Loading summary
Transcript43 lines
- [00:01]
A
Welcome to the New Books Network.
- [00:05]
B
I'm Caleb Zakrin and you're listening to the Truth About Bullshit on the New Books Network. Taking inspiration from Harry Frankfurt's seminal work on Bullshit, we take on big ideas with brilliant scholars and practitioners who seek to dispel myths, clarify confusions, and cut through conflicting perspectives with clarity of thought. Like many people, I've been following the developments of AI, testing out new models and following the deluge of news stories about the fight for supremacy. Much has been written about the existential and economic risks posed by AI but the political implications of superintelligent systems has often been sidelined in the United States and elsewhere. AI companies steam ahead with little regulation or oversight. Meanwhile, politicians appear flat footed and unsure about the best way to integrate AI into the government. To make democracies stronger and more responsive to the needs and will of the people. AI will undeniably change how governments work. But how can we ensure that democracy and individual rights are safeguarded amidst the most transformative technological revolution in more than a century? Today I'm speaking with Andrew Sirota, Head of research for the office of Eric Schmidt. Andrew has written extensively about the relationship between democracy and artificial intelligence. His writings have appeared in outlets like the New York Times and Noema Magazine. Andrew will dispel many myths about AI where he looks to call bullshit on the idea that democracy is a system heading fast into the dustbin of history. Andrew, thanks for joining me today on the Truth About Bullshit.
- [01:31]
A
Thanks for having me, Caleb.
- [01:33]
B
It's really a pleasure to have you on the Truth About Bullshit. I feel like we've been having various conversations over the years about politics and then obviously AI it was around when we first started talking, but it's really just become this incredible behemoth in the past few years that seems to be transforming every single field or at least posing challenges to every single field. And I really do think that when it comes to AI, people are always talking about it in terms of the technological and economic aspects of it. Is it going to end humanity? Is it going to make us all unemployed people living in fully automated luxury? Communism? Who really knows? But one of the areas where I feel like it hasn't really been well discussed is the area of democracy and how it's going to impact democracy either for bad or for good. And I was just wondering if you could just talk a little bit about yourself, how you first got into studying political theory and democracy before we even jump into AI.
- [02:29]
A
Definitely. Yeah. I should first say thanks so much for having me on. And I Agree. This is a very important intersection that more people should be thinking about and exploring. When I got to university, I was drawn to political theory because I was really interested in the question of how institutions then shape what we believe, how we come to believe, how we come to make decisions collectively. That's really a question of democracy. And in these questions of democracy, there were several dimensions that I was interested in. So on an epistemic side, I take this from my mentor, L.M. landmore. How do we design institutions to make good, good and better decisions? But I was also interested in this historically. What were the philosophical debates that led us to our current instantiations of democratic institutions? Then there's the normative element of power and who gets to decide and through what structures. But really it all kind of like cycles back, or rather circles back to these epistemological anxieties around our beliefs.
- [03:33]
B
And I think that that is a really important framing that you take to AI, because I think that a lot of the people that are working on AI, working on AI research, have more of a tech background. And as a result, they're not thinking about political systems, they're not thinking about the structures of democracies and why democracy is an important system to defend. If humans, with all of our flaws, we make bad decisions sometimes, why would we want to empower human beings to be able to make decisions when we can allow a super intelligent AI system to make the decisions for us? Wouldn't we want to seed that? Right. I oftentimes feel like this is an underlying belief that might not necessarily be fully articulated by certain people that are working on these systems, but you definitely can sense it when you hear them talk. And, you know, over the past several years you've done extensive research on AI, how it's reshaping democracy, because lots of governments around the world are looking at how they can implement AI to improve their decision making, to improve their activities, their various ways that they interact with their citizens. And a lot of the analyses as we've talked about, look at economic impacts, existential risks. Why do you think examining AI's effects on democracy is so important?
- [04:43]
A
I want to be clear that I don't want to say that a focus on existential risk or economic impact is for naught. Right? I think all of these things are interrelated and should be studied and prepared for. Right? In some ways, democracy is a luxury, right? You can't really exercise agency if we're not alive, for instance. Same goes for economic impacts. I definitely understand the focus on economic impacts. I Think, in many ways, these questions of job displacement are how many people are going to first experience AI transforming their lives. And in fact, I think our institutions are not well equipped for this transformation. And there should be more focus on those kinds of social and economic questions. But the reason why I wish people were also thinking about democracy is because when we talk about things like AI's relationship to the economy, we talk about these things with a certain degree of inevitability, Right. That this relationship is deterministic. And really, the outcomes that we will see are outcomes from a series of decisions that we make along the way. Right. And what is democracy other than a system that we impose to make collective decisions? Right. And so these questions of AI and its relationship to democracy, or AI and its relationship to collective reasoning, in my view, are at base of many of the rest.
- [06:15]
B
Of course, when we talk about democracy, we're talking about particular types of democracies. We don't have a pure democracy. And many people question whether or not we would want to have a pure democracy. It might be a little bit too much work for everyone. So we have at least the United States. We have representative democracies. And in this system, voters choose the politicians. The politicians then make the decisions about everyday issues. And I was wondering if you'd talk a little bit how the democratic systems that we already have sort of mirror the way that we are using AI.
- [06:47]
A
I think the best book that kind of traces the genealogy of our modern representative democracies is a book from 1995 called the Principles of Representative Government by French political theorist Bernard Manand. And what he shows is that through the evolution of modern representative democracy, we may call it democracy, but what we really have is a mix of different democratic and aristocratic elements. Right? We have these moments of democracy, these moments of mass voting, but many of the actual decisions are made through political elites, through this kind of filtering that happens. Right. I think James Madison put it best in the Federalist Papers that elected representatives are supposed to refine and enlarge the public views. Right. This is like a historic debate between trustee and delegate models of representation and things like.
- [07:36]
B
Like that.
- [07:37]
A
Of course, this was definitely true for Madison, and it was true to an extent for Manon. This filtering that happens through representative democracy is supposed to be a good thing, Right? And we don't want a pure democracy, they say. But what I argue is that this leads to a sort of deferential reflex among citizens, right? Citizens come to internalize the assumption that the most important decisions that shape their lives are decisions that other People or other entities are supposed to make on their behalf. And it's really that foundation that I believe then becomes difficult or complex, at least when you throw AI into the mix and people start developing a new relationship to AI. Right. I want to be clear that I'm not advocating that we should ban AI from all decision making processes or even that we shouldn't think about how AI can fundamentally transform the way that democratic institutions work. Work. I actually think we could and should be doing that. And in fact, I think what we're seeing in real time is that the goalposts are already changing around what we think is an appropriate question or task to delegate to AI models. Right. And these are social norms that are going to change and continue to change. But I do think it's a basic question of what is the democratic reflex that we are cultivating and how do we pull that in from existing representative institutions. And if that isn't sufficient for cultivating democratic habits in the age of AI, then what are the other models of democracy that we need to be thinking about?
- [09:15]
B
I think it's relatively easy for human beings to detect the flaws in other humans reasoning. You know, there's like some innate understanding that we have about the flaws that people have, even if we don't necessarily have the theoretical language or the psychological research to back why someone's making what decision that they're making and why they're making it. You know, we're human beings, we kind of understand each other. This isn't necessarily the case with large language models. You know, the, the people that are developing these systems constantly refrain, refrain that they don't exactly know why the AI model says what it says when you ask it a particular question. They, you know, they, they understand at some basic level how it's operating, but they don't necessarily know why. And you have some examples of instances where algorithmic decision making, algorithmic tools, which are, you know, AI is a type of algorithmic tool, more advanced type, have been used, have, have failed. And you know, obviously this is part of the idea is that governments might be taking AI tools and turning over human decision making to AI decision making. So you have this example of the Netherlands where they use an algorithmic, an algorithmic tool to monitor welfare fraud. So can you talk about this case and why you really focus on it as an example of why we should be cautious about turning everything over to AI tools?
- [10:32]
A
Between 2005 and 2019, I believe the government in the Netherlands used an algorithmic system to flag childcare benefit fraud. So this was essentially instances in which people were falsely claiming childcare benefits and in a sense, abusing the system. Right. We hear that language all the time, abusing the system. What ended up happening, of course, is that some people were falsely accused. And we're not just talking about a few people, we're talking about, I think it was 26,000 people over the course of this period. And the people who were falsely accused were also not selected randomly. I believe there were disproportionate number of non Dutch nationals who were flagged. So these algorithmic systems were endowed with certain racial biases and so on. And so maybe that's the first lesson, which is a fairly obvious one, which is that this system wasn't ready for deployment. Right. Technically speaking, we have built and can build better, less biased AI systems, but governments, for the sake of efficiency, are just so eager to jump to deploying algorithmic systems that they may do so before they're even ready. But the more interesting lesson that I think this example shows is that in the Dutch case, so when a citizen was flagged for potential welfare fraud, it actually required that a civil servant still had to sign off on the flagging and do their own manual review of it. But this requirement was essentially useless because the civil servant was given no reasons as to why these people were flagged. And so they were liable only in name. Right. They had no real tools at their disposal to be able to keep the AI system accountable or in check. Right. It's what we might consider to be like an impotent clerk. And that is really the relational thing that I'm interested in, which is like, what are the systems of accountability and verification that we're building around AI systems when they're inevitably going to be used in government to some extent. Right. Whether this be humans or other AI systems that are checking to ensure that decisions are reached fairly and in ways that align with the stated objectives of these different systems.
- [12:59]
B
Now, I know that people that work in AI development and AI researchers, they always say that these systems are very complicated. They don't know exactly why they're saying what they're saying. Why is it that these tools are so opaque and it's so hard to actually understand what they're doing.
- [13:16]
A
Yeah. So there's really a technical difficulty here. Right. Which is just that the complex networks that lead an AI to have a certain output are so incredibly complex, people often call them a black box. And one way to get around this, there's an entire field called interpretability, with researchers that are trying to understand the inner workings or the brain of large language models. The verdict's still out, I think, on how successfully they can do that. There are some heuristics that are created to try to do this as well. There is something called chain of thought monitorability, which is essentially. It essentially allows researchers to see the large language models stated reasoning steps along the way. I think there's an open question around whether the stated reasoning steps are a perfect reflection of the inner workings of the model. But that's still something, right? And honestly, that could be said for humans as well. We provide reasons for things all the time. And whether or not they really map onto inner workings, I think is a question really. But beyond the technical difficulties, I really think it's a question of institutional design. And that's really the lesson from the case in the Netherlands, which is what is the relationship here between human and AI? Even if we can understand the AI perfectly, which is a big if, but it's an if that I hope comes to be true. There's a question around what is the path for recourse? How do you reclaim power in that instance, when you don't like the outcome of a decision?
- [14:58]
B
So I want to go back a few hundred years to the Age of Enlightenment, and I think that it's important to start here, because the Age of Enlightenment was really where the groundwork was laid for the birth of modern democracies. Obviously, the democracies that were established at the very beginning of the Age of Enlightenment were restricted oftentimes in many cases to landowning white men. And in time, the privileges of democracy, allowing the consent of the governed, people actually being able to engage in a social contract that's expanded more and more. But it really does feel like we're in this moment where this social contract that was established in the Age of Enlightenment is being questioned by people. And I think it's especially being questioned by the fact that people look at these new AI systems and look at the CEOs and the tech leaders of these companies as the potential future monarchs in a way that we can almost imagine a world where there's some super intelligent AI that runs the US government, or there's different AI systems that run each government, and we don't really know what they're doing, but they just tell us that the AI system instructs us, tells us what the laws are, tells us what to do, and we don't really know what it is, but we know that it's doing it for our best interests or because it has the power Much in the way that a monarchy did prior to the revolutions that overturned that system and established democracy. So do you think that democracy and artificial intelligence are oppositional forces? That artificial intelligence will essentially bring back an era of monarchy, or that we can actually fuse the two, fuse artificial intelligence democracy in a way that actually will. Will benefit people's ability to still engage, still express their will, still feel that they have a say?
- [16:49]
A
Right. I definitely fall on the latter end of that question. I think there's often an assumption that AI, because technologically it is a system that centralizes information in order to make decision, that it needs to be a tool for authoritarians or surveillance. And I just think that is not necessarily the case. But I want to kind of go back to the first part of your question around the social contract first, because I do think it's true that we are in a moment where there is this broad social reckoning around the question of whether the social contract that we currently operate under is sufficient for the age of AI. And you talked about consent of the governed. And I do think we may in some sense be seeing the shortcomings of a social contract that is based around consent alone. And by this I mean consent really matters when you're operating under a system that is imposed on you against your will. Right. And that was true of the pre Enlightenment days, and that's why it was such a radical transformation ideologically. But whether or not you consent to something says very little about the regime form itself. Right? So even today, authoritarians use the language of democracy, that they are operating on behalf of the people. Right. They in a sense have their consent. And we can imagine a world in which people consent to AI rule. Right. And just because they consent to it, that doesn't make it a democratic system. Right. And so what I would argue is that we actually need to figure out a different way to conceptualize what we mean when we call a system democratic. In my view, it's not only consenting to power, but actually being able to exercise that power. Right. The Athenians knew this well, they understood democracy as a system that was really about forming citizens capable of self rule. Right. And if you use that definition, then you aren't really starting from the premise of, okay, you have people making decisions and you have AI making decisions, and which one is better, Right. In that sense, you are treating AI as a potential substitute for democracy. But if the way you start with thinking about democracy is that, no, it's about forming citizens capable of self rule, then really the question is, okay, how does AI advance that. And how does AI open up new opportunities to do this at scale? And that's what I'm really excited about. And we already see instances in which that's happening.
- [19:15]
B
And I think that people have experienced positive ways that AI is able to empower individuals. Obviously, people can, without coding skills, can now essentially code their own website. People can, I can learn.
- [19:30]
A
Yeah.
- [19:31]
B
It's quite incredible. And for, you know, for anyone who hasn't tried using these tools, like, I really do recommend people try them instead of being reflexively frightened by them. I think that there's many reasons to be frightened by AI in a broad scope, like you said, like, we should still care about the existential risks aspect related to AI and still think about and plan for a world where jobs start to disappear because of AI, as we're already seeing. But of course, like, like many technological advancements, we also have to be aware that, that you can't really turn the tide back. Exactly. We can just help to fashion the kind of the corridor, the path that it goes down. And I think that that part of what, what you're saying or what you're suggesting is that, you know, we don't necessarily need to accept the narrative that AI is going to be so, so powerful that it's going to invalidate the point or purpose of protecting human value. That human, that humans. We still matter, at least in our own context. Even if AI is dealing with a universe of ideas and thoughts that we can't possibly comprehend, it doesn't mean that our lives don't matter too. And when you find yourself, I'm sure you find yourself talking to people that are skeptics and just reflexively feel like, oh, well, AI is bad. People say we need to shut down the AI data centers, we need to have a moratorium on new developments. When you encounter these sorts of arguments, what do you say to them as to why they shouldn't necessarily be afraid, but they should be cautiously optimistic or cautiously embracing of the potential for AI?
- [21:11]
A
So I think there is a positive vision for AI that people are not talking enough about. That kind of extends beyond the realm of politics or at least the domain of democracy. Although, of course, depends how you define democracy. But in this sense we should be thinking more about AI for scientific discovery, for improving health outcomes, for expanding access to education, for uncovering new solutions to climate change. These are all really important things that we should be talking more about. But I think even within the realm of democracy, there is a very clear argument that could be made in favor of having a more positive relationship with AI, which is that there. Which that people are very unhappy with the existing system. Right. And we see this in waves of populist discontent where people are demanding a new form of collective rule. Right. Or of self rule. And that requires something different. Right. It requires a divergence from that paradigm of deference that I was talking about earlier. And I think many people, because of this fundamentally antagonistic relationship that many people have with AI, they don't even know that is an option. Right. And I think that's one way to frame it, or at least that's how I think about it.
- [22:34]
B
And there certainly have been ways that people that you've been in close contact with have been thinking about how we can use AI or use the Internet as a tool for improving democratic outcomes. More sensitivity. Obviously, the way that democracy works in most places now is every few years people vote and in the interim there might be other special elections. But really it's polls that are quite fickle and easy to, you know, depending on the question, it's quite easy to. To manipulate it. And the Internet has been a source of, to a certain extent, democratic voice, but it's also been, I would say a bit of a. It's both simultaneously an anarchic place and it's also a place for demagoguery. And you've looked at, at various approaches like Taiwan's open source v. Taiwan platform, which is essentially an attempt at e democracy. And I was wondering if you talk about this system and what you sort of see the promises for. For it for enabling collective deliberation.
- [23:37]
A
Yeah, absolutely. So this actually ties in well to the previous question when I was talking about broad public discontent. So the origination of Taiwan's VTaiwan platform is a really interesting one. It dates back to 2014 when there was a massive student led protest movement in Taiwan called the Sunflower Movement, which was responding to a potential trade deal with Taiwan and mainland China that critics thought would undermine Taiwanese sovereignty and Taiwanese democracy. And what ended up happening is that the protest movement occupied parliament for several weeks. But what the government decided to do in this moment was not to clamp down with an iron fist, which is the response that we mostly see today, but rather to open up the government. Right. So they invited Audrey Tang, who at the time was a civic hacker in the Sunflower Movement, into the government and who would eventually become the first minister of digital affairs in Taiwan. And together they developed VTaiwan, which is essentially, as you said, it's a digital platform that enables broad public input into policymaking in Taiwan. The core technology architecture is based around this platform called Polis, which is a discussion forum where people can submit their own comments and vote up and down on other people's comments. And then it uses machine learning to essentially map points of consensus, or what Audrey calls the uncommon ground that otherwise may not be known in the public sphere. And so far, VTaiwan has been used on dozens of policy issues in Taiwan. The most famous case is when it was used for the introduction of Uber into Taiwan. But it's also been used on a variety of other things like AI regulation and other things. So I do think when we're talking about the ways that AI can be used to usher in a more participatory, inclusive, deliberative form of democracy, Taiwan is typically like the canonical example for this reason.
- [25:40]
B
Yeah, this example is really interesting. And obviously there are many other, I think, attempts at how can we implement some form of E democracy. I can think of pros and cons for a democracy. I, I, as as much as I find myself feeling fervently supportive of democracy, I also do understand the arguments that people made for why representative democracy is better than a pure democracy. That just logistically speaking, you know, if every single person is engaging in decision making on an everyday basis, we probably won't get much done. And also, you know, there's a question of expertise too. You know, people that are spending, devoting their entire lives and careers to certain issues, they might have a better understanding of how to actually implement something. So can you talk a little bit about what you see as the pros and cons of an E democracy system and how it might be developed to deal with some of these issues?
- [26:31]
A
Yeah, absolutely. So a few things. One, if the assumption is that citizens should be voting on each and every issue, not only would nothing get done, but that is incredibly tiring for the average person. Right. I don't necessarily think that's what we want, but when we are trying to balance different democratic values in the design of institutions, we might think of them as coming down on two axes. One is thinking about mass participation as an element of democracy, which you really do need if you're going to call something democratic. And rule of the people of for by the people. The other element, though, is what we might call deep deliberation. And this is when the other elements come in, which is that you want to hear from experts, you want to weigh evidence. You don't just want to have an aggregation of preferences, but rather you want to be able to spend time it's not just a question of efficiency. Sometimes that friction is built into the system because you need to let these ideas and these perspectives percolate. And so traditionally there's been somewhat of a tension between mass participation and deep deliberation. When we're talking about these in person forms of democratic participation, whether that be citizens assemblies or other kinds of deliberative mini publics. There's also an affective or an emotional element to this right, which is that there are moments in between the actual deliberative sessions where people go to their local bar, they're having dinner together and they socialize. And that too is really important for building what we mean when we say the capacities for self governance. Because it's not just an epistemic thing, deliberation, it's also the ability to figure out how to coexist with people who share different values. What's interesting to me though when we talk about E democracy is that many of them, and arguably polis as they currently exist, don't go as deep into that deliberation category. These are text based, asynchronous forms of participation that have maybe some elements of deliberation, but it's certainly not as deep as a citizens assembly or something like that. But I remain curious and hopeful to see how, how this technology evolves and whether it will actually be possible to replicate some of those deep deliberative qualities. Whether that be hearing from experts, developing emotional connections with other citizens, whether that can be replicated at scale using AI mediated platforms.
- [29:17]
B
Yeah, I think that it's really interesting to think about all the different ways that people can develop these sorts of systems and I would love to just see it.
- [29:27]
A
I would say also on that point that it's not as if there is a single approach that should be used for every country, for every locality, for every topic. There should always be sort of like a non dogmatism to these things, that different kinds of processes might work better in different moments. But I think fundamentally we just can't be so scared of the technology that we don't recognize the moments when it can be really, really powerful and really helpful.
- [29:55]
B
Right. And obviously in lots of places, like you know, I live in New York City and for people to give public comment on certain issues, it's, it's, it's either confusing or you have to, you have to go to a place and wait for hours and it oftentimes doesn't make a difference at all. And I think many people have had the experience of being told to call their congressman to no real solution. To no real, you know, impact at all. So I do think that there is an outlook about technology. Obviously you said that part of democracy or part of what's part of having a democracy is this idea that the citizens have to rise to a certain level. You know, maybe they have to have certain, certain virtues that make them fit to participate in a democracy. Like certainly the ability to engage with people in a civil way is an important part of democracy. Otherwise it's, you know, you have war instead. But, you know, we hear a lot about this idea, especially when people talk about younger generations who are in high school and college now that they're not really doing their, their homework, you know, they're not doing their readings, they're not writing their essays. They're just going into, into ChatGPT or Claude and saying, write me an essay about to kill a mockingbird. 800 words, please. Thank you. And what's concerning about this is this idea that there's a generation that isn't doing this sort of thinking that might be required to then be able to think critically about political issues once they become adults. And I was wondering if you would talk about this sort of pessimistic outlook about people that, you know, want to remove technology completely from the classroom because they see it as this threat to our ability to deliberate, to think that AI is going to basically turn us into Nietzsche's last Man, who just, you know, who is a pleasure seeking animal who can't think for themselves.
- [31:51]
A
Yeah, it's a problem. It's a problem I'm concerned about. Right. There's been some writing around the notion of gradual disempowerment, which is like a very subtle transition where over time we delegate more and more of our cognitive processes to AI systems and then deprive ourselves of the ability to do those things on our own. In fact, I just saw a paper yesterday that was comparing reading comprehension skills between a group that had access to AI and a group that didn't, a control group. And what they found was that group that had access to AI was able to perform better when they had access to it. And then when you took away that access to the AI tools, they performed worse than the control group. Right. So. So this is something that, that's playing out in real time. I will say though, that that the degree of risk here is a little unclear to me. I sometimes think about as a very reductive analogy, like the calculator or GPS systems, where certainly my ability to do complex long division is worse. Now with the advent of the calculator than maybe it had been if I lived before the calculator. And yet that wasn't like a catastrophic risk to my ability to reason. And so I think the question with AI is like, is this a difference in scale or in kind? And I don't think that is only about the technology itself and the capabilities of the technology. It always goes back to the social contracts that we build around these technologies. And what I think will be really important is the question of similar to the calculator, do I have the foundational skills to be able to do these things with or without the technology? And then when I use the technology, can it make me radically more efficient and more productive at those things and frees up my time for deeper kinds of learning? Right. I don't know that is necessarily the default outcome here. Right. And that's what I think my concern would be around. But I guess just personally I err on the optimistic side. And so I think given that that is a possible outcome, that is the one we should have in mind and we should be working toward.
- [34:13]
B
Yeah, I mean, that is a good point. I think about math skills. Right. Certainly we still, even though we have calculators, we still teach kids in elementary school their times tables because it's more efficient and helps them actually build those base skills. But then of course, I'm sure most people have had that experience of typing in something that they knew, you know, typing in two numbers that they could have easily multiplied together when they were a kid. But they, it comes a little slower now.
- [34:39]
A
Yeah. And, and I mean, to be fair, like, I don't think any of my family members would say that I am the, the best at navigating without my gps. And, and so there, there is a risk that like, that this shifts from like just a convenience thing to like in you, you lack the capabilities thing. You know, I, I don't think I'm actually that bad at navigation quite yet. But, but that's one plausible thing that we should be keeping in mind.
- [35:07]
B
We, we became drivers after, you know, GPS was basically a thing. So, you know, I'm sure nav people who had that, that the map navigating skills probably have a bit of a better sense than, than we do though, obviously. Is that, is that the, does that spell the end of our ability to make deliberative decisions in, in the democratic context, probably not. Right. So a concept that I feel like I used to hear a lot more about that haven't heard as much recently, is, is the idea of technocracy, which is Essentially ruled by elites, you know, technical experts, people that, that know better than the average citizen because of some skills they have or some additional knowledge. And this idea that this is, you know, it's a, it's almost like a form of, of, of aristocracy, some with maybe some merit thrown into the mix. And many people have criticized this in part because of, you know, very high profile instances of the experts getting things wrong, people questioning whether or not the experts actually no better than mass decision making. And one thing that you have written about is you wrote an article with Eric Schmidt in the New York Times about algocracy, which is essentially a modification of this idea of technocracy. But instead of rule by technical expert experts who we've seen make flawed decisions over the year, instead we'll turn over the keys to algorithms. So this to me seems like, if anything, you know, the more likely, more reasonable expected version of how AI could really take over government systems. Not necessarily as AI as dictator, but AI as technocrat. So can you talk a little about this idea of algocracy that you explore?
- [36:45]
A
Definitely, yeah. Well, I think as you said, technocracy and algocracy are in a sense cousins. I think they're very similar in that they operate under the assumption that all political decisions are in a sense, technical decisions. Right. Things that can be clearly and cleanly optimized if only you're given the right line of reasoning. When democrats, multi democrats operate under a very different assumption, which is that what politics and democracy means is to negotiate, to fight out competing values in the public sphere with other people. I would say, though, that algocracy differs from technocracy maybe in two ways. The first is, at least theoretically, rule by algorithm can just be much better than rule by technical experts. Right. I think there is an epistemic argument against technocracy, which is that when you only include a small subset of technical experts in a decision, you're necessarily excluding important perspectives from that decision making pool and then you reach worse outcomes. I don't think that necessarily needs to be the case with algorithmic rule. Right. You could actually simulate those preferences of all of the perspectives and reach better outcomes as a result of that. So that's one argument in favor of algocracy over technocracy. That also makes it more dangerous for this other reason, which is that when we were talking about technocracy, we are talking about humans, human, human beings that can be held accountable for their decisions. Right. And my sense at least is that it's not necessarily the case with an Algorithmic or algocratic system, which is that one likely outcome we could see is that systems that are already facing corruption will actually become more opaque, not less opaque, and that paths for recourse among citizens then actually diminish in the end.
- [38:45]
B
Yeah, I'm sure that we won't hear the last of algocracy. I would be certain that this is going to be something that as the years continue, we'll see more and more instances of algorithmic decision making being used by governments, especially amidst concerns around how budgetary concerns appear to be more efficient to outsource. And of course we'll see in real time their effectiveness at making decisions versus technical experts. One reading of the 20th century that I oftentimes feel like has been a popular reading that I do feel like is coming into question now was we had these competition, this broad bipolar competition between a. And it wasn't exactly this, but a bottom up market based democratic approach. Of course, course there's lots of caveats there. And a top down planning approach, you know, the Soviet five year plan, obviously, you know, China still practices that sort of model five year plan with some market based approaches integrated as well. And many people, you know, in the kind of the, the liberal end of history narrative would say that what the 20th century reveals is that the bottom up Hayekian approach was, was better. It won the day because individual buyers and sellers are going to settle on prices much more efficiently. They're going to settle on outcomes much more efficiently than any system that, that is basically going and telling people oh, here's what the prices of wheat is going to be in three years. It doesn't necessarily work that well, but it does feel at least that in certain instances if AI really does become super intelligent and become what people think, it could be that it could essentially undo this lesson of the 20th century and show that maybe there is a form of a top down approach that, that might be more, more effective that that collective deliberation, even if it does lead to good outcomes for neighbors haggling with each other, that it's actually inefficient. It's the broad, on the broad scale. It can't actually look at things like climate, you know, climate effects or other sorts of issues. So how do you think about this kind of general debate that was waged in the 20th century that's being reopened again in the 21st?
- [41:07]
A
I would mirror political scientist Sherry Berman's argument for the 20th century that in the fight between capitalism and communism, what actually wins out is not free market capitalism but some form of social democracy, something that brings together elements of what Fukuyama and ultimately Hegel praised about liberal democracy, which is not only good outcomes, but also this recognition that you are worthy of a decision making being and good outcomes that can come about through efficient means and so on. And I do think that as the allure of the authoritarian model becomes stronger with integration of AI systems, that makes it not only more efficient, but actually reach, quote, unquote, better decisions. We will need to figure out new ways to balance these two ideals. Right? We can't just, as I was saying earlier, we can't just totally shun the use of AI because we assume that it threatens democracy. In fact, we should use it to make certain elements of our decision making processes more efficient. And at the same time, we need to preserve what is so compelling about democracy in the first place that it recognizes you as a being with integrity and value and agency.
- [42:26]
B
So this is a very live topic and there are so many politicians, so many researchers and people that are thinking about these topics that you have been exploring. You've obviously been looking at it particularly from this democratic political angle. And I feel like that's, as I said at the outset, that's not often how people are thinking about these topics, but it really is partially how politicians should be thinking about, about these issues. If there is a wave of unemployment as a result of AI, politicians need to be prepared for there to be a reaction from people. And how are we going to think about those that the social contract when unemployment goes up or if unemployment goes up as a result of new systems? And how do you think about, you know, when, when you're, you know, working with people that are, that are working on the cutting edge, people that are politicians themselves or people that are advising politicians. How do you approach them when you're trying to give them advice or help them think about the ways that they can both preserve deliberative politics, preserve democracy, while also embracing the benefits of AI.
- [43:38]
A
The first piece of advice I would give is, is to use AI to do that very thing, right? Use AI to enlarge, deliberation, to expand the public sphere. Right? Because right now when people think about AI, they don't even think about that as a possibility. AI is incredibly unpopular. It pulls worse than ice. As you said, the most prominent policy proposal around AI is for a moratorium around data centers. And this is because people only have very scary images of what AI means and what AI is going to mean for their lives. And I do think there's a real opportunity right now to show people a different path, a path where AI brings in a new form of representation. When we think about democracy and a broad historical perspective, this may be just one more evolution of what the system looks like. And the politicians and the policymakers that lead the way here are going to be historic. And that's a very exciting prospect. I don't think it needs to be strictly a scary one.
- [44:41]
B
Yeah, we're certainly in a time of immense change. I feel. It feels like the world order is shifting under our feet. And in no small part, that's because of this new technology. And I think it's always important to remember that we often talk about the backsliding of democracy, as if democracy, the high watermark was of the past. But, you know, it was never really perfect to begin with. And it's still, it's still, still you. It still could be improved and it still, I think, will be improved. So.
- [45:15]
A
Yeah.
- [45:16]
B
Andrew, it was really wonderful to get the chance to talk to, with you about this topic. I feel like it's just, it's such a rich subject and every single day there's, there's more news, more developments that I think are changing. People are thinking about it. So I'm glad that you're really on the front lines studying this stuff and that you're, you're a, you know, one of the lone voices that are saying,
- [45:35]
A
what about the people?
- [45:37]
B
Instead of what about the money?
- [45:40]
A
Yeah. Thanks so much, Caleb. This was a lot of fun.