attention space. And if we're interacting, then that attention space is like a shared comments, and we can eat it up. And so you're right. If we if all of us are constantly talking all the time and try to, you know, benefit from, this cross pollination, then difference disappears. Right. It's the difference which on which that was the benefits of that were predicated. So it's like you need to so in some ways, this is also it's dissing disciplines, but it's it's actually suggesting the importance the importance for continuing difference and minimum communication and, like, independent activities that then can be explored and rehybridized

But nonetheless When it's cut off forever. Exactly. Right. But but sometimes it needs to be cut off for a while, you know, for something. So I would say almost every meaningful speciation in the scientific space has occurred as a result of, of forced exit. You know? So, you know, people organic chemists felt like biochemistry, really, at the turn of the twentieth century was, many of them, that that they were looking for high level enzymatic events. They were kind of glossing over all kinds of underlying organic, reaction differences. And the same for genetic event. Genetic events involved all you know, a ton of enzymatic and other events, and so they're like, oh. But but looking at that different level of analysis, you know and then they cut them off in some ways. They stopped publishing in their journals, but they published in other journals. And and both of those moves happened to be enormously productive moves. You know? I mean, they they I mean, I guess it became a meaningful science, but I I think, you know, it's like sometimes, you know, it it would the best thing to happen for for both of those fields in some ways was to be exiled from their home journals, you know, and to stop conversing with people who saw true and correct limitations to their approaches.

Well well, I would say I mean, I think, you know, it it you need you need to find your place in that boundary. If there's not enough difference, you're gonna hit those diminishing margin returns really hard because there'll be no place to go. You know? You'll hit the top of the tree, and there's no one and there's no one else looking at other trees, and it's just the end. You know? Right. And, so I think anyway. So I I I complete I mean, I think finding that boundary is a is is is a challenge.

Exactly. Yeah. You know? I mean, I love this. I think you're put you're you're you're you're pushing me in in really profound ways. Because on the one hand, we wanna network all of our societies in ways that we can, you know, we can, you know, reduce inequities and that we can enable wealth flows and and communication. But the flip side of that is everyone's speaking the same language, so all the cultural knowledge, which is in all those different languages disappears. Everyone's using and engaged in the same set of political institutions, which systematically are benefiting some kinds of activities and kinds of things and not others. And, you know, so we don't yeah. I say so it's it's very tricky because diversity also facilitates and perpetuates inequality. So we're we live in a in a complex multivalent moral landscape, which values each of these things, but the balance of those values dramatically differs in how it is that we engage in in managing the world. Absolutely.

Yeah. Just just one thing, you know, that that you know, it's it's like elections themselves are a fundamental paradox, right, regardless of how you run them because something is at stake. A decision, an important decision needs to be made. You pull the populace. In pulling the populace, you unleash a social influence process. And so at the very act of trying to pull the diversity of opinions and perspectives, you're reducing that diversity Right. Through this process. Right? After right after the election, people are much more certain and much less diverse collective than they were before the election. And so, I mean, I believe in argument, debate, and persuasion, and yet the scales on which it unfolds also dramatically reduces, in some cases, very productive diversity, which is which is, like, flushed from the system in the very process of of assessing that diversity.

that that touches on this. It's been a while. I'll just say it. Okay. The, So refresh me.

Yeah. Yeah. I think so I think this this is one of those challenges is I I'm I would argue that they're also diminishing margin returns to most particular sets of values.

Right? I mean, you can say I value this, that, and the other, but if you turn the crank on making in a decision making engine, that satisfies those values, at some point, you realize, you know, I I I didn't think I cared about the environment, but, you know, now that, like, all the forces are gone, I I I do I I did care. I should have cared about the you know, like, there like, so any finite, moral commitment itself has, in some ways, diminishing marginal returns to commitment to that moral array. And so, so so so I agree that it does presuppose these these kinds of things, but but this there's a bubble that results by just by picking this, you know, discipline, you know, of, of virtue and pragmatics. I mean, by this, I'm kind of, like, combining, like, the virtue ethics. What's good to do on the one hand and what good does that produce, the kind of utilitarian ideal. You know, like, that's that's your method. You know? You've got a problem. You got a method. So in some ways, we're we're posing these kind of moral disciplines, and they're diminishing larger returns to moral disciplines. So, I think, anyway so I couldn't agree with you more, but it but it it it, in the end, poses the same, the same problem of of cultivating diversities that can hedge against those limits.

So it's interesting. I mean, I'm a sociologist, but I I'm I'm really, I'm excited by AI, not only because of some of these possibilities, but also because and and this is this may be a strange way to get into it. I've got I've got a book that's coming out, hopefully, at the end of the summer from O'Reilly on thinking of deep learning. I'm, you know, systematizing and exploring the development of certain of these models. But one of the reasons that I became interested in them was before they became successful, and and interesting in high performance, I was interested in them in the same way that something like a decision tree is built and not just a decision tree, but a random forest. You know, it's built on the idea of elections. And and it turns out, historically, Oliver Selfridge, one of the key participants in the idea of artificial intelligence, was inspired by voting systems, by social systems, and kind of creating, these ensemble models. For me, connectionist models and, deep learning models and and also, I mean, deep models of other kinds. I I I don't mean just, you know, kind of back propagation deep neural networks, but but especially deep neural networks. They actually suggest a kind of rather than, like, an electoral process where you're just, like, polling everybody, a discursive and influential process by which you develop, like, sub models that are then, like, speaking with and arguing with each other. And you're, actually, if you optimize, you know, front to back in a feed forward neural network, you're you're you're gonna create diversity of models in the middle of that model that then, you know, come back and vote, you know, to kind of produce the optimal outcome. So there's some neuroscience on these deep learning models, which suggests that they're doing some of the exact same kinds of things that we're talking about. So so I for me, the models are interesting because they work, but they also take collective and social principles seriously by, you know, building you know, so g p d three, for example, this is this this, you know, transformer model, by OpenAI has hundreds of billions of parameters. It has a 175,000,000,000 parameter. And that's that's, you know, that's like a bunch of worlds of little people. Each of those parameters are a separate little model inside that space, or their, you know, ganglia of them are, like, separate models. So so I actually came to some of these forms of computation, because they actually reflected, I would say, social principles, and some advances that are taking place in them now actually are rediscovering some of those advances of or or things that are known about social systems and discursive systems. So, I I kind of come to the they have, like, a homological appeal, in the sense that they they take into account some of the things that successful social systems that are computing answers and computing rationality, are doing. And so one of the things that's really interesting about their history is just the way in which, they have achieved success in popularity, in the broader popular imagination is, you know, by the the possibility of them approaching human intelligence. And so I think the computing machinery and human intelligence, Alan Turing writes about and talks about this imitation game, and this really became, it just became an enormous thing. I not just the idea of the imitation game as a whole, but, like, he you know, just ignoring humans per se, but just arguing that humans are the standard. You know? That's the standard of intelligence. And so if we're gonna create intelligent machines, then that's the standard we have to satisfy. And that was, like, I would say, even more deeply captured in Arthur Samuel's work over the nineteen fifties and finally with his encapsulation of it as machine learning that, you know, we're not just gonna target a human like or humanoid outcome. We're gonna we're gonna target human processes. Like, we're gonna use from we're gonna use stepwise human data, for example, to learn a a checkers playing machine, which is what he was developing. And so it's not only, like, you know, trying to achieve the same outcome, but it's trying to achieve it by human like steps in this process. And I think, you know, that was enormously popular games. You know, people engage in games, and that seemed to be, like, a nice model for human intelligence. But if you just push that to its logical conclusion, successful models it it's just like we're just putting, like, a bull's eye on the back of human capacity. And that's you know, it just seems neither the most ethical investment nor the most efficient investment. It's not the most ethical investment because I was in China in the 2019 when I'm trying to remember the the town in the South, but it was a huge automaker that completely automated its, its efforts. And and it put out of work I mean, I don't know if it was '25 or 02/1950 I mean, it was it was it was, you know, at least tens of thousands of workers and maybe over a 100,000 workers. It was it was a big deal in China. We barely heard about it in The US, but it was a big deal in China. And I think the idea that we're growing, machines that are just like, you know, reinforcement learning against human capacities, is is really an effective way to decrease the value of a whole bunch of human capacities. And, and you can see it's interesting. In places where we really value those humans, like radiologists, we've chosen a different path. You know? Right. Like, if if if we just let the machines, you know, run on, radiological images, then I'm not saying the radiologists would completely be out of jobs, but we'd have a lot less employed. You know? But no. No. We were very careful to kind of make sure that we retooled them so that they'd be able to be in positions where they're designing and manipulating these machines. And so we could kind of, like, re, facilitate a a long term tooling that would allow them to retain capacity, but machines to slowly and now more quickly, gain that advantage. But for most humans whose value whose labor we don't care about, we don't protect them. We don't try to find ways for them to kind of, like, reidentify complementarities. So I think that's, you know, that, you know, that's the problem is that we've got this and then and then the missed opportunity is that, like, we're not optimizing on collective benefit, And so we're not, exploring, like, knowledge and capacity, and engagement, you know, that would really, you know, fundamentally, alter the human machine equation. Right.

but it might seem that way for a period of time that is, like, relevant to, you know, the trajectory of human society. Does that make sense? For those individual humans, I think that's a beautiful point. I think that's not the point that I was describing, but I think that's a beautiful Okay. Auxiliary point that that absolutely, absolutely, you know, if I mean and the the illusion part is critical. I mean, I would say that, you know, radiologists actually you know, they're flexible, intelligent humans. We can create a new job for them. They can we can create new capacities for them, that they didn't have or that, but that benefit from the former capacities that they have, because we value them or because we've created a monopoly over their skill. And so they have a guild, and they're they're they're able to absorb and retrain. Whereas, you're right, in these other contexts where we could have, augmented individual skill, we could have created a new brokering of of complementarity. We we create this artificial substitute of logic because of where we've been placing the bull's eye. Right. Absolutely. And, and so, obviously, human skills are going to adjust and evolve and change. It's gonna be a lot easier for them to adjust and evolve and change if, the capacities that we're trying to expand are systematically optimized as a function of their contribution to the collective. In the same way that in some ways, you know, capitalist machinery and, like, large scale organization and these kinds of things actually blew open certain forms of productivity, you know, in the second industrial revolution, right, the turn of the twentieth century. And and so I think when those things also created waste that I think we need to to think about from other perspectives, but I think that, they like, we're trying to explore something

centralization of, for example, you know, data and you know? So so we've got, you know, companies that produce platforms that that generate lots of data. They use that data to drive artificial intelligence or intelligent agent systems. That data centralized, those algorithms are centralized, the wealth, the and the rents that they get from, you know, those algorithms are centralized. And so that creates all kinds of inequalities and all kinds of potential problems. I would say, at the same time, it's a lot and and Cyndal Mullenathan made this point beautifully in a, New York Times opinion in which he stated, yes. These algorithms are biased. This was after a a wonderful science piece, he wrote with several colleagues where they showed that there was a big insurance, algorithm that systematically underscored the ill health of, you know, of participants of color. Right? Because, historically, from historical reimbursement data, LUS was paid on them for, like, the same level of disease. And so going forward, LUS was recommended to to to cover those that same level of disease. So they'd be much sicker to be scored at the same level of sickness. And, but what his argument was, yes, they're biased, but if we undertake a kind of program of machine behavior, where we just monitor their behavior that's really been promoted by by some others, then it's actually much easier to debias them than it is to debias, like, the social system. Like, imagine just telling a judge, you know, you know, you you just here's some data. It turns out that, you know, you're biased against those who come before the bar, you know, who are Hispanic, who are female, who are women. It's all these insidious biases. Can you just stop those biases? And, of course, they don't know where those biases are Right. Where how they change them. I mean, they can look at the data, but they don't, you know, they don't know where they came from. With an algorithm, we don't necessarily may not need to know. We can either change the objective function or we can, like, add a constraint. Those are the two strategies. And so I think centralization actually facilitates the kinds of monitoring and control, which we don't have in a pluralistic and distributed society. But they also, you know, create all kinds of inequities and and lack of distribution and lack of so it's like the same thing, we see manifest over again. I I think yeah. So I I I think it's it's it is it's challenging. It's challenging to but but I I completely agree that, even the platforms we have and especially the the increasingly monopoly platforms we have, you know, so places like Google, you know, kind of advertisement market behavior. I mean, it's like, you know, there's nothing else that exists out there. And so we don't have other kinds of data that could be cultivated into alternative artificial intelligences, which could produce alternative values. Yeah. And so we need a kind of a competition policy without question that facilitates and preserves diversity across platforms, across data, across algorithms. And I and I don't see that happening nearly as fast as I think it should, you know, when I look at SCC and these other organizations that are monitoring organizations at the level of very old categories of, like, horizontal and vertical integration and not at the level of these kind of data complementarities and potential monopolies that are very helpful.

I agree. I I well, it seems to help me understand. It seems this this democracy can mean different things here. So it it could just mean, you know, like, control of the people. It it could mean diversity. We were actually trying to preserve the diversity which is represented within the context of the people, whether or not that's something the people have sought to control or not. Like, what just help me understand your commitment there. I think both of those are important. Right? Diversity of control and, like, diversity of platform and input. Right? And and I think those are, in some ways, almost two independent values because they certainly don't reinforce one another necessarily.

Yeah. And I think and I think to really do that seriously, we also need a diversity of systems. Totally. And and it's but it's but it's very hard in the current environment to achieve that because of short term successes. Right? You have one search algorithm, that, you know, for example I mean, you know, PageRank, is it's a beautiful algorithm because it harvests all kinds of local knowledge from the web. You know? Like, I what what it does, basically, is it allows me, when I know something, I'll be willing to search through a few pages of search results to, to get my answer. If I don't know something, then I trust the local community of people who do know that something. So I just the first one. And, and so I can, you know, I can, like, pick a good restaurant. I can, you know, pick the right citation. I can pick, you know, all these kinds of things. But this have, you know, like, two obvious biases. One, they collapse the world, or they have the potential to collapse the world, you know, so that everybody if everyone's believing in everybody else's, you know, tastes, then it, like, reifies and, like, hardens, like, those tastes. It makes it really difficult for alternative tastes to arise. And the second is that the entire that entire system. Right? There are other ways to organize, you know, search and certainty and assemblage of information and of possibilities, you know, in in cultural domains and political domains, you know, economic and inventive domains. And, but if one at one moment in time is enough better than the others that it achieves monopoly power, then you see what you see with all the big tech companies, which is to say, they cultivate lots of projects. They kill lots of projects. They kill those projects not because they're they wouldn't have been a successful business for somebody else, but because they're not big enough to marginally move the needle. And so in some ways, like, they're the best. It's like it's like Clayton Christensen's innovation innovator's dilemma. You know? Like, they're in the worst position to disrupt themselves. And, so I I I I couldn't agree more that that we, you know, we need we need more democratic control. We and and we need a better competition policy that monitors, like, the diversity of the systems alone so that we have a reservoir of alternatives available to us, in a serious way, which which is is not gonna I mean, under current policies is not the case. Right. And I mean, I think that sometimes

Yeah.

Right. Yeah. Yeah. Yeah. So yeah. So so this, we yeah. We have a if if we have a republic, there's there's a lot of layers Right. There. And each of those layers basically has its own set of heuristics and institutions, which kind of collapse a lot of potential and real diversity in the experiences and insights and the desires, right, of the populace that they supposedly serve. No. I I I couldn't agree more. The government representative Republican, government is not enough, basically, to ensure the diversity

Of both control and and of and of continual inputs Right. Right required for these things to both generate value and for that value to be distributed and appropriated widely. Yeah.

Oh, well, I should say, you know, one thing about, just in thinking about the alternative of, and and and, actually, this is, like, the missing link. It's it's not so much that we it's just from here to there. In the same year, interestingly, that, that this idea of artificial intelligence, like, the term artificial intelligence emerged with McCormick's conference on artificial intelligence nineteen fifty five is when he proposed that. We've got this idea of augmented intelligence, amplified intelligence that also comes out. But what it means at the time, which was important at the time and has become arguably much more commercially important than artificial intelligence were like interfaces. You know? How how can we reduce the friction between individuals and information so we can just make them superpowered? You know? We can make them, you know, the like, they can control with their minds. We can amplify all the things that they might otherwise want to do. Just, you know, just turn up the the efficiency dial. And, and I think I think these intuitions and and insights about diversity, you know, just suggest that they're ultimately diminishing marginal returns. I mean, at the limit, you get to the limit of human capacity, and you certainly get to the limit of your horizon of of, you know, what the possibilities are that you want to explore politically or otherwise. And so finding ways to, like, systematically, break those prisms and burst those bubbles those bubbles is in some ways the next step in that augmented, collective intelligence story, which is is is really, like, cultivating, valuing, generating the diversity required for us to see past ourselves in our current moment. So I think there I think there's, like, a link in history to that. And I would also say it's not like nobody's doing this. I think, you know, current reinforcement learning approaches are ways I mean, you know, these are creating some alien things that produce things like what we want, but through the creative articulation of rewards, do things very differently than what than the way we do them. And I I'm I've been promoting, you know, human aware AIs that that are, you know, like, aware of of the diversity of opinions and perspectives that are out there. On the one hand, can represent them, can understand where they're coming from. On the other, can hedge against them and actually provide and generate novelty to that, you know, in ways that could advance their causes. So I I think these aren't just, like, fanciful ideas. Like, these are actual, you know, machines. So I've got a paper that it's up on archive, but it's, you know, under a few of a few places where we, by just adding human awareness to an existing system, you know, that was published the year before for discovery, you know, we have a 100 to 300% improvement in performance immediately and a validation that, if we avoid, you know, that human capacity, we can discover a whole host of things that that would either never or not until the distant future become possibilities. So I think this is not just social science fiction. This they're actually operational things that can be done. Some people are thinking about them. I think there's much more that can be done that would, facilitate and broaden their capacity. But I can't really I I I'm trying to think if there's there anything, if there's anything else. I mean, I think this there's I I mean, just this conversation alone has generated this question about the the relationship between the willingness to valorize failures or sustain failures and the preservation of diversity. You know? To to me, I you know? I mean, it's like the way these things seem to fit together that I maybe thought about in exactly the way that I'm thinking about it now as a result of this conversation is is you you don't wanna rush into a winner take all situation.

You you don't wanna rush into a situation where, you know, like, one platform or one governance system or one metric or one data source, like, or one search efficiency is, you know, even even if it gives you, you know, those enormous advantages in the short term. Because if you if you do, if you jump too quickly, then you really forsake enormous amounts of opportunity in the long term and and enormous amounts of value that could be redistributed and broadly shared in the long term. So, like, being able it seems to me that being able to, to withhold the desire to place everything on that one short term bet is is gonna be a critical issue. It's just difficult. It's it's it's it's just so and it's so delicious. You know? That's just that's a great return. You know? To science and society and the economy just to to to the creative destruction of of, you know, letting the winners win. And, and so it's but it's like, basically, we need to slow down the process in some sense of creative destruction, so that we preserve the diversity that facilitates, you know, like, new generations of creative destruction and possibility in the future.

Yeah. Or or my telos is illusion. Yeah. Well,

Yeah. But there are other things that are right. Right. You know, besides those we've defined, other things that are no. Absolutely. And and and this is, in some sense, I think, part of the process of discovery. It's not just about executing our values. It's about discovering the values that we should have. And in executing is, in some sense, how we validate the hypothesis that that was a value to pursue in print. So it's like it's like the process unfolds itself only through the realization of of of the process. So, but I and I think opening ourselves to that, you know, is is hopefully opening ourselves to, new opportunities along the way, you know, that would change, and and new voices, you know, that would change, the the space of possibilities. Well, thank you. This was a lot of fun. Yeah.

big

A big thank you to James for that conversation. Thank you to Radical Exchange Foundation's supporters. This would not be possible without you. You can continue to support Radical Exchange Foundation at radicalexchange.org. Thank you also to the producers of Radical Exchanges, Jennifer Marrone and Leon Erickson. Thanks for listening, and have a great weekend.