Super Unsexy By Design: Ben Green on Algorithmic Realism

00:00:31 Andrew Keen: Hello, everybody. One of my favorite all time books is one by a Czech writer, Josef Škvorecký. It's called The Engineer of Human Souls. It's a wonderful parody of the promise of communism. Of course, it's borrowed from Stalin's, definition of the engineers of human souls being writers. But I wonder whether, the idea of engineering and engineers of this human soul has come back into fashion. There was a recent piece by, Stephen Witt who's appeared on the show. He's the biographer of Jensen Huang, the CEO, the chief engineer of NVIDIA, a piece in the New York Times entitled “He's the Most Powerful Engineer on Earth. Here's Why He Thinks AI Is Going Great.” And what it occurred to me, having read the piece, it's a good piece, is that engineers now, for better or worse, are running the world, and it's their way of thinking about reforming the world that seems to be becoming conventional. This all comes to mind because my guest today on the show, is a professor of data science, Ben Green. He has a new book out. It's called Algorithmic Realism: Data Science Practices to Promote Social Justice. I'm not accusing Ben, of course, of being a Stalinist. But, Ben, would it be fair to say that we live in an age of the engineer, for better or worse, and that's where you're coming at us from in terms of your book and your work? You're a professor of information and public policy at, the University of Michigan in Ann Arbor.

00:02:22 Ben Green: Absolutely. We are in an age where the most powerful companies in the world are tech companies and particularly AI companies. And the sort of imperatives of technology development and in particular AI development are really at the heart of how society is being reordered and how the economy is functioning. If you look broadly at the economy, most of economic growth these days is really just, capital investment tied to data centers, and all of the companies at the top of the, you know, stock market are, big tech companies at a level of concentration that we've never seen before. And I'm deeply concerned about the direction that AI development is pushing us in as a society. And so my book, Algorithmic Realism, is trying to chart out a different direction for technology development and, in particular, a different way of thinking about engineering and how to actually develop algorithmic systems and AI systems that work for the betterment of humanity rather than just accruing power and profits.

00:03:33 Andrew Keen: Although, I'm not necessarily a defender of big tech, but they would at least argue that whether or not they have a bad reputation, they're doing their best. Recently, we had, Matthew Botvinick on the show. He's a senior guy at Anthropic. He has a new book, AI and Political Freedom: The Risk to Democracy and How to Respond. And then it's an engineering attitude to how to fix democracy. Is, I mean, you're coming at us, Ben. It's no surprise given your subtitle about the promotion of social justice. You're coming at it from the left. But from the left as a progressive, are you suspicious of big tech or the idea of engineers and engineering being the thing in itself? Are there good and bad engineers in your mind?

00:04:28 Ben Green: Yeah. I am definitely, not rejecting the idea of engineering writ large. And in fact, this book is in many ways trying to provide a new model of what engineering can look like that, is working toward social justice. Because, right now, AI development is sort of weighed down or brought in a dark direction by both the methods of the field, which is sort of the core critique of my book, but also the political economy of tech. And, you know, I think it's very hard to say that these tech companies are doing their best when, there are so many examples of them, clearly having evidence that their systems are harmful and not taking action. Or—you know? And we've seen this for years. Right? Even Facebook, many years ago having knowledge about how harmful social media platforms were for the mental health of teens and continuing to do things anyway. And now more recently, you know, we look at OpenAI's agents hacking Hugging Face and hacking other, websites and platforms. A lot of that can just be tied down to negligence, rather than them, you know, doing their best and making innocent mistakes. And then, of course, they were, really quite slow in actually revealing this information. So I think there is a—I'm deeply suspicious of big tech, and I think, we need to create directions for engineering that are divorced from big tech and have a methodology and a culture that is much more grounded in social impact and societally important questions.

00:06:13 Andrew Keen: Yeah. Let me revise the question.

00:06:15 Ben Green: Mhmm.

00:06:16 Andrew Keen: I'm not defending Botvinick, but he runs a supposedly nonprofit, institution within Anthropic. Who knows whether it really is nonprofit? Who knows much about the institution? It's hard to call one way or the other, and I don't wanna personalize it. But is it possible, Ben, that guys like Botvinick who are working with big tech companies, like OpenAI and Anthropic. Can they do good with their engineering brains?

00:06:54 Ben Green: I think it's really difficult to do good on the inside, of these companies because they are really just trying to push as much as possible for superintelligence and, you know, a massive IPO in the case of OpenAI and Anthropic. And to them, safety is really just something that is a PR problem and a liability problem rather than, I think, a deeply held, ambition. And we've seen now, you know, just over the last few weeks, people leave—whistleblowing and leaving these companies saying that, you know, I tried to work at the inside of a place like OpenAI, and the culture is broken. And they are rejecting concerns about safety, and there's been also report—you know, there was a report in the New York Times about this, a week or two ago about OpenAI engineers raising safety concerns and essentially getting ignored and shot down. So I think, you know, while there are many well intentioned people at these companies, the broader sort of business decisions are leading the way, and it's really hard to make effective progress, I think.

00:08:03 Andrew Keen: But, maybe I keep on repeating the same question, and I apologize if I am. There was a good interview with Ezra Klein, and Jensen Huang on the New York Times. I'm not sure if you saw it. In which Huang was actually quite critical of OpenAI and Anthropic saying that if their products were dangerous, they should shut them down. That's the mind of an engineer. So perhaps from the left, as someone committed to social justice, should we be suspicious, or should you be suspicious of the kind of thinking of men like Jensen Huang, if you can separate him from NVIDIA for a moment?

00:08:46 Ben Green: Mhmm. Well, I mean, again, Jensen Huang is both being critical of those companies, but still ultimately calling for, leaving all decision making in their hands. Right? He's saying, well, OpenAI and Anthropic should just be responsible stewards and not release a model if it's, you know, seems dangerous or seems like it might cause problems. And, I think this is very wishful thinking, and, you know, it's very much a way of defanging calls for regulation and, the sort of broader, sort of tumult that you have right now with people pushing back on AI. I mean, I guess you're sort of also trying to get at this idea of, like, an engineer's mindset. And, obviously, in this case, it's very difficult to separate the mindset of an engineer from the mindset of, you know, a business leader and the business decisions that are being made in these cases. But I do think kind of in general, the mindset of an engineer is very much to assume that there are, like, neat optimal objective solutions to problems and that facts and reason will win the day and not to sort of think about questions of power and who actually gets to make decisions and what their incentives are, or to think about sort of what the value might be in democratizing decision making around these technologies. So, I mean, broadly, this—the engineer mindset is really just to say, the goal—our goal is to build more and more powerful tools. And as our tools get more powerful, they will inevitably work for the betterment of society. And I think that is just a fundamental, misunderstanding of the relationship between technology development and social impact.

00:10:36 Andrew Keen: So who's gonna determine what is good for society or what social, quote, unquote, social justice is? The—that term's in your—in the subtitle of your book. I looked it up on Wikipedia, what social justice is. I mean, it's obviously a term that means different things to different people. I'm not saying it doesn't have any meaning.

00:10:57 Ben Green: Yeah.

00:10:57 Andrew Keen: You've been very critical, surprise, surprise, as I think most of us have of the DOGE catastrophe of Elon Musk.

00:11:05 Ben Green: Yeah.

00:11:06 Andrew Keen: One of the pieces you wrote was “Using AI to Reform Government Is Much Harder Than It Looks.” And you warn of the enormously dangerous consequences of DOGE's plan to push AI across the US federal government. And, clearly, your warnings were right. But where is social justice coming from? You've been critical of startup AI companies like, OpenAI and Anthropic. You're no fan of Elon Musk as most of us are. Where is—and I'm quoting from the subtitle of your book. It's an MIT book, so the subtitle is rather dry. Data science practices. Where are they supposed to come from? From within the academy, guys like you, from private enterprise, from the government, from citizens?

00:12:01 Ben Green: Mhmm. I mean, there's no single place that is going to be the, you know, the origin of social justice. Social justice is something that is a sort of larger ambition of change rather than, you know, like a single output. I think it's definitely not going to come from, you know, the big tech companies. I think it can come from the government. You know, I think there are, ways in which the government is—does have an—a very important role to play, and they sort of are at this, intersection point where some of the things that they do are very bad and some of the things they do are good, and there's no kind of single description that the government is only an impediment to social justice or the single tool for social justice. And then, you know, absolutely coming from researchers and citizens and so on. I mean, I think really right these days, I mean, a lot of what we're seeing in terms of social justice movements and tech are not, really happening in the development of AI, but actually in the pushback to AI development where you have communities pushing back against data center development, pushing back against AI in schools, AI in the workplace. And so there is, and those are by and large, in my view, you know, oriented around social justice and oriented in particular around rebalancing who has power and authority in society. Because what we have with AI development now is a world where we have these small number of companies and engineers, as we've been talking about, who kind of hold all of this power to dictate what does the future look like. And so social justice broadly to me means, you know, balance—providing more of that power and decision making authority, into, you know, regular people and in particularly, you know, marginalized and oppressed groups. So, again, there's not a single answer here, but that's sort of the direction that we need to move in.

00:14:10 Andrew Keen: What about, Ben, this new alliance of left and right populism, which was manifested, I think, most clearly by Steve Bannon and Bernie Sanders appearing at the same conference and seemingly to agree to condemn tech oligarchs and demand AI reforms. Are you concerned with that? There's the—both the populist left and the populist right, if that's the best way to describe Bernie and Bannon, are uniting against technology, not just big tech, but what you might think of as algorithmic realism?

00:14:52 Ben Green: Mhmm. I mean, I'd say I'm a mix of excited and concerned. I think it is exciting to see bipartisan pushback to AI, and we've seen that, you know, with Sanders and Bannon also in settings like data center pushback. And I think that is exciting to see because if pushback to AI is just partisan, if it falls into a left right divide, it's going to be really hard for it to get anywhere because then it's just kind of in the s—the same trench warfare that we see on everything. But the fact that this has been bi—bipartisan is, I think, very important for, sort of paving the way for more regulation to come in the future. You know, I think one of the things kinda more broadly that I am trying to, I don't know, say with this book about technology is that we need to sort of shift from thinking about AI as a single thing. So often we talk about AI in this sort of monolithic way, and, you know, even as myself and we have been doing in this conversation, whereas both, like, AI sort of is actually like an umbrella term for a bunch of different technologies and a bunch of different types of systems. But more importantly, for my purposes, like, AI development is not a single directional thing. AI development could move in many directions. And so I am both opposed to the direction that it is currently going, But, also, do think there is a potential path for AI development to work, in service of people? And those are—to me, those are not in tension with one another. We don't have to choose between AI and no AI. I think the—really, the bigger choice is what type of AI do we want and who do we want to be in control of, developing and managing those systems.

00:16:49 Andrew Keen: So what kind of AI would you like? I take your point that there are no generalizations we can make. There's good and bad AI, presumably good and bad AI companies. A few years ago, Jennifer Pahlka was on the show. She's the founder of Code for America and the author of, I think, a really good book, Recoding America: Why Government Is Failing in the Digital Age and How We Can Do Better, arguing, she was the, I think, deputy tech—technology director [ed.: Pahlka was US deputy chief technology officer] in the Obama administration, arguing that government does have a role to play. What's your take on, I guess, the Pahlka argument? One might think of it as a fairly mainstream liberal argument about how government can use technology to improve society.

00:17:36 Ben Green: Mhmm. Well, I think where I would really align with Pahlka's argument is the idea that we don't need technology to be this incredibly flashy thing. You know, when we talk about AI and—the drive to develop AI is so much oriented around building the most powerful systems, achieving superintelligence or consciousness or AGI. And that—there's actually, like, a very tenuous relationship between a system that is most powerful according to some benchmark or idea of consciousness and the system that is actually useful to people in practice. And a lot of what I talk about is sort of this shift from general AI to narrow AI is one way to think about it, where general AI is the types of things that the big labs are doing where they're trying to build consciousness and have a single model that can do everything, whereas narrow AI is something that is more, specifically tailored to an individual use case. And in doing that, you can actually, like, help people make, you know, just meaningfully better decisions, without it needing to be the sort of be all end all of changing the world. And so I think, yeah, Pahlka definitely talks, I think, in that same vein of just finding ways to build technical systems that overcome real pain points within a operational workflow or a decision making process rather than trying to, you know, automate something entirely or replace humans entirely or build some system with whatever we might think of as sentience.

00:19:17 Andrew Keen: Yeah. Your last book was called The Smart Enough City. I guess you could have called this new book, The Smart Enough Technology.

00:19:26 Ben Green: Yeah. Absolutely. I mean, in many ways, I feel like I've been having the same fights and arguments for many years now. There's a, you know, somewhat frustrating sense of deja vu where, you know, where I—when I wrote that book, that was very much trying to convince, local government leaders that the way to use technology well is not to invest in these flashy smart city solutions, but to, focus on how technology can be integrated in more mundane ways in everyday life. And so, like, one of the basic things that I sometimes talk about is just, like, clean up and manage your data better. A lot of cities have just really messy, disorganized datasets, and there's very little you can do unless you just, like, lay a nice foundation of data management. That's one example of something that's super unsexy, but actually can have a much larger impact than investing in some, you know, fancy AI proprietary platform. And so, yeah, I mean, today, it's much the same. And, you know, even with organizations, I think about all of the millions of dollars that companies have wasted on, you know, quote, unquote, token maxing over the last year as everyone has been convinced, oh, you know, we need to adopt AI immediately and as much as possible or else we're going to fall behind. And then, you know, there's a pretty high price tag that has come with that. So they feel the cost a little bit more viscerally than they otherwise might if it was just wasted time or something like that. But, yeah, even within an organization or a company, you don't necessarily need to use AI for everything or use the most sophisticated model. You need to actually have a real sense of what are the specific challenges that we're facing, and how can we integrate AI or data driven processes into that rather than, just kind of throwing that—you're throwing AI and technology at everything and assuming that it naturally will, improve whatever you're up to.

00:21:28 Andrew Keen: So maybe an alternative title for the book is algorithmic realism is unsexy.

00:21:35 Ben Green: Yeah. I mean, I definitely am not making promises of utopia the way that, the leaders of, you know, the AI developers will talk about superintelligence or curing cancer or solving climate change. I mean, I think those promises are, pretty absurd and, you know, they're not really—especially the—actually, really all of them. But they're making those types of promises, and that gets headlines. And it's not what I'm doing because I don't actually think that's a realistic or grounded way in how technology can or will actually affect organizations and people's lives.

00:22:14 Andrew Keen: Yeah. The Economist recently had a cover story on effective altruism is this century's biggest I—idea. Of course, the effective altruists are not only powerful in, organizations like Anthropic, but also as critics of AI. But I'm sensing, Ben, that you wouldn't be a big fan of effective altruism because it's also, a utopian philosophy or a philosophy with great ambitions. It's a sexy philosophy as opposed to your algorithmic realism, which is by definition unsexy. I'm guessing you're not a great fan of the philosophy about effective altruism, are you?

00:22:56 Ben Green: No. I'm not a—not particularly a fan of effective altruism, but I should say I'm not opposed to utopian thinking. I—actually, one of the books that really influenced me the most in writing algorithmic realism was a book called Envisioning Real Utopias by Erik Olin Wright, who is a sociologist. And that book is all about how to achieve this idea of what he calls real utopias, rather than, you know, utopias that are just—we often think of as, like, these fantasies, but how to achieve utopias that are from the left, you know, socially liberatory, but, actually realistic. And, you know, I think for me, that really helps me sort of provide a counterweight or see the flaws in the way that a lot of, these groups talk about utopia, which is often this sort of, like, snap of the fingers thing that will eventually occur. Like, oh, well, we're building AI. It's going to, you know, cause all of these problems. But, eventually, there will be, you know, universal basic income, and we'll have solved disease, and we'll have solved climate change, and it will be wonderful. But they have no realistic pathway for getting there. They just kind of dangle this promise out in front of us. And what Erik Olin Wright describes is the need for accessible way stations on the way to real utopias, essentially that we need to develop a path towards getting to utopia rather than assuming that it's something that will just emerge out of nowhere. And so that is—you know, what I wanna see in thinking about utopia is not to view it as some endpoint but actually as a goal that we are iteratively working toward and making meaningful progress towards. But, yeah, I think—you know, you asked me, like, where social justice comes from. I think that that goal is in many ways a utopian one. And what I'm trying to build toward is how do we make steps that are both radical in their ambitions, but incremental and realistic in what you can actually do as one person or organization, in a given moment in time.

00:25:08 Andrew Keen: But, Ben, doesn't everyone have definitions of—I don't wanna sound too much of a relativist, but everyone has definitions and disagreements about what social justice is.

00:25:18 Ben Green: Uh-huh.

00:25:18 Andrew Keen: You have it in your title, data science practices to promote social justice. It implies, and maybe this is just in the title, that everybody knows what social justice is, and you just somehow key it into the computer or the algorithm, and it will come out as social justice. Whether or not that's algorithmic realism, it—not everyone's gonna agree that's how politics or policy should work.

00:25:44 Ben Green: Mhmm. Absolutely. I mean, I, yeah, don't think social justice is, something that everyone agrees on or certainly that not that it's something you can just put into an algorithm. It's contested and people will disagree, and we have to kinda sit with that. But, you know, what I'm really trying to work with here is to say, you know, how would you actually develop algorithms in a way that is oriented towards social justice? You know? Like, yeah, not everyone is going to be trying to do that. But let's say we—if you are motivated to do good with technical skills, which a lot of people are, then how do you actually do that in a meaningful way? And how do you teach students, who are learning data science how to approach their work in ways that are more likely to align what they're doing with the public good. And, yeah, like, there is no single definition of social justice. It's certainly not up to engineers to decide on their own. It should be a, you know, a dialogue, a conversation. And I don't think that, again, that sort of, like, endpoint we get to will be, oh, we've all come to a single definition of social justice and, you know, everything is great and no one's fighting about politics anymore. But, you know, I am trying to give practical steps for how we might reenvision the, you know, the work of a data scientist to be more sort of pushing us toward social justice. And, you know, my definition is certainly one that is more, thinking about, sort of economic equality and, sort of redistributions of power in society. And that's what this book is trying to accomplish, and not everyone's going to agree with that. And that's also okay with me.

00:27:37 Andrew Keen: We did a show recently with, an MIT professor of anthropology, of psychology, Sherry Turkle [ed.: Turkle is MIT professor of the social studies of science and technology], a friend of the show. She has a new book out, Artificial Intimacy: Who We Become When We Talk to Machines. She's a self declared humanist. She thinks that we're losing a lot by talking to machines in terms of who we are as human beings, our humanity. In terms of your algorithmic realism and this promotion of social justice, do we need to leave some space in this for the human outside the algorithm, Ben?

00:28:15 Ben Green: Mhmm. Yeah. And so, you know, what algorithmic realism is is focused on is, sort of beyond the specific social justice angle is to say that the way to understand algorithms is to look at the social impacts that they have. You know, it—it's not enough to just say, oh, well, this algorithm satisfies this technical benchmark or has these incredible mathematical properties or abilities when we run it through some test simulator. What we actually need to do if we wanna understand that algorithm is to look at what impacts it has on people and on society. That's sort of the broader conceptual frame of what algorithmic realism is is pushing for. And so, yeah, I'm—I am quite concerned about, how AI is being sort of chatbots and how people are building relationships with these tools. I think in particular, just because we can—we've already seen evidence of how, it is isolating for people. It is not actually beneficial to their mental health. Certainly, there's been the more high profile cases of chatbots, you know, sort of prompting people to kill themselves or harm other people. And I do fear for both the sort of social skills and the identity formation as well as the critical thinking abilities of, you know, sort of everyone, but especially the younger generations like the students that I teach and, how much they are turning to AI for friendship, for advice, even just for accessing basic pieces of information. I think that that sort of has clearly documented negative effects on people, even beyond sort of, like—I don't rely so much on sort of just, like, a sort of general humanist idea of, like, oh, we need space for the human just because, but actually focus on—well, if we just look at, you know, the evidence of what these interactions are leading to, it pretty clearly seems negative.

00:30:22 Andrew Keen: Ben, can you give me one or two examples, concrete examples of data practices that in your view at least can or have or do promote social justice?

00:30:33 Ben Green: Mhmm. So the broad, sort of framework—I mean, the—so a lot of the work that I describe in the book is, well, a, working closely with communities and nontechnical stakeholders. So a big piece of it of doing data science for social justice is actually to recognize that you don't have all the answers. Right? You need to work closely with, other stakeholders, whether that's, you know, the end users of your system, but also the communities of people that are going to be affected by a particular system that you're building. And so that might be—you know, if you're working with, you know, a public health agency, who are the social workers, and what are their actual problems that they're dealing with, and what are the challenges faced by the community that they are trying to work on. So definitely building in more of a sense of humility in data science and working closely with—

00:31:33 Andrew Keen: But, yeah, I take your point. But could you be a bit more specific? I mean, you brought up health care and health system. How is this being implemented that will do good in your view?

00:31:45 Ben Green: Mhmm. Yeah. So one of the, sort of stories I tell in the book and that I can talk about from my own work was, a project I worked on with, the city of Boston when I was working with them as a data scientist. And, there was a goal to use data science to improve response times for the emergency medical services department. So, essentially, you know, the group that when you call 911 for a medical emergency, they respond, and we wanted to, you know, help optimize the efficiency of that group. And, you know, the initial idea was, okay. Let's develop some sort of algorithm that would optimize ambulance deployment and make ambulances get to incidents a couple of minutes faster on average. And that sort of, like, would be a very typical approach. But the approach that I took in this case was to really build relationships, with the stakeholders and get more of a sense of what was actually going on and sort of dig beneath the surface of a superficial understanding of the data. And so what I discovered, through these conversations was that the sort of response time and efficiency issue was actually an issue more to do with inequality and policy where the reason that Boston EMS was receiving so many more calls for service than they could handle was that so much of the increase was incidents related to people struggling with mental illness and drug addiction and homelessness. And so if all we did was simply optimize response times and ambulance deployment, we would not have been actually providing support to the—these very vulnerable individuals in the community. And so what we were able to do by both analyzing data but also thinking about the sort of larger organizational workflows within EMS was to create a new team that was specially trained to respond to these types of incidents, to handle someone in a moment of crisis and, you know, provide connections to counseling and social services and homeless shelters. And so, you know, that—that's just one example of how by—as a data scientist working in a way that's not just sort of like I can optimize everything, so a sort of traditional engineering mindset, but stepping back and saying, what is actually the problem here? Let me go talk to the people who are involved in this and learn more, can open us up to, ideas and responses that still are using data. I mean, I had to do a lot of data analysis in this project, but actually can do more to respond to meaningful challenges and inequalities that are present in the world.

00:34:39 Andrew Keen: Yeah. And you answered my question perfectly. I asked for a very concrete example. You've given it, but it's a very narrow one, as you know, and that was the nature of the question. Can we—and I use this word with a degree of irony, Ben. Scale, that example, because, of course, the idea of scale is one that Silicon Valley has popularized. It's an increasingly, controversial idea. But could you—could we scale that one example of the ambulance and create local policy or national policy, involve governments, big tech companies perhaps with—that are nonprofit or certainly with, some credibility on the nonprofit side. Can we imagine a new deal, a twenty-first-century New Deal with AI to address so many of the problems of inequality and injustice in America today?

00:35:37 Ben Green: Mhmm. I think we can. I mean, that's very much the ambitious in—of this book is to scale that type of project, not in the sense of, you know, that one solution. We need to scale that up to everywhere, but to say, we can scale the sort of locally targeted solution, in many places. You know, I think a lot of that comes down to sort of how we're training data scientists. And so one of the major focus areas for me with this book is to really change how data science is taught so that, you know, the next generation of professionals in this field are actually sort of equipped and have a sort of intuition to ask these questions and build these relationships rather than just jumping into some, you know, algorithmic solution. I think another really important direction is for governments to build more, in house teams for doing technical development. And some cities in particular have had this for, you know, ten or fifteen years, but, you know, you need to have embedded technologists sitting in the government agency or really sitting with the government stakeholders to understand, okay. How do we really solve a particular process challenge and make this system or policy better. So often the government relies on external vendors. And there's both the problem of what the interests of the vendor are often not really aligned with the policy goals of the agency. But, also, there's just this really big divide that is present between the vendor and the agency. And so you're—you often get these technical solutions that aren't really tailored to the specific data or workflow within an organization. And so a lot of the failures just come from that disconnect. So I think, like, those are the types of things where, yeah, I think you can scale these types of methods and teams, to a lot of places and in particular, you know, government agencies at all levels.

00:37:50 Andrew Keen: So final question, Ben. We're coming up to the midterms that probably gonna be seismic, probably shift the nature of power and conversation in America. And then we'll be, on the run—in the run into the presidential election. Lots of talk on the left about what progressives should and shouldn't do with tech. You've already noted that you're ambivalent about this weird Bannon Sanders alignment on AI. What kind of policy would you like to see AI companies? What kind of policy, sorry, would you like to see progressive candidates, left centrists, maybe, Ossoff or those types take on AI. The Economist recently had a piece about a no rules approach to AI as mad politics and worse policy. What kind of rules approach to AI would you like to see that will enable social justice in a realistic sense, and to borrow your term, algorithmic realism?

00:39:03 Ben Green: Mhmm. Yeah. I mean, I think someone like, AOC has been talking about AI in a more sophisticated way than Bernie Sanders. I think she's—

00:39:16 Andrew Keen: More hard [unclear], actually.

00:39:18 Ben Green: Yeah. I mean, she's more skeptical. I mean, Bernie Sanders, like, really buys into some of the, like, the superintelligence and, the claims that AI is going to replace all of the jobs and the sort of, like, doomerous rhetoric of these companies in a way that, you know, I am more skeptical of those particular claims, and I think someone like AOC does a better job of situating the power of these companies sort of more broadly. So I think one big direction is on really holding companies liable for the harms that their AI tools create. Right? We have these conversations about, like, rogue AI right now, but these are not rogue systems. They are actually doing exactly what they are trained for. And if they are hacking into, you know, a government server, OpenAI should be held liable for those things. And OpenAI should be—

00:40:09 Andrew Keen: Yeah. And that's Jensen [unclear], for better or worse. That's Jensen Huang's point too.

00:40:13 Ben Green: Yeah. Yeah. So you need that type of enforcement. You need, you know, sort of, strict policies like, that really level the playing field in terms of negotiations around data centers where, you know, so often you have a race to the bottom between states in terms of tax breaks, and you have communities that are really disempowered when they're fighting against the data center just because of, how much money these companies have to kind of get into litigation. And, you know, if you're a small township of a few thousand people, you just—whether or not you have a strong case, like, it's just really hard to get into that type of fight. But I think there's also a broader shift that we need to think about in terms of even our model of AI development. Like, I think the sort of the US approach is both—right now has been both very, obviously, very corporate, but very scale oriented. The whole idea is, oh, well, we can just get bigger systems with more data and more compute power, and that will naturally be whatever we need. And I think we've—we're really hitting a sort of threshold in terms of how far the scaling approach to LLMs can actually get us in terms of socially useful or even economically productive AI systems. And so I would like to see a more sort of government support for a more diverse ecosystem of AI development, that is striving for more, like, small scale and efficient models that aren't so expensive to run and, you know, reinvigorating funding agencies like the National Science Foundation. Right? Like, if you want the US to be a leader in AI development from the left or the right, you should want, the US academia to be a place where there's a lot of money for researchers to actually train and study and do—you know, build AI systems, so that we are continue to be on the cutting edge of research, from that sort of broader perspective, not just cutting edge of, sort of the labs where they're like, well, we have all the data and all the compute, so we're just going to, like, outmuscle you because our sausage factory is more powerful than your sausage factory. I think that's a very shortsighted type of industrial policy that is setting us up to, you know, sort of not be innovative in the long term.

00:42:36 Andrew Keen: Speaking of sausage factories, Ben Green's algorithmic realism is, in his language, super unsexy. I think he's right, which makes it credible, down to earth, sensible, maybe not suitable for the current insanity of American politics. But, Ben, great conversation. You're very down to earth, and you bring a lot of these issues. You make these issues much more concrete than either the utopians or the doomers. Thank you so much.

00:43:08 Ben Green: Yeah. Thank you. This was a lot of fun.