Nobel Prize Winner on Threat of AI Dictators and Big Tech’s Risk to Democracy

Like many dictators, you don't need to be absolute. You know, look around you. Many dictators still have to contend with forces. They don't control everything. But if you have a very powerful AI model that influences a great majority's choices, that is a dictatorial position. Who's making the decisions about AI? Right now, just a very small number of companies and perhaps in China, some government and non-governmental forces. But most people feel disempowered. And I think the great promise of AI is to actually provide better information to people, to make workers more productive, to make citizens more capable of controlling their information intake and participate in politics. But that's not where we're heading. Up until this point, perhaps up until the summit, the main obsession with AI has been on existential risk, on the idea that we might get some extinction-level events from it becoming out of control, from the mere fact that it might become more intelligent than humans. That's not necessarily your area. However, there was an institutional structure designed to focus on safety. When you look at what's happened with the tech companies, it does look as if they've just given up on that, that they're all madly now competing to try and get the fastest AI capacity out there to make as much money as possible. And they've really given up on some of the safety. Is that the sense that you get? That's the sense that I get. But I was never very convinced that all of that safety industry was really going after the important issues. I'm not denying that there could come a time where AI could become so powerful as to overwhelm all of our institutions and everything. But before we have to worry about AI as an autonomous agent ruling over us, we have to worry about AI under the control of a few companies, AI under the control of a few governments ruling over us. So before we get to AGI, we're going to get to a point where AI models are so powerful and there are people controlling them. So a lot of the issues about democracy, jobs, inequality come before the worries about existential risk. So the AI industry would argue that it's the ultimate form of democracy to be able to get the maths and the algorithms to work out what the right answer is and to make that available to everybody. And it's automatic. Is it a democratising agent? What's your sense on that? If we delegate all power to algorithms and for that matter, algorithms controlled by one or two companies, I would say that's the opposite of democracy. Now, there is a scenario and I don't reject that scenario. There's a scenario in which AI models could be an equalising force because they can provide capabilities to people who have fallen behind economically or socially. But I also don't think that's the direction we're going. There is a future where AI as an information technology helps people gain greater expertise, perform more sophisticated tasks, perform new tasks. Everybody can be just like the best in their occupation or chosen profession. But that's not where we're going. We're going to a place where a lot of jobs are going to be automated at some point, not next year, not the year after next, but at some point. We're going to a place where a lot of power and information are centralised in the hands of a few companies or a few governments. And that, again, is not democratic. And so you're talking about the companies we know about, the big tech companies that are currently controlling AI. Yes, the companies that shall not be named. Yes. Okay. And when you see, if you like, the scenes in the US where you have these tech companies so close to the new president, what does that make you think? Oh, I worry, I worry. But, you know, just to be clear, tech companies have been very close to several governments before the current administration. And I think Biden, even though he did have an executive order on AI, did absolutely insufficient amount to think about AI risks, AI direction and all of these things. But of course, right now, Trump and his administration have declared an accelerationist approach to AI and to cryptocurrency that I think has the risk of exacerbating a lot of our problems and accelerating us towards wherever the few tech companies wanted to go in the first place anyway. Do you buy into the idea that he or she who controls AGI could be a dictator? Yeah, I mean, I think who controls powerful AI even before AGI could be a dictator. And that's part of what I was trying to get to with centralisation of information. And like many dictators, you don't need to be absolute. You know, look around you. Many dictators still have to contend with forces. They don't control everything. But if you have a very powerful AI model that influences a great majority's choices, that is a dictatorial position. So you don't really need to get to existential level risks to be really pretty worried about where this is going. Absolutely. That's why I'm more worried than the existential risk people. And that's before even we get the meat of what an economist would look at, which is the impact on the jobs market. Again, I don't think this is going to be overnight, but part of my worries are about loss of tasks, loss of jobs and loss of meaning for people. And by meaning, I mean the fact that most of us derive social meaning from the fact that we contribute to something. We contribute to our jobs. We contribute to our families. We contribute to the democratic process. We contribute to our social networks. And again, in an AI driven world, I don't know how these things are going to evolve. AI will change many more jobs in the short run than completely eliminate them. In fact, my prediction would be that if we were having this conversation again in 2030, you won't see any occupation that has disappeared in the intervening six years. But I think if AI goes down this path, there will be people who lose their jobs and there will be people whose incomes are not going to increase. Look, robots, industrial robots are much more specific in what they do and they affect very more, many more, much less widespread tasks, those that involve some manual manipulation of things on the factory floor. But the data show very clearly that when robots are introduced very fast, people lose their jobs and their earnings fall. There is much greater inequality. So at some point, AI is going to start having effects like those of robots that are even more pronounced. Not next year, not the year after next, because its spread is going to be slower than what people expect. But we could get there unless we again find a direction of AI that is at the help, at the service of workers, to try to make workers more productive. And that's what I would really like to see come out of this summit. So in terms of government policy responses here, some have talked about universal basic income. In other regards, perhaps, you would want to tax jobs and labour less. Potentially, do you try and tax the AI in some way? I don't think taxing AI is the solution and because I want better AI. So if you have an across-the-board tax on AI, that doesn't really achieve the kinds of objectives that I want. In the States, President Trump has pronounced a golden age for the US. Now, you've written something quite provocative, saying, actually, we may be seeing the start, perhaps, of US decline, or you've theorised about that. You're not convinced by the golden age narrative? No, I'm not, because, look, I think for AI innovation, we need the right institutions. For any innovation, we need the right institutions. Institutions matter, especially for innovative countries. And the United States has, until now, over the last 80 years, been powered by innovation, by entrepreneurship, by new technologies. And if you damage US institutions, and that means US democracy, that means the rule of law, that means the right regulatory environment, you're going to hurt innovation. And if you hurt innovation, US leadership is threatened. That's the situation we are in right now. Professor Samoglu, thanks very much for joining us. Really appreciate that. Great to hear from you. Thank you. Thank you, my pleasure. Thanks for having me.