AI and Political Freedom: Anthropic's Matthew Botvinick on the Risk to Democracy and How to Respond

00:00:31 Andrew Keen: Hello, everybody. It is Wednesday, October 7, 2026. We live in strange times, apocalyptic times, or at least the—on the part of many people, a fear of the apocalypse. And when you get the details from them on that supposed impending apocalypse, it's usually either associated with the death of democracy or the taking over by Silicon Valley of all our lives, particularly associated with AI, the apocalyptic flavor of the month these days. My guest today, Matthew Botvinick, understands this. He has a front row seat on this supposed apocalypse, although I'm not sure he thinks of it in quite so apocalyptic terms. He works for Anthropic. He's a scholar and a gentleman, and he has a new book out, came out yesterday, AI and Political Freedom: The Risk to Democracy and How to Respond. Matthew is joining us from just north of me, Marin County, in Belvedere, just on the bay. Matthew, congratulations on the book. It took you a while. How long did it take you to write?

00:01:47 Matthew Botvinick: Thanks, Andrew. It's great to be with you. Yeah. It's something that I started working on, during time I was spending at Yale Law School 2023–2024. And, it's quite remarkable how quickly the world has changed around the project. In fact, I had to update my thinking about what was plausible, what we had to worry about most, as the project went along.

00:02:14 Andrew Keen: You couldn't get our mutual friend, Matthew, to, write the book for us, Claude?

00:02:19 Matthew Botvinick: No. I—I'm not down for that. I will say, Claude, I asked Claude to write many things for me, as part of my work. It's wonderful at summarizing, areas of the scholarly literature or introducing me to an idea, or area that I'm unfamiliar with. But, no, I still believe in human written books, and this is one.

00:02:45 Andrew Keen: Oh, that's good to know. The book is called AI and Political Freedom. Is democracy, in your view, leaving AI out of it for the moment, is democracy in America in crisis?

00:02:58 Matthew Botvinick: Well, I mean, I pivoted my whole career because I had that specific worry. I think regardless of which political party you adhere to in the United States, or if you adhere to neither, of the major parties at all, you have to be worried about the trend over the last, fifteen, twenty years toward a disregard for what were previously taken to be fairly well established political and legal norms. You have to be worried about the concentration of power within the executive branch. Again, I think stating the problem that way, there are people on both sides of the political spectrum who, you know, who will see signs of those trends from their own perspective. They may identify different, political actors. They may identify different political events. But I think there's a general sense that the foundations of democracy in the US, and by the way, in many other places around the world, feel less solid than they used to.

00:04:10 Andrew Keen: Yeah. Strikes me, Matt, at least in the way you presented that you see this as a bipartisan problem coming from both sides. A lot of people who come on this show suggest that the real problem is coming from the right. Do you not believe that? Is the problem not an ideological one? It's more broadly a crisis of democratic institutions, norms, ideologies.

00:04:35 Matthew Botvinick: Well, I mean, taking the point of view that I, you know, that I try to take up in the book, I'm trying to talk about problems that are coming at us in a few years. So they're problems that are not on the agenda. Well, we can talk about whether they're coming on the agenda actually, you know, this month, but, problems that we're gonna be facing down the line. So I guess as a strategic matter, I'd like to talk about this as something that we should all agree on across the political spectrum. Something—you know, to the degree that we believe in the foundations of democracy, you know, we should recommit to particular actions to deal with this possible future. One thing that I will say is, that's a little—maybe a little bit more substantive and connected with current events. Many of your listeners will have, watched or at least read reporting about this recent appearance on the same stage in, you know, in back to back talks of, Bernie Sanders, and Steve Bannon, at a meeting in Washington that was about the risks of AI. I think that was a very, important and telling moment about where we are, in terms of the politics around AI, specifically with respect to, you know, left versus right or Republican versus Democrat. The fears that AI is creating about job security, about social upheaval, about just, you know, honestly, what it means to be human, are ones that I think are starting to eat away, at, people at—across the political spectrum. And in fact, in this case, for two people who are really at opposite ends of the political spectrum. I have concerns about that political moment, which we can talk about, which will relate to, populism. But, yeah, I think the issues that I talk about in the book and that I'm personally concerned about really do go beyond, partisan alignment.

00:06:39 Andrew Keen: Yeah. And that meeting or the appearance of both Bannon and Sanders on the same stage—

00:06:45 Matthew Botvinick: Yep.

00:06:46 Andrew Keen: They condemned, at least according to The New York Times, tech oligarchs like your boss or your distant boss, Dario Amodei at Anthropic, and demand AI reforms. Do you take it personally as a relatively senior person at Anthropic that guys like, Bannon and Sanders are coming after you, describing you as tech oligarchs?

00:07:11 Matthew Botvinick: I don't take it personally. Working in the AI industry is a complex—it's a complex proposition. You know, people in the AI industry—I don't know whether people will be surprised to hear this, but people in the AI industry share many of the same concerns, and, you know, are focused on many of the same issues about our shared future, that you read about in the newspaper with attribution to, you know, ordinary Americans. I don't speak for Anthropic here. I wanna be very clear about that. I'm speaking my personal voice. My experience of Anthropic is that, you know, many people around the organization and up through the leadership actually welcome this public debate. This is a conversation, in fact, that people who care about AI safety have been trying to get going for quite a long time. And, of course, as many of your listeners will know, Anthropic in particular has been honestly asking for more, government regulation, partially as a way of coordinating with the other labs to slow the pace of AI development so that we can focus more on AI safety. So, no, I don't take it personally, and I think, my feelings are probably similar to many other people in the AI industry. At the same—

00:08:26 Andrew Keen: Let me come back on that.

00:08:27 Matthew Botvinick: Yeah. Yeah. Sure.

00:08:28 Andrew Keen: Richard Waters of the Financial Times, longtime tech correspondent, also based in San Francisco, asks in a piece [unclear] last week why Anthropic's IPO feels so weird. Now I would go beyond that. I mean, Anthropic sounds weird because it's full of guys like you, for better or worse, Matthew, who claim to be almost critical, not just of AI, but of the company itself. You had a—an X post recently, where you talk about being concerned about, quote, unquote, concentration of power. I mean, you're talking—you work for a company that's one of the two superpowers in—certainly, startup superpowers, a company that's rumored later this year or early next year to do a $2 trillion IPO. So let me ask this. I don't wanna personalize it too much because, as you say, you're not here to represent Anthropic. You're writing a book. But is there something weird about our current moment where so many people who work for OpenAI and Anthropic, including Dario and Sam, seem to get away or at least try to wear two hats simultaneously both as supporters of AI and of critics, not just of AI, but of their companies?

00:09:52 Matthew Botvinick: There's a lot of complexity there for sure. Let me say one thing about my understanding of, Anthropic's mission as the term is used. Again, I'm speaking here in my personal capacity. And then I can say a word about why I personally, you know, how I personally made the choice to work at Anthropic and how I see that work. So Anthropic, if you dig into its history, was founded, by a group of people who were very intensively concerned about AI safety and about the potential societal impacts of AI. And the decision that they made was, that if they were going to have an impact on, you know, making that go well or as well as it can go, the way to do that was to actually be on the frontier building AI systems because in a sense, that gave them a seat at the table in these discussions. So, I think it was a complicated idea. It was an idea that was guaranteed to put them in uncomfortable, positions. And we're definitely at a moment where I think it's sort of publicly visible that there are tensions in that, you know, in that mission. But in any event, that was the idea. Right? That was—that's not the idea behind all of the AI labs, but Anthropic specifically chose that path. And, you know, for what it's worth, my experience of Anthropic is that people are, you know, pursuing that original idea, you know, quite sincerely. Let me just say a word about why, you know, why—what it feels like for me to be working in the AI industry. And then maybe I can say, you know, we can take this where you wanna go, but I have some thoughts about the relative roles of industry and government, and maybe that can bring us back to the issue of democracy, more directly. So my attitude about working in AI industry is, you know, I started doing what we now call deep learning research, I guess, before it was cool in the early nineteen nineties. And back then, it was really part of cognitive science and neuroscience.

00:12:11 Andrew Keen: That was the ice—was that the ice age of AI, basically?

00:12:14 Matthew Botvinick: Well, it was before that. It was the dawn of the field that was called connectionist modeling back then. It went through, yeah, it went through what was called the connectionist winter, later on. And it was only aft—and, you know, through that whole period, it was simply an area of cognitive science and neuroscience research. You know, nobody really believed that it would lead to, important commercial products or have, you know, societal impacts. We were doing it as a way of modeling how the brain operates. But here we are. And, you know, sometime around 2012, there was enough computing power and enough training data, and computer scientists and engineers started coming into this field, and scaled these methods up and, really revealed what many of us suspected was possible, but which hadn't yet been realized, which is these amazing, capabilities. So, you know, when I look at my own personal, you know, life trajectory, I've been following the development of this technology, you know, since early on in its development. And I guess I would say I feel a certain amount of personal responsibility, having been there, and been involved all that time to do what I can to make this phase of the technology's, you know, trajectory go well for society. Of course, I also have very serious concerns about, democracy, you know, which is something that I cherish.

00:13:46 Andrew Keen: Yeah. And I wanna get to that. Let me just very briefly just—really more for my own clarification than anything else. You are part of Anthropic's—of the Anthropic Institute. You are one of the founding hires. You are, you were, as you say, you were a resident fellow at Yale Law School. You got background at Google, professor of, neural computation, and you're joining the institute to lead its work on AI and the rule of law. That was the official announcement by Anthropic. What exactly do you do then there?

00:14:26 Matthew Botvinick: Yeah. Well, I mean, we're making it up as we go along, but the mo—here's the motivation. So the institute, is a, you know, a group of teams inside Anthropic, each of which is trying to think through some aspect of how AI, may impact society, may impact the—our economy or, you know, the way that we deliver education, aspects of our social life. So it's—you know, you can think of it a little bit like an internal think tank, which is there to help then steer the way that Anthropic develops our products, affects the things that we advocate for, externally, including, different forms of regulation. And in the case of my team, we're thinking through essentially the political impacts. But let me unpack that because that term is—you know, it can mean many things, especially in the American context. So there's a team at—in the Anthropic Institute that works on economic impacts. Some of your listeners may have seen, what Anthropic calls the economic index, which is, a regularly updated public report about, measurements of how AI is impacting the job market and other aspects of economic life. So it's there to share information that really is only accessible from inside Anthropic otherwise, that may be useful to voters, may be useful to people in civil society, may be useful to policymakers. That's the idea. And then, the idea for my team, I was invited in to start this team by one of the founders, a guy named Jack Clark.

00:16:10 Andrew Keen: English guy. Right?

00:16:11 Matthew Botvinick: Yes. Yeah. Was to think through the political impacts. But I—you know, the way that I think—I—when people hear the word political, especially in the US, they think about political parties. They think about Democrats versus Republicans. They think about particular political issues. Like, what about immigration? Yes. That's the economic index I was referring to. Absolutely. What I mean by political is something a little bit more like, what a political scientist would mean. So, my team is interested in, how AI is gonna interact with what you might call the foundations of democracy, the things that allow us to have political discussions, the things that allow us to have free and fair elections, the things that allow us to hold our, you know, our elected officials accountable. So, for example, how is AI gonna affect the political marketplace of ideas? This is—you know, it's a long standing, insight into the nature of democracy that in order for democracy to work well, ordinary citizens have to be able to discuss things openly without worrying that they're gonna be thrown in jail for their political opinions and without worrying that they're, operating in a setting where information is unreliable or is manipulated or is propagandized. So one question that my team is interested, for example, is how is AI gonna affect that marketplace of ideas? When people ask AI questions about politics, do they get even handed answers? What questions are people asking, AI systems about politics? These are questions that we really need to dig into. We—things we need to measure in order to foresee how AI is gonna affect our—the democratic institutions that we wanna protect. And I don't—when I say we, I don't mean Anthropic only. I mean, we as a society.

00:17:59 Andrew Keen: Although some people, Matt, will be watching this and thinking, here's a guy who's part of this institute. Presumably, you got involved with Anthropic. You got stock. You're very well paid. You're probably about to become very rich, like everybody who works for Anthropic on the IPO. Why would we trust the Anthropic Institute? Why would we trust Matt Botvinick to say anything that will somehow undermine, Anthropic?

00:18:27 Matthew Botvinick: That's a great question. There's no reason that you should trust, anybody from Anthropic. You just need to listen to what people are from Anthropic say just like anybody else. There's no reason to trust any newspaper. There's—the whole point of a marketplace of ideas is it's not based on trust. It's based on strength of argument. It's based on, information shared. It's based on, opinions expressed. And everybody needs to take those, you need to take those—

00:18:57 Andrew Keen: Oh, no. You're presenting this idea of a marketplace of ideas as some sort of basic truth. That is also contestable. It's the position of some people who think about politics. Some people would reject the idea of a marketplace of ideas, but that's another matter. Let's talk more about the book, Matt. You talked about trust. I asked you why we would trust you, and you—you were very honest in your response. You said you gotta listen to what we're saying. You begin the book with what you call democracy and the new distrust. Is this the big danger? You talked earlier about, Bernie and Steve Bannon appearing on the same stage? These are guys we may—many, many viewers and listeners might not quite like their ideas, but they are very good at being trusted by their own political constituencies. Is what's gone wrong in your view? You said you're a student of democracy and of American politics. What's gone wrong with trust? Why is there this thing you call in your first chapter the new distrust?

00:20:03 Matthew Botvinick: Yeah. I think it's the impact of the actions of, many leaders across many democracies to undermine the foundations of, democratic governance. There's a pattern that's been unfolding over the last ten to twenty years that, sometimes referred to as the democratic recession, the global democratic recession, where after decades of a growing number of democracies, you see that number shrinking. You see measures of democratic freedom across multiple governments, now regressing. You see elected leaders in democracies, taking steps to dismantle, what had been, regarded as trustworthy, and durable, institutions. An example would be, Viktor Orbán, during his time as leader of Hungary. There are other reasons that people, think that faith in democratic institutions may be declining, and that—those have to do with the efficiency of, the delivery of state services. Just the basic impression that, you know, maybe democracies are just too inefficient, to deliver, you know, the services, and, you know, to make the gears turn. And it—it's in this setting that people start to find authoritarian government, alluring, which is a big concern. One word about, Bannon and Sanders, though, is that, yes, I—you know, they are—they stand as trusted, spokespeople for particular, political subcultures. But at the same time, I think both of them, the language that they use is the language of distrust. It's the language of populism. It's the language of there are ordinary Americans. There are, you know, us, the real people, and then there are these unaccountable elites. And, you know, it's a tricky moment because as you say, there are highly empowered individuals and organizations in—especially in American, on the American scene. So it's hard to bat away the term elite, as if it were not relevant to the discussion. On the other hand, this language is the classical language of populism, and it's been used by leaders over, you know, over the decades to undermine, political freedom, to drive a wedge, into society, to separate people from one another politically, and also to undermine faith in political institutions. So when you say everything's being hijacked by these elites, then you're in a position where peep—people become politically vulnerable to appeals from demagogues. Right? People who say only I can fix it. People who think—who say, look. These institutions that you've been working with, they're outmoded. They're inefficient. They don't work for us, the real people, because they've been hijacked by these elites. That's the moment at which, elected leaders can draw on what's sometimes referred to as the demagogue's playbook. And, you know, again, Viktor Orbán is a perfect illustration. We could talk about the specific things that he—

00:23:36 Andrew Keen: Although he's no longer in power because he's very—

00:23:38 Matthew Botvinick: That's true. That's true. And we can talk about—by the way, we can talk about what—

00:23:42 Andrew Keen: Your point on this, but, yeah, some people are gonna be watching and thinking this guy sounds really conservative, just a defender of elites. I mean, the foundations of the American system is based in part, at least, on a distrust of elites, colonial elites.

00:23:56 Matthew Botvinick: Yep.

00:23:57 Andrew Keen: Jefferson challenging one kind of economic elite or another, Jackson, even FDR in his own way in terms of his critique of Wall Street. So isn't democracy also one element of democracy? Requires, in a sense, a degree of distrust. Maybe not the epidemic of distrust that we're now experiencing, but, a democracy which is entirely trusting is probably no healthier than a democracy which is entirely distrusting.

00:24:30 Matthew Botvinick: I couldn't agree more with that. So just to be clear, the term as I use it in the book specifically is about a distrust of government, a distrust of especially democratic institutions. It's not about a distrust of, say, corporate power or distrust of, other forms of concentrated power in society. And that is something that—a degree of trust in democratic government is something that has changed very much over time. As you say, it's an inherent part of democracy and certainly part of American democracy over the years, that it involves a distrust of, other forms of elite power. So for example, you might compare what's going on right now, say, with, the movement against data centers. You might compare that to aspects of the progressive movement of the late nineteenth century where the focus was on, railroads, right, where the idea was that there were these, business elites who ran the system, the railroads, and, the, say, the farmers who were dependent on railroads for shipping their goods were simply at the mercy of these economic elites. And, you know, democracy worked in that instance, and the kind of distrust you're talking about worked in that instance in order to, drive policymakers to, promulgate, reforms and regulations that actually did serve the public good. There's one important variable here, that I wanna highlight, which is specifically that faith in democratic institutions. So when you have this populist, sentiment, right, when there's a concern that things are going off the rails, ordinary people are being pushed aside, and elites have hijacked the system. As I said, that can actually go quite well in a democracy. Does—the progressive movement is an example. But that depends on there being a certain amount of faith in democratic institutions themselves to sort through the issues and to dry—and to, you know, to, bring about reform. When you have a populist atmosphere combined with very low degree of faith in democratic institutions, that becomes a rather dangerous situation. That's a situation in which authoritarian appeals become appealing. For example, you know, to take an extreme, instance, Weimar, Germany, where, people felt that the system wasn't working, elites were in control, ordinary people were being pushed aside, but there was very little faith in democratic institutions, and that provided an opening for authoritarian leadership.

00:27:26 Andrew Keen: So how would the Musk assault, the DOGE assault on government fit in? I'm assuming that would be a manifestation of the kind of unhealthy distrust that you write about and think about.

00:27:39 Matthew Botvinick: Yeah. I mean—and, of course, that was, you know, Musk was invited in to do that work. Right? Was invited in to take actions that, you know, I think it's safe to say were in many cases unlawful. And in any case, were, you know, were disdainful of ordinary, you know, the ordinary processes that had been established for running the administrative state up until that point. So, yeah, that's a good example of what can happen. Of course, if you look at other—so let's go back to Viktor Orbán. You know, when he came in, his first actions were to revise the whole constitution, in order to remove, barriers to executive power. He took steps to change the structure of the judiciary so that it was much harder for the constitutional court, to question his actions and so forth. So this is the demagogue's playbook. This is—you start with populist appeals. You start with, by shouting, the elites are—have hijacked, our country, and our democracy, and, ordinary, like, the real people, are being screwed over. Usually, when those appeals are made, there's a grain of truth to it, and it often involves labor, disruption, which is another thing we should talk about in connection with AI.

00:29:07 Andrew Keen: So would it be fair? And we've done a number of shows on this, Matt, as well. I'm sure you're familiar with some of the literature, people like Jeff Kopstein and John Rauch. But what guys like Orbán or maybe even Trump do, Modi in India, is rearchitect the democratic state into some sort of patrimonial vehicle for their own interests, their interests of their cronies, their families, their allies. Is that one of the consequences of this new distrust is the replacement of the democratic legal state in Weberian terms with a patrimonial state?

00:29:46 Matthew Botvinick: Yeah. Exactly. So, you know, it's—it starts with populist appeals that allow the leader to kind of divide the populace, and frustrate, political collective action. And then once—according to this playbook, once, an aspiring authoritarian is in office, there are two, like, high level goals. One is to reform the way that government works, in order to entrench their own power, and the other is to muzzle dissent, right, to attack civil society, to attack the press. And, yeah, we've seen both of those happen, in a number of places over recent years. And I've mentioned Hungary a couple times. That's not the only place. So we've seen similar moves in Turkey, some related actions in India. Of—and it's not only from the right. You know? Venezuela—I mean, Venezuela is, of course, a very good example.

00:30:51 Andrew Keen: Maybe Cuba in a weird way. So let's—I mean, you're not the first or the last to have this critique of democracy. It's a good one, Matthew. You've obviously given it a great deal of thought, but I'm sure you're familiar with, as I said, the work of guys like Jeff Kopstein and—

00:31:04 Matthew Botvinick: Absolutely. Yeah.

00:31:05 Andrew Keen: —John Rauch, good friends of the show and of mine. Let's get to your expertise, which is AI. So is AI gonna be digging a deeper hole for us, or is it the solution? I mean, obviously, that's the question that you ask in the book. Yeah. Is that a fair summary given the pressures, the challenges of democracy in the late twenty twenties?

00:31:31 Matthew Botvinick: Yeah. Absolutely. So first of all, you're, of course, absolutely right that the—this demagogue's playbook that I'm describing is not something that I'm—that, you know, that I came up with. However, the structure of my book looks at the ways in which the advent of powerful AI systems might intersect with that strategy. That—that's really what—you know, one way of thinking about the program of the book.

00:31:56 Andrew Keen: Do you think, be—before we get to that, in terms of how AI can help or maybe hinder—is it coincidental that we—we're having this great crisis of trust in our traditional rational legal institutions at the same time as we have this AI revolution in smart machines? Hegel famously talked about, the—I can't remember his exact words, but in—at dusk, the crow of history, or the bird of history. I can't remember the exact terms. Crows. [ed.: Hegel's image, in the preface to the Philosophy of Right, is the owl of Minerva, which spreads its wings only with the falling of dusk.] In other words, history has its own logic. You know, you may not be a Hegelian or a Marxist, but you don't have to be one to understand that things don't always happen by accident. Is it coincidental that we have these two things going on at the same time?

00:32:53 Matthew Botvinick: It's a fascinating question. I—I'd have to reflect on that. I think to some extent, it may be coincidental in terms of the timeline of the difficulties of democracy and the development of AI systems. On the other hand, now that both of those two things have arisen side by side, they're being interpreted culturally through, you know, related lenses. Let's put it that way. So when AI driven, job loss begins or even anxiety about AI driven job loss, which is something that we can hopefully talk more about as we go forward with this. When that happens, it's being seen through the lens of this democratic recession. Right? You know, when people worry about their livelihoods because of AI, you know, their political reaction to that is of—is naturally gonna, play out in the setting of this moment of political distrust in democratic institutions. So whether these two things came together for a reason, is something I'd have to think more about. I—I'm open to the idea. I'll think more about that. But now that they have come together, you know, they're not being seen as independent relationship. As—

00:34:11 Andrew Keen: And they're—that—they're entangled like two balls of yarn. You talk about democratic recession. That, of course, is the term of a very distinguished scholar of democracy, Larry Diamond, old friend of the show's, down at Stanford. So your book deals with, your—the chapters of the book include, the age of AI phase one, the age of AI phase two. As you noted in the introduction, this AI revolution is happening a lot quicker than any of us imagined, including insiders like yourself. Let—

00:34:42 Matthew Botvinick: It's true.

00:34:42 Andrew Keen: Maybe you can provide us with two scenarios in broad terms, Matt. The best and the worst case scenario with AI. On the one hand, how can AI save us? Is it a—can it address this new distrust and create a new age of what you call democratic stability, if that's a good thing? I'm—was a little wary of the idea of stability.

00:35:09 Matthew Botvinick: Yeah. I—so let me answer that by kind of mapping it onto the two phases that, are mentioned in the chapter, headings in my book. So, I'll answer your question about, a worrisome scenario in the short term, and then let me talk about the longer term and how, things might be brighter there. I'll just say in passing that there are positive and negative versions of that—of the outcomes in both of these phases, but this will be maybe an easier way for me to proceed. So in the in this—in the immediate term, say, over the coming twenty years, that's what I refer to as phase one in the book.

00:35:53 Andrew Keen: Which is, I mean, in Silicon Valley terms, twenty years is 20 centuries in other—

00:35:58 Matthew Botvinick: Yeah. Yeah. Yeah. So, I mean, the timelines are just incredibly hard to think through in this—in this setting, partially because it's hard to know how the technology itself is gonna develop. But there are other variables that are maybe even harder to predict. For example, how quickly AI will diffuse through the economy, how quickly employers, will, adopt AI and use it to replace la—it's just very hard, to predict these things. So let's just say for the sake of argument, we're talking about the coming twenty years. What I mean by phase one is a period in which AI starts to cause serious job displacement, which is a future that I think is, you know, is coming at us. And during—

00:36:42 Andrew Keen: Already has, or we're in the very early stages of it.

00:36:44 Matthew Botvinick: Yeah. And there's a lot of debate about this, and we can get into that. But my personal view is this is coming. Let's prepare for it. So in that setting, my argument is that there's a significant danger from AI, not because it creates some, you know, exotic new form of threat to democracy, but rather that it raises the risks and the stakes of vulnerabilities that democracy has essentially always had. So—and what I'm referring to is this demagogue's playbook. It's the hijacking of democratic, states by, aspiring authoritarians. These are concerns that even the founding fathers of the United States worried about. So what do I mean by raises the risks and raises the stakes? Here's what I mean. So we've already talked about one way in which AI raises the risk of this sort of, authoritarian, direction, and that is people feel insecure. People feel vulnerable. People look for help from government leaders to take care of problems that they can't take care of in their own private lives. Here, it's insecurity—economic insecurity, job insecurity, and more broadly, just insecurity about one's position in society. If AI is doing all these things that we used to do, you mentioned even writing books, then, you know, what—you know, who are—

00:38:06 Andrew Keen: What do you mean, "even," man? Yeah. We know how hard it is to write a book. You just spent the last few years doing this. Particularly, you know, writing books rather than "even."

00:38:17 Matthew Botvinick: Yeah. Particularly writing books. Good. So that's one sense in which the risk, even though these are, in a sense, traditional long standing risks to democracy, the risks are elevated. What do I mean when I say it raises the stakes? This is a different problem. What I mean by that is if we act—if these—the cons—the dangers that I'm talking about actually materialize, if we have an—if authoritarian leaning leaders, demagogues who come in and try to capitalize on this anxiety in a way that is hostile to democratic institutions. AI gives them tools. So let me give you two examples that I focus on in the book. One is AI will cause national security crises. For example, it will cause, moments of, crisis around cybersecurity. In that setting, governments can declare states of emergency. In fact, in the US, there have already been states of emergencies declared about cybersecurity. States of emergency are not inherently bad. They're not inherently anti democratic, but they're very easy to abuse. They're very easy to exploit to authoritarian ends simply because they're intrinsically designed to expand executive power. And then the sec—a second way and a more serious way in which AI raises the stakes is that it is an amazing tool for surveillance and censorship. So once an aspiring authoritarian leader comes in, we—I talked about an important ingredient in the demagogue's playbook is to change the system in order to entrench power. Well, AI is a wonderful tool for that. It's a wonderful tool for domestic surveillance, for keeping track of political opponents, for silencing those opponents, for preventing coordinated action and protest. We see this, of course, in the present day in places like China. So in the short term, this is what I'm worried about, is that AI creates a demand for populist demagogues and then also hands those kinds of political leaders these pretty frightening tools that would allow them to entrench power once they have it.

00:40:35 Andrew Keen: Okay. So that's the worst case. So let's get to the best case. There was an interesting piece. You may have seen it in the New York Times, a few days ago. About religious scholars who met with Anthropic people, maybe—you may have been one of them. What they heard stunned them about the way in which some people, at least at Anthropic, are seeing your AI, Claude, as a kind of god for better or worse, is one way, and I don't wanna bring in the god stuff to this conversation because it's probably not that pertinent, but is one way for AI to solve all this, to fix all these big problems is having this objective, quote, unquote, god that everyone can trust as the state. Is that the long term vision?

00:41:22 Matthew Botvinick: No. Please, no. No. I—that's not my—that's not the vision that I favor. So I—

00:41:26 Andrew Keen: So that's another—that's a long term warning. So what's—what could AI—how can AI help in the short term in your phase one? What can it do right?

00:41:37 Matthew Botvinick: So in the short term, I—so let me just quickly say what I think the defenses are against the risks that I just mentioned because they don't really actually have that much to do with AI. They have to do with reforming our institutions with, for example, fixing the way that emergency law works in the US or surveillance law, and their analogs in other countries. So I don't—I—I'm not a, you know, what you might call a techno solutionist in this regard. I think in order to protect democracy, we need to think about democracy and think about democratic resilience. Having said that, I actually do think that AI could be a wonderful tool for enhancing democracy both in the short term and the long term if we can avoid these risks that flow from AI. So first of all, let me just respond to the scenario that you painted a moment ago. I—and I'll—I'm—as elsewhere here, I'm speaking for myself in my personal capacity. I really don't like the idea of AI systems taking over the key functions of the democratic state. To me, the essence of democracy is that it's, government by the people. It's, you know, it's we the people. And I also think that it's important for us to keep track of which activities that we're engaged in involve moral innovation. So when we make policy, when those among us who serve as judges, contribute to, settling legal cases, there's a certain sense in which that is moral—they're exerting moral agency. They're exerting their humanity in those moments. They are leading other people, as members of the polity. That's the beautiful thing about democracy to me. It's government by the people. And, I don't wanna see AI take over those roles. So for example, I personally don't like the idea of AI judges, except in the most narrow administrative, setting.

00:43:48 Andrew Keen: Well, you've listed a lot of stuff that you don't like. What do you like? Where can AI help us remake democracy? I assume you share some of the ideas of people like, Jen Pahlka, who was the deputy, head of SEC [ed.: Pahlka served as U.S. Deputy Chief Technology Officer in the Obama administration] under the Obama administration, who, unlike a lot of other people on the left, still has a great deal of faith in technology—

00:44:11 Matthew Botvinick: Yep.

00:44:11 Andrew Keen: And AI reforming government. Is that the core area where AI can make the state more efficient, fairer, more transparent?

00:44:22 Matthew Botvinick: I absolutely think that if AI is deployed very carefully and designed properly, it can play that role in a transformative way. So, you know, simply in the delivery of services, I think AI could be, incredibly, incredibly, impactful. And by the way, I touched on this point early in our conversation. When democracies are able to deliver services efficiently, that bolsters people's confidence in those democratic institutions, and it makes democracy stronger. So it's not just a question of people getting better services. It's also a question of making democracy, more durable. But it goes beyond that. So, we've been doing some experiments at Anthropic looking at, how good Claude, might be as a, essentially an analyst to support, legal reform.

00:45:14 Andrew Keen: So good Claude as opposed to bad Claude?

00:45:19 Matthew Botvinick: This is good Claude, I hope. So in these analysis—

00:45:22 Andrew Keen: [There] is a bad Claude too, if there's a good Claude.

00:45:24 Matthew Botvinick: Well, you know, everybody's working on AI alignment and trying to make AI good in all the ways that we want it to be good. But we don't know perfectly how to do that yet, and I think that, everybody is aware of that and as a—as an area of work in AI. But let me talk about this legal analysis work. So the idea was, we simply said to Claude, can you please read the entire US legal code, including, any relevant court cases, regulations, and so forth, and point out things that it might be smart to change, in order to keep things working well, as AI itself becomes more powerful. And Claude came back with a 200 page document analyzing many areas of law and identifying cross cutting issues, that it was fascinating to read about. Now, of course, the next step is for human legal experts to read this sort of stuff and decide whether there's, you know, there are ideas in there that we, in fact, consider wise and actionable. But one example is simply pointing out that, what counts as a record inside the federal government, is gonna be quite complicated with the advent of AI. So it really matters what an—what a record is because that is something that, feeds into, laws that make, internal records, available to the public, like, freedom in—Freedom of Information Act, guidelines. It also, governs what Congress can see in exerting its oversight role over the executive branch. So it really matters what the definition of a record is inside the government. And so one of the points—

00:47:15 Andrew Keen: Clinton's definition of sex, it depends, I guess, on where you stand or where you don't stand. Very briefly because we need to move on. You've been very generous with your time, Matt. There's a couple more questions I wanna ask.

00:47:28 Matthew Botvinick: For sure.

00:47:28 Andrew Keen: Last week, one of the heads of security at OpenAI quit because he claimed its culture is broken. He wrote an influential essay on The Atlantic. Anthropic's security leader also quit recently. There's a lot of fear that these companies—when it comes to good and bad Claude, they may be good at encouraging their good Claude, but maybe policing bad Claude is another issue. In terms of your arguments about AI and political freedom and the defense of democracy, can we trust these private companies, your company, OpenAI, Google, you've also worked for DeepMind, to actually police these, to police these systems to guarantee and protect democracy? Adam Tooze wrote a very good piece earlier last week about AI not being similar to the Manhattan Project in the sense that the Manhattan Project would, of course, was a government project, whereas AI is a private project. So he compared it to our fate being dependent on Dr. Strangelove, which I don't think anyone wants. So in your view, given all the pressures and opportunities, the potential of good Claude, the dangers of bad Claude, can we trust private companies like yours to remake democracy, or do we need to rebuild the state? I understand that for you, AI is kind of the state, but you—you're—

00:49:03 Matthew Botvinick: No. No. No. I wouldn't put it that way at all. And so to answer your question, no. There's no reason—I mean, to the extent—no. Look. And I don't—I'm not speaking for Anthropic, but I can simply point to public statements that the company has made that in fact, greater public oversight is very much needed for this industry.

00:49:22 Andrew Keen: But more than oversight. I mean, going back to Tooze's point. I mean, for better or worse, I think he's put his finger on something really important that—I mean, imagine in the forties if the Manhattan Project had been privatized. God knows what would have happened. We probably wouldn't be talking now. Is one of the challenges in terms of defending democracy rebuilding the state, maintaining its credibility, its trust? And the challenge or the problem, what makes this so complicated is the reality of all this AI technology, which is both part of the problem, but also a kind of solution.

00:50:00 Matthew Botvinick: Yeah. I agree with all of that. I think, yeah, I guess the thrust of my book is very—is very much what you just said. We need to rebuild the foundations of democracy. What—to connect it with the issue that you just raised around AI safety, I'll make two points. First of all, I think, you know, people are right to be thinking about what it means for private organizations to be driving the development of this technology when it's gonna have such profound social, economic, national security implications. That's—those are legitimate concerns, which honestly are shared by people in the industry, and this is why Anthropic has made the statements that it's made publicly about pacing the frontier and so forth. On the other hand, I think we need to worry a lot about what it would mean for government itself to have direct as—direct, access in any sort of global way to this technology for exactly the reasons that I cited before. These are tools that are very ready to the hand of, authoritarian leaning leaders. We see it happening in other countries today. One thing that—like, I guess the bottom line for me is this. I'm delighted that the—that discussion of AI safety has sort of moved now outside of the labs where it's lived, outside of the labs in academia where it's lived for, you know, ten or more years now into the public debate. I personally think that's a fantastic development. We all need to be thinking about how we're gonna manage this technology as a society. However, what worries me a little bit is that we're not considering a very important aspect of AI safety, which is specifically democracy. The bottom line is AI safety requires democracy. If you have—you can build your AI system to be as good, you know, it can be the good Claude as you say. It can assiduously avoid, hacking. It can assiduously avoid, sharing information about how to make, bioweapons and so forth. It can do all of those things that we think about classically as AI safety risks. But if you introduce that, quote, unquote, safe AI system into an authoritarian setting, it's not gonna be safe. It's gonna be used for, automated weapons. It's gonna be used for surveillance. It's gonna be used as an excuse for emergency law, to quash dissent. This is what I'm worried about. I want the conversation about AI safety to expand so that people see that democracy needs to be part of it.

00:52:49 Andrew Keen: We did a show recently about Hannah Arendt, of course, the great political theorist, perhaps the greatest political thinker of the twentieth century, certainly as a critic of totalitarianism of one kind or another. She famously argues that politics is not a business for engineers like you. She wasn't personalizing it. Of course, she didn't know you. But you present democracy as an engineer. You are an engineer. Your background is at MIT [ed.: Botvinick's academic career was at Carnegie Mellon, Penn and Princeton], at Google, now at Anthropic. Do you think that you and types like yourself, people with engineering backgrounds, are able to talk about politics and democracy without coming up with one kind of technocratic solution or another. Democracy, as Arendt said, it's a dirty, messy business, and that's just the nature of it, for better or worse.

00:53:45 Matthew Botvinick: Yeah. It's funny. I don't experience myself as coming at it from an engineering point of view. I mean—

00:53:50 Andrew Keen: You sound like an—I mean, I don't mean this as a—I don't mean this as a criticism.

00:53:54 Matthew Botvinick: No, no.

00:53:54 Andrew Keen: You sound as if you're coming at this whole problem like an engineer. You—you've been given this project. Okay. Democracy doesn't work. What can we do to fix it? And you have all these phases and programs. I mean, that—maybe that's one way of addressing the problem, but you are thinking like an engineer.

00:54:12 Matthew Botvinick: Let me own that then. So, I mean, on the one hand, so I actually went back to school and studied, got a law degree because I was worried that maybe, there were insights to be gained from, you know, kind of looking at the law from the point of view of—in its own terms that I didn't wanna miss. On the other hand, even coming away from that additional education, I do think that getting institutions right is a matter of—there is an engineering aspect to it. So let me give you an example. One of the things that I talk about in the book is a movement called the deliberative democracy movement. And the idea here is that, we need to revive the way that people were meant to dis—discuss political issues with one another.

00:55:03 Andrew Keen: Yeah. We've done many shows on deliberative democracy. I assume you work with the Yale historian [ed.: Landemore is a political theorist], Hélène, maybe. Hélène Landemore, who's the Yale—friend of the show.

00:55:14 Matthew Botvinick: Yeah. One of my heroes. One of my heroes. And so, yeah, so the kind of work that she does, the kinds of institutional designs that she's interested in, I mean, you could call them a form of engineering. Of course, the goal of the engineering in that case is to foster human autonomy. Right? Is to support human moral action. So I—you know, there's a—the risk that when you say you're designing institutions as an end—in an engineering, as if you're ignoring the humanity of the whole thing. I hope—I know Hélène doesn't do that, and I hope that—

00:55:47 Andrew Keen: Well, Hélène's a political theorist. I mean, so that explains your arguments, in the book, the vision of, you know, when you're dreaming of reimagining democracy for what you call phase two, which is really the long term—ancient Athens at scale. So, essentially, in scale, of course, is very much of an engineering Silicon Valley term. Great. You're scaling the citizen assemblies, and Landemore has done a lot of work. Citizen assemblies are back in vogue, pre-AI in Ireland, Belgium, Canada. I know that, Hélène's championing one in Connecticut. So could you imagine—I mean, just dreaming. We gotta [unclear] in a second, Matt. But dreaming in the long term, is that the real promise of AI, ancient Athens at scale?

00:56:38 Matthew Botvinick: It's a picture I try to paint in the book as a possibility. So the kinds of citizens assemblies that Hélène champions and researches, I'm convinced that they can be quite effective at putting people in a position where they are ready to have open minded, fact based conversations about issues of public concern. But there—at present, we only know how to do them on a very small scale at the level of, like, you know, maybe a hundred people. So how do we scale that up so that it has democratic legitimacy so that everybody—the idea of democracy is that everybody who wants to participate in politics can participate in politics.

00:57:14 Andrew Keen: But isn't that the very definition of the citizen assemblies of antiquity, or is that these republics were small, and that the idea of, quote, unquote, scaling them undermines their very natures.

00:57:29 Matthew Botvinick: Well, that was one of the concerns, around the founding of the US was that, you know, yeah, maybe democracy could work in, you know, in these, city states of old, but there was no way it was gonna work across 13 colonies where it took days for information to travel. And the genius of the original designs, you can think about them as engineering designs. They did involve checks and balances, by the way, was that they essentially worked. They did allow democracy to operate across this large scale polity. Now it's all an experiment. Right? Democracy is an experiment, and it will continue to be an experiment. But this is one thing that we can try. One of the things that I talk about in the book, and maybe this goes against the kind of techno solutionist, approach that you referred to earlier is that I think we need to start experimenting. We need to work—you know, we can't imagine that we can just throw AI at our political situation and imagine it will revolutionize our situation overnight. We need to try things. We need to test things out, see how people react to them, see what people like, see what they don't like. And by the way, let democratic institutions guide that. Let's let regulation into the picture. Let's let democratic, you know, elections play a role in this.

00:58:38 Andrew Keen: Let a hundred flowers bloom. Finally, [unclear] about what you call a technological republic. If you can keep it, of course, it's a reference to the Benjamin Franklin, one of America's great innovators and founders who, I think, after—was it the Constitutional Convention when he was asked about what kind of American, what system America had. He said a democracy if you can keep it. [ed.: Franklin's reported reply was "A republic, if you can keep it."] Is that the existential question here, Matt, very briefly to end? It—can America maintain its republic without this new technology?

00:59:27 Matthew Botvinick: It's, for me, it's the key question. AI is here. You know, as Marshall McLuhan said, technology—you know, once the cat is out of the bag, with technology, it's very hard to put it back. So we need to figure out what we're gonna do with it. And, you know, to come back to my point about AI safety, if we can't sustain democracy as a political context around AI, then we're kinda done, because once an authoritarian regime, gets its foot in the door, AI allows it to entrench its power, indefinitely. So we need to think hard about how we're gonna sustain our democratic institutions through this moment. Maybe AI can help, but in the—in any event, we need to be thinking about how these two seismic events that—as you portrayed them—

01:00:16 Andrew Keen: That's where we began [unclear].

01:00:18 Matthew Botvinick: How they relate to one another.

01:00:19 Andrew Keen: So we began on an existential dilemma, and we will end on one too, America—American democracy. If it's to survive, at least according to Matthew Botvinick, who works for Anthropic, so he knows his stuff inside out, needs to use AI, the good AI, the good Claude versus the bad Claude. His book is out this week, and it's important book, certainly because it's written by an insider. AI and Political Freedom: The Risk to Democracy and How to Respond. Matthew, congratulations on the book. I know it took you a while. You didn't rely on Claude or OpenAI for it, so congratulations on that. Let's get you back on the show. This story is only just beginning. Thank you so much.

01:01:04 Matthew Botvinick: Thank you. Thank you very much.