How is AI shaping democracy?

0:03Welcome to the Practical AI Podcast where we break down the real world applications of artificial intelligence and how it's shaping the way we live, work, and create. Our goal is to help make AI technology practical, productive, and accessible to everyone. Whether you're a developer, business leader, or just curious about the tech behind the buzz, you're in the right place. Be sure to connect with us on LinkedIn, X, or Blue Sky to stay up to date with episode drops, behind-the-scenes content, and AI insights. You can learn more at practicalai.fm. Now, on to the show.

0:48Welcome to another episode of the Practical AI Podcast. This is Daniel Whitenack. I am CEO at Prediction Guard and joined as always by my co-host, Chris Benson, who is a principal AI research engineer at Lockheed Martin. How are you doing, Chris? Hey, doing great today, Daniel. How's it going? It's going good. You know, lots of interesting things to talk about as we head into the new year and especially, you know, a lot of people thinking about how technology and AI especially is impacting both our daily life and geopolitics and all sorts of things. And we're really privileged today to have with us Bruce Schneier to talk about some of these things.

1:31Bruce is a fellow at the Berkman Klein Center for Internet and Society at Harvard University. And also, we'll be discussing a little bit his new book, Rewiring Democracy, How AI Will Transform Our Politics, Government, and Citizenship. Welcome, Bruce. Great to have you on the show. Thanks for having me. Yeah, yeah. Of course, as I mentioned, there's a lot of these topics that are on people's mind going into this year, just looking at things that are happening in the news, how AI is factoring into those, how there's various disruptive things happening across our world, how AI might factor into those.

2:13Maybe you could just open us up by kind of setting the stage for why this combination of things was important for you to think deeply about, write about this kind of intersection of AI, democracy, citizenship. You know, I've been writing about AI and technology for a while, and it felt important to talk about AI not just in a corporate context or in a financial context, but in a democracy context. I mean, AI is going to affect kind of every aspect of society because society is about people, and AI is in some ways an artificial people, you know, of varying quality and capability. and we can think about them in terms of companies and consumers and workers, but we also think of it in terms of citizens.

3:04And I wanted to look at AI and democracy, how the tool interact, how AI will affect democracy. My co-author is Nathan Sanders. He and I have been writing about AI and democracy. And someone very smart once told me that you should think about writing a book when you start having book-length ideas. And when our essays sort of turned into something more in our head, we thought about writing a book because there's a lot going on here. I mean, everyone thinks about deepfakes and they stop. But to me, that is the least interesting thing about AI and democracy. They're so much more interesting. And we tried to cover all of that in the book.

3:50And I think we did a good job. It was a lot of fun to write. It feels like we're at a very particular point in that there are several kind of key elements coming together in all of our lives right now. We obviously have, you know, AI disrupting lives, changing the way we live and work and what security means, obviously. But we also have these geopolitical events that are happening in the world in terms of what is democracy, both here in the United States and abroad. I'm curious, as you're looking at all these big, big world events that's impacting everybody on the planet now, how do you see all these coming together in a big picture?

4:31And it feels like your book is very relevant right now. And so can you talk a little bit about the book in the context of all these things that are happening in our lives and in the news right now? Yeah, so it's interesting. There's a lot that the book doesn't say about what's happening in the news. the book is really about how AI can make democracy better. Now, AI, in my mind, is a power-enhancing technology. If you like democracy, AI will help you make democracy better. If you hate democracy, AI will help you make democracy worse. It doesn't have an intentional stance. It takes the stance of the people who wield it.

5:13Now, I guess I want to start by laying the breath of what I'm talking about. So again, more than deep fakes, the book is divided into five basic parts. In the first part, we look at AI and elections. And that's everything from authorized AI avatars that are used in Japan and in Brazil and other countries to interact with voters, AI being used for different aspects of campaigning, setting up websites, doing messaging, AI in polling, AI helping with get out the vote, door to door knocking, so all of AI in politics. The second part is AI and legislation, how AI is helping write, amend, debate, and pass laws.

6:04And this includes a French AI model that lets legislators write better law to a Chilean model that looks at legal interactions, sort of all the different parts of AI and legislating. Part three, we talk about government administration, ways that we can use AI to make government more responsive. Now, right, Elon Musk can use AI to make government less responsive if he wants, but there are ways to use AI to make government more responsive, to figure out benefits, to audit contracts or citizens or different compliance documents, to help the patent office look for prior art. I mean, all sorts of things.

6:47Part four is AI in the court system. Different ways AI can make the courts more efficient. And this ranges from Brazil using AI to help schedule judges and cases to make their courses run more efficiently, to judges using AI to maybe query the plain meaning of a term. And our fifth part is citizens. How AI can help citizens organize, make their voices heard, advocate, figure out who to vote for, who to support, sort of be better citizens. So that's the breadth. We have examples from all over the world. This is not a US book. This is very much a here are cool things happening in so many different countries.

7:35And we're largely looking at the good things. We don't talk about how AI can be used to save democracy or destroy democracy. This is really, if you have a democracy and you like it, here are ways AI will change what's going on. Some of them positive, some of them negative, but it really is AI and a functional democracy. We wrote it mostly in 2024 and then into 2025. So it's not really about what's going on in the United States right now. It's more about democracy in general. Yeah. I really like how you're framing. Number one, it looks at this across these different areas or sectors or geographies, what have you.

8:25But also, you're talking about, I love what you said, AI more than sort of deep fakes. There's kind of two elements that I'm thinking about that triggers in my mind when you're talking about this. One is AI itself is quite a broad term and covers many things. And secondly, the ways that it can be applied to your point might be positive, might be negative, or there might be security related issues related to that. If we tackle the first piece of that, some people might think, oh, I'm hearing like AI as applied within the courts or within legislation and thinking about their own personal experiences interacting with what is most visible as AI, these chat systems, general purpose chat systems and understanding that they make mistakes and all of this stuff and sort of be worried about, well, is AI good enough?

9:18Is it a destabilizing force to those systems? Is it a stabilizing force? And also, is that the type of AI that we're talking about even? Do you have any thoughts for those people out there that are kind of coming at it from that perspective of generative AI only and maybe this perspective of it not, wondering if it's not good enough for these kinds of contexts? It depends what kind of contexts it are, right? So people's most common interaction with AI are probably their mapping app on their phone, Google Maps or Apple Maps, and the algorithmic feed of their social media. Those are probably the two places that people interact with AI the most.

10:01Notice most people don't realize that as AI. Notice that neither of those are chatting, right? Neither of those are producing text. They are both predictive AI. which video will keep you on the platform longer, is turning left or right likely to get to your destination faster. When people look at chatbots, I think largely their opinions are formed at a moment in time, probably two years ago. And maybe they use it once and they made a mistake and said this thing is stupid. And these technologies very broadly are changing all the time. So when you say, is it good enough? The question is, compared to what?

10:41So, this is a non-democracy application. I was at a conference in Toronto about a month ago. It was about medical uses of AI. And this is an experiment with AI in an emergency room. So in an emergency room in Canada, it might be true in the US, I don't know, they have to, after a case comes in, a bleeding person, leg severed, whatever it is, they do whatever they do, they have to write up what they did and pass that on with the patient to wherever they're going next, right? Whatever part of the hospital they're going next. Turns out doctors are terrible at this. They leave stuff out, they make mistakes, they forget things, they do an awful job.

11:18So the hospital experimented with having an AI that passively listened in the emergency room to all of the noise and the chaos and everyone screaming and talking and saying things and wrote up this after event report. It's probably got a name. And then the doctors would look at it. They would correct it. They would approve it. It would go on. It was orders of magnitude better. The doctors loved it. And they would say, oh, my God, I forgot that. We did do that. It made fewer mistakes. So there's an example where the AI is way better than a human. There are going to be other examples where the human is much better.

12:04You probably don't want your AI as a doctor because a human doctor is better. But again, compared to what, you know, fast forward to some parts of this planet where there are no doctors and your choice isn't an AI doctor or a human doctor. Your choice is an AI doctor or nobody. And there you might say, well, give me the AI. It's going to make mistakes. But, you know, my brother is even worse. He'll make more mistakes. He won't know what to do at all. So it's very context dependent. And it depends on what you're doing. The AI that's feeding you your TikTok feed, I mean, who cares if it makes a mistake?

12:42If it's an AI that's putting policemen on the street, yeah, mistakes matter more here. So there's never one answer to that. It always depends. It seems to me that those are great examples. And they also, tying back to the book, you make the point that democracy is kind of understood as an information system and AI is operating on that system. And as you just pointed out, these capabilities where you're talking about agents and chat capabilities, being able to capture what's happened much better than the humans will and not miss the things. As you're looking at things like free speech and stuff, how would you apply kind of that?

13:26You take that emergency room monitoring and apply it with AI agents into the context of free speech and how does it change it? How does it amplify or tamp down on it? What does it mean for us? I don't know if it affects it that much. Basically, it's already so bad. We're already living in a country where money equals speech. The more money you have, the more speech you have, the more power you have. These social media algorithms, even pre-AI, very much affected how you were heard and what you heard. I don't know if AI makes much of a difference, right? AstroTurfing, the notion that a company is going to fake a grassroots movement, way older than AI.

14:12I mean, they can use AI to do that, but that is not an AI problem. And a lot of times in our book, we come across this. We come across the problems that AI exacerbates, but doesn't cause. And in all those cases, like we know the solutions. They're not hard. Technically, they're just hard politically. And, again, depending on the country, you will do different things. So I'm going to pull an example from Germany. Germany has a lot of political parties. That's actually kind of confusing to know what they stand for. And for decades, they have had a system where the government, some government agency, would summarize political parties and voters can go to this nonpartisan voter guide and figure out what the parties stand for.

15:01Last year, they experimented with a chatbot to do the same thing. So instead of going to a static web page, you would have an interactive conversation with an AI that would tell you kind of what the party stood for and likely who you support. Younger voters like that. Now, there's something where it's not going to make mistakes. It's working on a very confined data set. You know, these systems, we're learning that if you constrain them, they don't go widely off script because the script is very narrow. So we are now seeing that, you know, you don't want a massive AI model. You want an AI that's a good travel agent, right, or a good investment counselor or, you know, a good something else.

15:51So this is all changing. changing the hallucinations, the making mistakes. We talk about mistakes and what they mean. And again, it always depends compared to what. You have to say that again and again, compared to what. AIs aren't great drivers. Humans are terrible drivers because we drink alcohol. We get tired and AIs don't.

16:23Well, friends, perplexity builds AI that can answer almost any question. Miro powers visual collaboration for millions, and Mixpanel processes billions of analytics events every single day. And none of them use engineers to update their marketing websites. Think about that for a second. The company's pushing the boundaries of what software can do. The side of the smartest move for their dot-com was to get engineering out of the loop entirely. Well, that's Framer. It's a website builder that works just like your team's favorite design tool, real-time collaboration, a robust CMS built for SEO, integrated A-B testing, the works.

17:00Changes go live in seconds with one click, no ticket, no sprint planning, no, yeah, we'll get to it. And this isn't some startup compromise. We're talking about enterprise-grade security, premium hosting, 99.99 uptime SLAs. That's four nines. The infrastructure is serious. The workflow is just fast. Learn how you can get more of your dot com from a framer specialist or get started building for free today at framer.com slash practical AI for 30 % off a framer pro annual plan. That's framer.com slash practical AI for 30 % off framer.com slash practical AI rules and restrictions may apply.

17:40Well, Bruce, one of the things that I think about immediately when I hear the words, you know, whether it's democracy or citizenship or and especially sort of AI is questions. You even, I think, talked about force multiplier or power multiplier. And AI seems to be kind of having this tendency to, in some ways, centralize power in the sense that if you have the data centers, if you have access to larger cards, or if you have a certain scale, you can maybe do things. Do you think that's an inevitability of the technology? Do you think, and the other element I'm thinking of here is actually the drivers of research on the AI side are really the corporations, right?

18:26The larger tech companies that are driving cutting edge research versus traditionally kind of the academic side. Not that the academic side is not doing good research, but you see a lot of this good and cutting edge research coming out of industry, maybe where there's more access to compute or other things like that. So any thoughts on this dynamic of this centralization of power in relation to AI, how that affects the question of democracy and citizenship one way or the other? So it's 100 % not a function of the tech. It's a function of the way we choose to use the tech. It's a function of the market.

19:07This tech does not have to be concentrated in the hands of five monopolies in the United States. We just chose it that way. And the monopolies are powerful. And of course, in the United States, money equals policy. If you're rich, you get the policy you want. So that is being perpetuated. But it's changing naturally. I mean, DeepSeek taught us that you don't need the cutting edge chips. and all the money to make a competitive core model. In the book, we advocate something called public AI, which are AI models not built by corporations, not built under the profit motive. And our book, it's largely theoretical.

19:47We talk about how this could be done. The neat thing is a few months ago, it happened. Switzerland, ETH Zurich, the Supercure Center, funding from the government and other organization, I forget, produced a core model, a purchase. It was not corporate funded at all, not built on the profit motive, no illegally stolen training material, no poorly paid third world labor for fine tuning. It is free to use. You can go online and use it right now. It is competitive to the best models of last year. It's a little bit behind, but you're not going to notice it. And here's an example of doing it much cheaper.

20:33And my guess is we're going to have 20 of those in a year or so. I think that the dominance of the big corporations, these hundreds of millions of dollars for core models, we're going to laugh at that in a few years. It's turned out to be much cheaper. You don't need to spend all this money, all this compute, that you can be smarter. And especially we're going to need models that are more specific. We're going to need a model that is sort of a good physics teacher. We're going to need a model, right, that is a good, you know, restaurant chooser for me, right? You know, something that will be, you know, my agent.

21:13And, you know, my butler model is going to call any of these dozen or two dozen specific models, any of what we want. And in this world, the Claude's and the GPT's and all these massive models become archaic. So I don't know if this is true. That's my guess. But there's nothing in the tech. These are corporate decisions. Just like the models don't have to be overconfident. They don't have to be obsequious. They can say, I don't know. The tech allows them to say, I don't know. The tech doesn't require them to produce child porn on demand. Those are choices by corporations. And we can make other choices.

22:01It's hard that we're living in a world where corporations run the planet, but it is technically possible. I'm curious. I think that's a really interesting idea. And I think it's one of those things. I think we have a bias where we automatically think of the good option as being open source. But open source and open weight models are still, you know, trained on corporate dollars and trained on corporate information access and stuff like, you know, as you just pointed out. And so like, as you have planted the seed in my own mind of kind of going, what if, can you talk a little bit? What's playing in the back of my mind is like, how would such an ecosystem come to fruition, especially in the political climate that we have right now?

22:48There are many places in the world, but as we sit here in the United States and the fact that there is forces of influence being applied in all sorts of different ways, how would you create such an ecosystem to kind of equal what the Swiss have done in the United States or a similar nation that doesn't get those fingers of power being applied to them? You know, it may not be corporate dollars, but there are so many other ways. How do you envision that coming about? I think it's a wonderful dream, but I am, I will confess I'm making a struggle on seeing how do we get from here to there. Yeah, the U.S.

23:27doesn't do this. Of course not. I mean, we can't even fund government weather service anymore. I mean, the U.S. is done. Just assume that U.S. government will just decay and not do any of this. I probably couldn't before just because we believe corporations should do all the things. Other countries don't have that bias. But you in the United States could use the Swiss model. Right? So, I mean, the ecosystems are going to come from the ground up. I know people who are using DeepSeek on their phones. It's not fast, but it's on their phones. So I think the environmental concerns will largely disappear as these models get smaller and easier to run.

24:09We saw this with search. Search used to be environmentally expensive. Now it's very cheap. I think the US will be dominated by the tech giants, by the monopolies. That's not going to go away. But other countries are trying to divest themselves. I mean, Europe's trying to build an entire tech stack. Singapore is working on a public model because Southeast Asian languages get short shift in these American models. They have a model called Sea Lion, Southeast Asian languages, something, something, that is optimized for their region. And that's what they're doing. France built a model. I mentioned it before that is optimized on French law.

24:55Taiwan's having a problem because the models in Chinese largely are trained on Chinese data, which is often translated from Russian. and they use the wrong words for democracy in these models. So a lot of politics here. You go on Hugging Face right now, there are two and a half million public domain models. Now, a lot of them are people who are personalizing Lama and a lot of them are small and toy models, but there's good stuff there and that's going to continue. you. So to me, the hope is this happens from the ground up, that as this becomes cheaper, more organizations can do this. Switzerland is not a big country.

25:45I am spending the year in the University of Toronto. I'm normally at Harvard. Canada could do this. They pledged a couple of billion for AI infrastructure. Now, they could build their own model. OpenAI is offering to house an instance of their model here. Say no to that. You can build a Canadian model. You don't have to rely on, right, Lama or, you know, any of the U.S. Claude or any of the U.S. models. So I don't know. But my hope is this happens naturally as things get cheaper. The normal way I interact with these models is perplexity. And that just gives me a choice of models. Seven or eight, I can choose from.

26:30I can imagine there being 100 I can choose from. I can imagine there being a perplexity scheduler that looks at my query and decides, you know, which model is best for me. I mean, there are some I've paid for. We can make up how this works. Apple, you know, is trying to figure out how to put models on people's devices. I mean, they are the privacy preserving of the big tech monopolies. I mean, sort of unique of all of them, they don't make money spying on you. They make money selling you overpriced electronics. So they want something for that overpriced electronics to do. And putting the AI model on the object will be that.

27:11So I think a lot of things are happening. We just had a few years where only a few could do it because it was very expensive. And I think that is naturally going to change. And I'm wondering how much from your perspectives, as you've seen the industry evolve, we talked about one level of evolution of the industry, which is the model, right? which is evolving in the ways that you've mentioned. And different types of models are coming out and that sort of thing. But there's this element outside of the model, as we've seen these kind of agentic systems develop, which are like all the things around the periphery of the model, which could be data sources, right?

27:58Which brings up a kind of access to data element. It could be other systems, right? I want to analyze my tax return, and so I need some access to some maybe IRS system or something like that. And then there's interactions with maybe non-AI computer code or rule-based systems that are tied into these agentic things. How do you see that element developing? Because there's a kind of this infrastructure or access side of things, which if you have access to a good model, it might not mean that you have access to the right systems that kind of are their own kind of force multiplier to the model. So I think there's a lot here.

28:46I mean, you know that there's a lot of software between what you type in the chat window and what the model receives and a lot more software between what the model produces and what you see. And all of that middleware is where alignment, rules, I mean, a lot of things get applied where the system figures out what the person wants. You know, AI doesn't get math wrong because the math is grabbed outside the AI and just done. So the AI doesn't mess that up. So I think there's going to be a lot there. Access really feels important here. I think we're going to have haves and have nots. People have access to good models and people who don't.