C (2:23)

Yeah, absolutely. So I'm Nathan Sanders. I'm an affiliate of the Berkman Klein center at Harvard University and. And my background is really in science and data science. I did my PhD in astrophysics. From there I was really excited to apply the statistical and machine learning methods involved in AI in a wide variety of fields. So I've done research in applied work in industry, in public health, in environmental science, in media and entertainment, in higher ed, in the publishing industry, and in biotech and chemical machine learning as well. So I really just enjoy finding ways to apply the technologies of what today we call AI in lots of different domains and to understand what their implications are. But the thing I'm most passionate about is public policy and especially finding ways to help vulnerable communities have a larger voice in the future of the laws and the policies that govern all of us. And so my work at the Berkman Klein center has really focused on that. My colleague, my co author, Bruce Schneier, is a fantastic faculty member at the Harvard Kennedy School and also a fellow and affiliate at the Berkman Klein Center. His focus for many years has been cybersecurity. He's really known as one of the leading voices and technologists in not only innovating some of the technical systems that power modern cybersecurity, but also calling out their societal impacts. And he's really made a focus of his own in the last few years in thinking about how technology is changing democracy. He's been the convener of an annual conference focused on reimagining democracy. What would governance look like if you invented it in the 21st century instead of, let's say, the 18th century thinking of the American Revolution? And that's been an amazing convening of academics and practitioners from all different fields and from government to put those ideas into action.

B (4:10)

How did the two of you come together to write this book? What was the moment where you decided that you should write it? And also I'm wondering, did you use AI at all while writing it?

C (4:19)

Yeah, great question. We met through our respective appointments at the Berkman Klein center at Harvard. We were put together there and what really inspired the seeds of what became this book is Bruce's last book which is called A Hacker's Mind. He went through systematically all the ways that the idea of hacking, which we traditionally associate with computer science and software development is actually more of an ethos that lives out in all different aspects of life. He gives great examples of how athletes hack the rules of sports. And he talked about how politicians and political actors hack the rules of government and electoral systems. He gave a great example that really inspired my thinking, which was thinking of gerrymandering as a hack of the electoral system invented centuries ago. And at the very end of his book, a hacker's mind, he thought forward to how will the practice of hacking change with the advent of AI technologies? How can AI become a hacker in the same sense that humans have been in the past, and how will that change the dynamics of hacking? So we got together to think about, okay, now that Bruce has kind of laid out that idea of AI hacking, what would it mean to do that in practice? What actual machine learning and AI tools that exist today would be used for that? What specific modeling technologies? What are the specific capabilities of those tools, and what would they lead to? So we initially started writing together about that. And then as the technologies of AI have evolved and had so much of a broader impact on democracy, we tracked that really closely. We wrote about it, and it broadened our thinking into the much larger subject that became this book.

B (5:54)

Right. And as far as the topics that you go and explore, the book is broken down into a variety of parts. And then from there, you know, you have these, these very short chapters that, that very directly address a variety of questions that, that people might have. So I think, I think it might be helpful for you to just sort of situate people now to, you know, in a very broad, broad overview, you know, how is AI currently being thought of implemented by governments like the US Government? Is it, is it something that, that they're really pushing or is it, is it confined to certain areas and not to others?

C (6:35)

Great question. We think of the application of AI in governance as very much a present issue, not a hypothetical future state, but something that nearly every government around the world is actively investing in and doing. While we were writing our book is primarily during the final year or so, the Biden administration and during that time, the Biden administration released an inventory of more than 2,000 specific applications of AI throughout nearly every agency of the federal government. So you can imagine for us, that was sort of a goldmine and learning about how governments are thinking about using AI today. And we cite a lot of those examples in our book. And of course, the US is not the only such government. Just to lay out a few examples that were really interesting to us. To our knowledge, France was the first Government whose legislature directly developed an AI tool and distributed to the legislative staff to augment the legislative process. They developed a model called Llama Amendment, based on Meta's llama model, which helps legislative staffers to summarize and understand the impacts of legislative amendments. And we think that's a great example of an assistive technology that seems to us an appropriate, useful application of the technology of the eye to make that one aspect of government function better. And there are thousands such examples from around the world that we discuss just a few of in our book.

B (7:56)

Right. I'm curious if we could sort of think through the process together of how AI is used. It is sort of a very, in a very basic way. So, you know, the first way that someone interacts with democracy is just by being a citizen of a polity. You know, they go and, you know, they have this right that's enshrined where they get to vote, they get to make certain decisions. Maybe they, you know, as you show in the show in the book, you know, in the past, in the past, in ancient, ancient Greece, you know, you might have gotten a lot pulled and then you, you know, were, you know, when we're expected to serve. So at a very basic level, how is AI changing how voters interact with their government?

C (8:38)

You're right. We opened the book with this anecdote of a society that developed a technology that would effectively supervise how their democracy functioned. A society that was sort of skeptical of giving individual citizens the elected or appointed power to control everything. And so they relied on technology to actually decide who would be in charge of specific functions for specific periods of time. And by the way, as their technology advanced, they made that system more and more complex to try and solve specific problems of fairness and security. And as you said, that society was not a science fictional story. It was ancient Greece from more than 2,000 years ago. And the technology was called a clariterion, a mechanical device that was used to decide who would hold an appointed office for a period of time, essentially by random chance. And we think of that technology as part of a long continuum that today has led us to the point of AI being involved in democracy. We don't see the introduction of AI to democracy as a completely novel encounter that we've never experienced before. It's part of a long lineage of technologies that also includes things like the telegram, the radio, the television and social media that have forced democracy to constantly reinvent itself. So we talk about what that means today. We talk about really interesting examples of citizens around the world leveraging AI in new ways and where that leads to. I'll give you just two examples that we cite in the book along that path. One, forgive me. I'll pick an example that I'm involved in that we cite in the book. We've got a project here in Massachusetts called maple, the Massachusetts Platform for Legislative Engagement. We're building a online, open source, nonprofit civic technology platform that just makes it easier for Massachusetts residents to learn about the bills proposed in the legislature and to convey their point of view, to submit testimony for the legislature to consider. That's a process that has existed in Massachusetts for hundreds of years, actually dating back before the American Republic. But it is a process that's largely then manual and often difficult for people to interact with. So we've just made it easier through a website. And we found AI to be a really useful tool in facilitating that kind of citizen engagement. One of the features that we've rolled out in the last few months is an AI summary tool that allows every Massachusetts resident to get a plain language explanation of what every bill does in the legislature. We introduce 8,000 bills per session in Massachusetts. No one ever has, probably no one ever will provide the service of having a human sit down and explain what every one of them means. But we found it very effective and worked with some great technologists at Boston University and elsewhere to provide a service that automates that using AI. And it works well. We extrapolate in the book to what that could mean in the future. Democracy involves a process of distributing decision making throughout a polity. One of the most important ways that many democracies do that is through ballot initiatives where a specific proposal is put to citizens who have the opportunity to make an up and down judgment. Should we do this option A or should we do option B? Should we do something? Should we do nothing? That's a very narrow exchange of information between citizens and government because of the way that ballots work, because it's difficult for all of us to take time off of work and out of our days to go down and vote. Most democracies do ballot initiatives very infrequently, and there's very little opportunity to iterate on that proposal. If an initiative fails, you might have to wait two years or four years for the next election to bring forward an alternative version of that proposal. That could all look very different. With AI, we talk about experiments happening around the world where people are trying to augment individual citizens with AI tools that help represent their perspective, that could help translate citizens express preferences into expected points of view on not Just one ballot initiative, but maybe hundreds. And to offer that point of view not just once, but maybe on an annual basis or a monthly basis or even a daily basis. Of course, there are huge problems with that. There are ethical concerns, there are fundamental concerns about the nature of democracy, there are technical concerns about how well an AI can represent a person's opinion. We try and work through those questions in the book, we take them very seriously. But ultimately the book is about exploring what are those possible future states of democracy you mentioned?

B (13:06)

In France they have this AI system that's trained using the llama large language model, which is metas. There are others. There's ChatGPT from OpenAI, there's Claude Anders Grok, for example. Sometimes people will ascribe a sort of a politics to these different chatbots. Or that depending on the user, depending on what your interactions have been, what preferences for certain political responses, it might give you one answer or another. So if I ask it, who should I vote for? It might give me a better answer than you if you ask it. So I'm wondering how you think about this as well, about the people's reliance on AI going to AI to understand where they should vote as opposed to, let's say, other sources, like listening to the radio, watching TV, going on TikTok, watching a creator. How do you think about this in terms of what chatbot someone might use, how this might affect their voting preferences?

C (14:10)

It's a very significant question. And when we were writing the book it was a little bit hypothetical whether or not an AI would have a systematic bias towards one point of view or another. There's been a lot of really interesting research over the years analyzing the specific political proclivities of specific models such as GTP 3.54 and all the way up to the present day, and comparing them across model developers to see if there's systematic biases from for example, OpenAI to meta, et cetera. And of course what's happened more recently is that some of the largest AI model developers have been really explicit about their intent to build in one type of bias or another. Some of those are sort of high minded statements saying that we are going to try to build a quote, unquote objective model. Of course we don't really think it is viable to build a truly objective AI model. It really depends on the values that the developer holds and the values that they try and encode in the model. But we also have developers like Brock, as you mentioned, that have taken an explicitly quote, unquote anti woke, which I think we should all read to mean ideological point of view on how models should be developed now, from the point of view of different types of users, for some, that would be a bug and a huge problem, especially if it conflicts with the values that they are bringing to a decision that they want a model to help them with, and for others it is a feature. It is desirable. We give examples in our book of how different actors in a political system independently make choices and independently have the authority to choose to use AI in their work. One of the examples that we talk about are judges. There are many examples of judges leveraging AI to help them interpret law and make decisions in the US and around the world today. And judges have always hired assistants, traditionally human assistants, with a view towards their ideology. And we give examples from the history of the U.S. supreme Court, where 70 years ago there was a big debate over whether or not that was appropriate. In any case, we all know that it happens that clerks are selected on the basis of their ideology and by the way they influence the decisions that judges make. You can see the fingerprints of clerks and the opinions of the Supreme Court or justices. It matters who their assistant is, and likewise, it matters what AI assistants. We all choose when we're leveraging them and making decisions, and we should expect to see their fingerprints on the choices that we make.

B (16:34)

Right. And as far as, you know, people like judges or politicians or other people that, you know, are involved in democracy and in the government in some capacity, I remember, I think it was during one of the Republican debates. I think Chris Christie, you know, attacked Vivek Ramaswamy and said that he sounds like chatgpt. And I, I, I am noticing this more and more that sometimes when I hear politicians speak, they'll say things or speak in a cadence or it just doesn't sound natural. And, and I know that, you know, obviously politicians are constantly reciting or reading speeches that were written by another human. But you do get the sense that, you know, if you're at one of, you know, one of the speechwriters, one of the aides. Yeah, it's probably useful to just plug in, you know, here. What would blank politicians say about this, this issue? And it might, it might come up with something, something pretty good. So what did you discover? What did you learn about, you know, politicians using AI, for example, for things like speechwriting? I mean, obviously you already gave the example of, you know, for example, writing opinions. But what about things like that? Are you noticing more and more speeches that seem to be written by one of the chatbots?

C (17:40)

Well, I think you're right to point out that it can be a tell when politicians speak in a certain way, and sometimes you can tell that they're speaking inauthentically and maybe trying to refer to something someone else or someone else wrote. And of course, that's always been true politicians, certainly in the lifetime of our country, in the last few hundred years for the United States, politicians have always been, over that timescale, sort of a front person for a much larger sociotechnical system. In the decades past, we would expect politicians to not speak their own words, but to rely on a speechwriter, help craft a message for them. And you might say that ideally that message would be based on their own point of view and their own values and their own policy commitments. But of course, that's never been true either, because we know that certainly in recent decades, politicians have leveraged quantitative data from the electorate, polling data, focus groups to help them decide what to do. And by the way, even before that, we couldn't trust that a politician is representing individually their own human point of view on an issue because they've also been part of political parties and they've been obligated and incentivized to represent the perspective and the policy platform of that broader group of humans and the systems built around those parties. So we don't view the idea that a politician may leverage an AI tool as part of decision making or part of their expression as categorically different than those other ways that politicians have always been a front person for a larger system, but it certainly changes how they do it. There are, of course, some interesting examples that we talk about in the book of politicians going a step beyond that, beyond just using AI as an expressive tool. There have been candidates who have gotten quite a bit of attention in the US and the UK by claiming, by committing that they will defer any decisions that they make in office to particular AI models, by running on the idea that they would be simply a front human for an AI model. Now, those politicians have not been successful. We don't think that is a good idea. But people are legitimately proposing that that is a platform to run for elective office. And I don't think they'll be the last ones.

B (19:59)

Absolutely. And that's really interesting to think about. This episode is brought to you by White Claw Surge. Great podcast pick, friend. No surprises there. After all, you're all about finding the tastiest flavors out there, just like White Claw Surge, and with big, bold flavors to enjoy like blood orange, BlackBerry, cranberry and more. It's time to go all in on taste. Unleash the flavor. Unleash White Claw Surge. Please drink responsibly. Hard seltzer with flavors. 8% alcohol by volume. White Claw Seltzer Works, Chicago, Illinois.

A (20:32)

Upgrade your laundry routine with a durable and reliable Maytag laundry pair at Lowes.

B (20:37)

Like the new Maytag washer and dryer with performance enhanced stain fighting power designed.

A (20:42)

To cut through serious dirt and grime. And what's great is this laundry pair is in stock and ready for delivery.

B (20:49)

When you need it the most. Don't miss out.

A (20:51)

Shop Maytag in store or online today at Lowe's.

B (21:00)

Also too, you know, these models, you know, there's obviously there's the few big ones that people know about, but also, you know, they can, they can, you can produce, you know, particular chatbots or particular AI tools for very specific purposes that are trained on very specific pieces of information, pieces of data. And, you know, one of the things that, that you talk about too, and has been, I feel like a huge talking point with the DOGE initiative to simplify government, make it more effective, is this idea that there are tons of tasks, tons of things that we can offload onto these AI tools that they can start making decisions, that they'll be more efficient than a bunch of human beings trying to argue out and debate things. So how have you seen, as far as actual implementation for efficiency is concerned and also just in terms of how these chatbots are actually produced that are being used for government, how different are they than the models that most people are using?

C (21:58)

Well, I'm glad that you raised the question about doge, the Department of Government Efficiency under the second Trump administration. I think, unfortunately, they've illustrated some of the ways that AI can really be misused in government in at least two ways that we write about in the book. One way is sort of structurally, we write in the books that part of what makes democracies function historically is layers of the civil servant service. The idea that you may have an executive like the US President who's sitting on top of a large bureaucracy, but that bureaucracy delegates decision making across many layers of civil servants, all the way up from a department head and all the way down to someone who's maybe delivering benefits down at the street level to individual citizens, and they all have different freedom functions and roles in decision making. With AI, it is possible to realize the vision that some on the ideological spectrum have had for a long time to unify the power of that broader executive branch in a single person. The idea of having A unitary executive is really enabled by automating decision making potentially at all levels of government. With AI, we see that as part of an authoritarian project that is a way to use the technology, that is not the only way to use it, but. But is capable of advancing those type of authoritarian goals. Of course, the second thing that DOGE has sort of grabbed headlines for is using AI in ways that are not effective, in ways that lead to mistakes, in ways that do not improve the responsiveness or the quality of service from government. And I hope that people won't take that as a necessary outcome either. Just as technology does not have to be used to centralize power or to advance an authoritarian agenda, it also doesn't have to be used in a way that is not protecting of or mitigating mistakes in a way that is not applied carefully. I think some of the places where DOGE have. Has tried to use AI and perhaps not performed well could be done better with a more careful implementation.

B (24:02)

Right. And what's interesting, too, is that because there are hundreds of governments, you know, thousands and thousands of municipal governments and local governments that are grappling with these questions about AI. We see lots of different approaches, different examples. Some, you know, as you say, as you pointed out, some politicians who, you know, completely want to turn over their job to AI, and then also, you know, tons of politicians that are very skeptical of AI that want to, you know, impose significant regulations. And, you know, one of the splits that we've seen is the split between how the US is thinking about AI and how the EU is thinking about AI. Feature in the book, this trip that JD Vance took, where he basically goes and admonishes Europe for handicapping their AI development. And he frames it, and I want to get to this afterwards, but he frames it very much in this competition between the west versus China as far as whoever gets to AGI or AI supremacy first will dominate the globe. So, you know, could you talk a little about the differences between the EU and the US just in their approaches? We'll just start there, and then we'll get to the US and China and the world and, you know, the new. The new global arms race that we're seeing?

C (25:16)

Yeah, let me try and just very briefly spell out the differences between the US and the eu, and then we can talk about what it means and why it matters. You know, I think clearly the EU has been the world leader at advancing comprehensive AI regulation in the form of the EU AI act, which I think was debated with an enormous amount of care and thought over A period of several years, and then finally passed by all three branches of the European Union government just within the past year, which is kind of an exciting milestone in the interaction between government and AI. Now, I wouldn't argue that the outcome of that legislative process was perfect by any means. We have just a few points in the book where we critique how it could be done differently, but it's a major milestone. In contrast, of course, the US Congress not only has not made substantive progress in regulating AI, not only has not made substantive progress in regulating other technologies that have been around for decades before that, such as social media and having a modernized data privacy law, but especially in recent months, it's actively resisted the idea of doing so to the point where we've had a lot of attention in the AI policy community in recent months, has been over this question of whether the federal government should preempt states and actually strike down the few regulations that some states have already passed on the technology. So what does this mean? I think it points to a fundamental difference of perspective between the U.S. and the EU, where in the U.S. policymakers, particularly in the Republican Party, are framing it as a choice between regulation, innovation in the eu. They embrace the perspective that I happen to strongly agree with, that these two things are not opposed to each other. And in fact, a strong regulatory regime can spur innovation. And we ultimately write in our book, in our closing section about a different and additional way that government can spur innovation in AI. The interaction between government and AI does not have to be only regulation. It does not have to be only constraints on how private actors develop and use the technology. Government can also demonstrate appropriate, ethical, productive uses of AI. Government can do that by adopting government responsibly within its own systems and show citizens directly how it can make government more responsive, make the outcomes better for people. Government can also directly fund and directly spur innovation of AI technology in the same way that government in decades past, and especially the United States, has spurred innovation in physics, in the Internet, and so many other technical fields. We could have a focus for the US Federal government in driving AI innovation by creating what we call. And there's a broader movement that thinks of as public AI. AI that is developed for the public interest, that is subject to public oversight and transparency, that provides a common baseline for the market of universal access at a reasonable cost to AI technology for what we believe is the fundamental critical infrastructure of the 21st century, that is guided by a different incentive structure than the corporate AI developed by private companies. Today, instead of being driven by private profit, AI could be developed and incentivized to chase public benefit.

B (28:34)

Right. And what's so interesting too is the extent to which, you know, AI and the competition for AI advancement, not just AI advancement, but also, you know, the underlying technology that's required to create or to run AI systems. So, for example, you know, they're the most valuable company, I believe at the moment in time it might be not the case anymore. I'm just. The episode is, is Nvidia, you know, which, which produces the sort of the underlying infrastructure to, to actually run these, you know, these, these computers, if I'm not mistaken. But they, they also, you know, a lot of, a lot of the, the, the GPUs are, are created and produced in Taiwan. There's, there's this kind of like a, a geopolitical angle to it too in terms of this competition with China, China's claims on Taiwan, even, even, you know, you know, regardless of that too, there's also just this general seemingly cold war competition between the US and China in terms of, you know, developing supremacy. So I was wondering if you'd talk a little about that, about just thinking about the US government versus the Chinese government's approach to AI and you know, what you think about this, this argument. Because I feel like a lot of this has been the argument that I feel like has been the most powerful in the halls of government of America about why it is that they need to embrace such a pro AI policy. So, yeah, I'll leave it there.

C (29:54)

Yeah. I think this gets to some fundamental questions about where advancements and benefits from AR are created. And people often talk about a triad of the elements that you need to drive advancement in AI. Different people define the triad differently. I'm going to define it as data, hardware and applications. And to your point, I think a lot of the international, the geopolitical discussion about AI has really been focused on a second piece of hardware. Who has the most GPU chips, who controls the Nvidias and the TSMCs of the world. But the reality is that we have widely available computing technology, actually not only GPUs, but also CPUs as well that are very capable of training and powering inference from AI models. That's already a highly, highly distributed technology. It is to me not at all obvious that the future of geopolitical power is determined by who has, you know, the incremental, the next most powerful chip. I think that the application layer, that third piece of the triad, is what's really important. Who's going to innovate on the ways that AI is used in society. Who's going to show how democracy can be improved the most by leveraging AI? Who's going to create the most economic benefit for their citizens by using AI? Who's going to manage and mitigate the harms and risks that are also created whenever there's a fundamental shift in the way society and economies are structured, as may happen with AI, I think the governments that do that really effectively, that manage the application layer, have a big advantage over the governments that are focused on gathering the most chips or getting the next most powerful chip.

B (31:32)

So when you're looking at where we currently stand in terms of problems that you really think that AI, thoughtful application of AI could but improve governance? You know, I, I, I think a lot often of, of, you know, small or like local governments that seem constantly strapped for cash and you know, not enough people, not enough talent able to, you know, help successfully man manage the city. You know, they're on, you know, they're using computers from 40 years ago and you know, there's, there's always, seems to be cases where they get hacked and there's some, you know, to pay, pay $36 million worth of Bitcoin ransom because nobody changed the password on a computer for, for 40 years. So, you know, are there any immediate applications, immediate things that you're seeing or that, that are exciting you about how AI is being applied to improved governance?

C (32:21)

Yeah. Let me give you an easy, maybe obvious example and I hope I can open it up a discussion about what can stem from that. So I think sometimes people almost skip over or maybe forget how far we've come in language translation and making our society truly multilingual. Because of advancements in AI over decades, but especially in the last few years, I mean, it is possible to have a real time conversation vocally with audio, with our voice, with another person who speaks a different language across hundreds of languages today, with the devices that we all have in our pockets. That is just a remarkable transformation of what it means to live in a multilingual society like we have here in Massachusetts and across the United States and really across the world. It is just a fundamentally different way that we can interact with each other that opens up opportunities for democracy that would have been so hard in the past. So of course AI language translation is not perfect. There are definitely cases where having a human perspective and the expertise of a human translator can get you to a better outcome than AI translation, without a doubt. But the fact that anyone who speaks any language should be able to interact with government services and with their fellow citizens. Because of this capability being so widely available and scalable as AI makes it is a really important advancement. And to me, part of what this illustrates is we don't necessarily require governments, especially local governments, to do the innovation to figure out how AI language translation works. We don't even necessarily have to require them to implement it. We can expect now people to walk into, I don't know, their local secretary of state or our V office and use their phone to translate a document that they may not otherwise be able to understand. We can rely on private corporations to do that innovation and put those devices in people's pocket. We can in some cases look for governments to do it. We can also support a vibrant third sector, a civil society sector, to bring those kinds of technology innovations to the democratic applications that matter. For decades, there's been a vibrant movement of civic technologists who are trying to find those pain points in the democratic process and to leverage technology to make them better. Often those civic technologies born from volunteer or nonprofit projects get merged into government once they prove effective. And I think that's a really effective model for this kind of innovation. I'd love to see governments support that more. By the way, that doesn't necessarily have to require enormous new funding mechanisms that take away from other things. My colleague Matt Victor and I have an op ed in the Boston Globe recently pointing out Maryland as a model for how to. So sorry, I'm just thinking about how to word this. We point out how Maryland has recently introduced a new tax that applies just a very small amount of the profit generated by big tech and social media companies to fund things like public media and civic tech innovation so that the companies that are profiting from our attention and maybe contributing to the coarsening of our political discourse and making the infrastructure of our democracy harmed can redirect just a small fraction of that funding to fund the kind of innovations that can apply AI to make democracy better. We think that's a really potentially effective model and we'd like to see more governments investing in that kind of innovation.

B (35:42)

Absolutely. Yeah, that, that makes a lot of sense. And you know, it's really so hard just in this conversation alone to go through every single aspect about how AI is being used in governance. Like, you know, there, there are obviously like, lots of examples too of AI softwares being used, for example, in warfare, in policing, in all sorts of other, you know, surveillance technologies. And I don't know if we'll be able to, you know, if we'll really have time to go through everything like that. But I was wondering if you could just comment on that a little bit about how you see AI used in those sorts of parts of the government.

C (36:26)

Well, you know, we talk about one example in our book that I think deservedly gets a lot of attention when people try and think about where could AI make things better. And I'm going to pull another example from the US Context. It's what I know best. So it's what I tend to turn to first. Our Social Security Administration is responsible for administering disability benefits to millions of people. If you become disabled and you urgently need help to pay for your care or to offset your income, you have a Social Security disability benefit that should be waiting for you to do that. However, our current process is so slow that it creates real harm for people. We give the exact numbers in the book, but I think something like tens of thousands of people in the US die every year waiting to be approved for that disability benefit because the process is so slow. I think that's an example of where the kind of automated capabilities of decision making that are enabled by AI could create real benefit just by helping to speed things up. Of course, there are huge ethical decisions involved in how you apply that. And depending on the values that you bring to the policy process, you'd make very different decisions. We advocate in the book for a process that tries to minimize harm by focusing on speeding up approvals, by trying to leverage AI to review the paperwork associated with cases that are very, very likely to get approved, and to just have the approval be automated and happen much faster. And there could be a secondary process. There should be a secondary process of review and making sure the decisions are usually coming to the right outcome. But you know what? If you approve a handful of people that maybe wouldn't have gotten the benefit with a full human review, that's probably okay. On the other hand, someone who's more focused on reducing costs for the Social Security Administration, who's more focused, whose values are more rooted in avoiding fraud, might implement that system in a very different way that can actually increase the amount of harm. It's really not about the technology or how well it performs in that process of review. It's more about the metrics that you're trying to optimize for and the values that the humans putting that technology into place are representing.

B (38:27)

Are there any big misconceptions around AI that you find yourself encountering? Like, you know, when you're. When you're having dinner with someone, or you're just, you know, you strike up a Sort of a casual conversation with someone who might not be as knowledgeable on AI and applications to technology. Are there, Are there any common misconceptions that you seem to find or any things that. That you find yourself just on a regular basis telling people that that. That might, you know, have one idea about it that isn't exactly accurate?

C (38:56)

I think the conversations I end up in the most are with people who are very knowledgeable about AI and often technology and often science, but who have maybe a more narrow point of view about how things could evolve. I think people who know the technology well and watch it closely see a lot of the harms and risks that absolutely exist in our AI ecosystem today and how it's being used in government. We've talked about a bunch of them just here. The fact that AI is a technology that magnifies power and therefore can be used by governments that have an authoritarian bent to make their system more autocratic. The fact that AI can be used to automate harm and encode bias. The fact that AI does reflect the values of not only who develops it, but also who uses it. And so when it's being used by people in positions of power who have harmful biases and prejudices, it amplifies those. Those are all very real risks and very real harms. And it makes sense for people to focus on them. And we should have vibrant research and communication programs that point out those risks. That's all good, but we hope that everyone will recognize that there's an alternative path that we could be on. If it is possible for us to use the same technologies to distribute power instead of decentralize it, it's possible for us to use the technology in a way that supports values of equality and fairness instead of encoding biases. And even if that's not the way the technology is being used in systems today, as citizens of democracies, we have some agency in changing that. We can fight to change the development model for AI so that it's not entirely within corporate hands for private profit, but can shift to public control for public benefit. We can change the applications of AI so that it's not advancing the plans of people who are looking to centralize power, but instead decentralizing it. We can use it to create public benefit instead of optimizing for outcomes that are harmful to many. And it's really in our hands, and it's our responsibility as citizens of democracy to fight for that. I hope the role that our book plays is by spelling out what those alternative paths could look like and opening people's minds to what actions we can take to try and lead to those better outcomes.

B (41:07)

I think, I think the book really does that and it provides different ways for people to think about how AI can be used, how it has been used, how it can be used in a productive way, it can be used in a harmful way. And I think that perspective, that approach that you bring is actually really refreshing because I find constantly that it just seems like people have either extremely optimistic, shinily optimistic views that, you know, everything will be all problems in life will be solved and will enter some new utopia or the dystopian vision that it's going to destroy all of humanity. So, you know, obviously it seems more likely that it'll be somewhere in the middle, you know, maybe, hopefully, you know, leaning in the positive direction. And I think that this book does a great job of kind of walking through the different ways that it is changing democracy. And I think that that's a subject thinking about how AI is impacting democracy. That has been under discussed. There's been such a focus on how AI is going to impact the economy and I think that that's great and extremely important. But the governance aspect is ridiculously important too. So I'm grateful that you wrote this book and I think that you and Bruce are great voices who hopefully will help shape how we think about AI moving forward. So, Nathan, thanks so much for being asked on the New Books Network. It was really wonderful to get the chance to speak with you.

C (42:31)

Thank you so much for inviting me to join you. It was a pleasure.