Transcript
0:10 I would like to introduce our speakers today. We have with us Francis Fukuyama, senior fellow at Stanford's Freeman Spogley Institute for International Studies. He is here today in conversation with Nate Priscili, professor of law at Stamford Law School and co-director of the Stamford Law AI Initiative. Francis and Nate, thank you so much for being here with us today. We're looking forward to your conversation. Please take it away. >> thanks very much Annette. So I want to talk a little bit about the course that we are planning. this course the course title is AI power and the new rules of global strategy.
0:50 it will be taught in person the week of March 22nd of 2027. I'll be teaching in it. Nate will be and many of the other professors that work on AI policy issues here at Stanford. I should say a little bit about who this course is for. This is not your typical AI course where we're just teaching skills or giving you background on the technology or how you're going to deploy it in your company. our unit is we I direct a master's in international policy at the Freeman Spogley Institute and we are interested in governance issues. we're interested in the question of how this technology relates to geopolitics to domestic politics how it's going to shape the international dimensions of the world that we are entering into. And if you are interested in those sorts of issues, you know, really what this means for the world, more broadly, this, I think will be the course, for you.
2:00 So, let's begin. my colleague Nate is one of the foremost experts on the question of AI governance. He's thought a lot about these the issues connected with that. So Nate in recent weeks there's been a lot of talk about needing to regulate artificial intelligence. you know Dario Amadai did a essay in which he said we need to pace the frontier. I have no idea what that means. but it evidently means you know you got to slow down. But it strikes me that a lot of that discussion has been very non-specific. if you did want to slow down the pace of AI development, these new frontier models, how would you go about doing it? would it be the labs themselves that did it? Would it be an international agreement? Would it be some government agency? And if it's a government agency, you know, who has the authority to make decisions like that? So maybe you could speak a little bit in general about AI governance as a problem and then you know what you see as some of the solutions.
3:13 >> Great. Well, thanks for having me. I'll say that I I've decided that my new diet is going to be called pacing the frontier because that's what I I'm trying to do. so AI governance is certainly having its day and AI is having its political moment. we're seeing it in the United States and around the world. what whereas in the 2024 election AI was barely discussed in this election whether it's because of data center buildouts or these questions about labor force impacts or existential risks which you are focusing on I think now people's minds are are being focused on it both here in Silicon Valley and in Washington.
3:52 I'll say that at at a general level when you think about AI governance and regulation you need to admit a few things. The first is AI regulation is inevitable and it is already happening. if you are you know riding in a way whimo in in San Francisco and it runs over my foot I need to decide who I can sue that is AI regulation. when it comes to the use of AI by banks or by pharma or others, there are existing regulations that are going to that that are triggered. And so we need AI regulation in order to sort of chart a safe AI future. But one of the central questions is whether you regulate the technology or you regulate the applications. For the most part, you know, in the last few months, because of these concerns about existential risk and loss of control, we're now having a more serious discussion about how you might regulate the the technology itself, particularly as produced by the major labs. But I do want to just emphasize that AI regulation is all around us because we need to adapt the existing legal order to to deal with AI. Now, when it comes to how you would try to deal with the the the new problems or the the problems that are are coming from the labs that with respect to the technology, you're right to sort of highlight the what I think are the obstacles, right? And and those obstacles are that this is a new and fastm moving technology. I mean obviously AI has been around for a long time but it's it's a lot of these new features are I think present unique regulatory questions. it is international in scope and so of course whenever you talk about say US governance of AI you have to put that in the geopolitical discussion with what would China do and we don't exactly have international institutions that are performing with great skill these days and so it's it's not clear how whether it's the UN or whether it's WTO or or whoever else to to craft an international agreement it really begins with the US and China, but but we here in the US really don't have the answer. There are many bills in Congress. the administration is is quite accelerationist. and so we're left with sort of the the the preaching from the AI companies about how dangerous this may be and asking for regulation. And I think there is m maybe momentum in that direction but there's a lot to to work on with respect to the specifics.
6:38 So if well okay let's let's go through this systematically. Could you rely on the frontier companies themselves to you know quote unquote pace the frontier? Could you leave it up to them to you know try to slow down the rate at which their development of these very powerful models is going. Well, I mean I we've sort of done that already and we've had the hugging face incident. So I think that that incident over the summer and then all of the kind of knock-on reporting of other incidents like that have suggested that at least the existing system is not working and that in you know that that thank God that that incident or the other losses of control were not didn't harm anyone.
7:26 but it does give you some sense that the the companies if left completely to themselves might not act in the public interest. Now it is still the case that existing tort law and liability regimes could punish the companies if for example their agents committed havoc in the financial system or pick your pick your victim. and so it's not as if they're completely unchecked by the law. but I think given that we're we're we're worried about new kinds of risks with these companies, whether you're talking about what we call the SEAB burn risks, chemical, biological, radiological, nuclear, or these loss of control risks or the cyber risks. there there is this sense that the the potential costs are so huge that leaving it to like the existing litigation liability tort system might not be sufficient. And so one other option is that while we could have collusion among the companies to to come up with best to slow the frontier to agree to do so even the spectre of that by the various press releases that we've seen has led to already an antitrust lawsuit against them. And so I think that it is it is going to be difficult for the companies to come up with mutually agreed upon safety standards if there is not some enforcable enforcement body that would keep them to that. Even so one of the challenges is also that you can do all you want with the labs in the United States but if the Chinese just continue a pace then the technology is going to be out there.
9:08 >> Yeah. Well, let's let's get back to China in in a little bit. All right. So, you can't trust the companies to do this on their own. So, that leaves the US government. this is one of the issues that I think about a lot. U we have a problem in this country with state capacity, meaning we don't have enough people in the government that know enough to actually sensibly deal with a new very fastmoving technology like this. I think the average age of an American civil servant is somewhere in the upper 50s. I have nothing, you know, I'm older than that, so I have nothing against people that are in their 50s, but they tend not to be, you know, up on the latest technological advances. And so you start with this big problem of where are you going to get the people into the government that know enough about this technology that they can regulate sensibly. I think that you know my students a lot of them have been hired by the companies themselves and are working at anthropic and open AI and you know other places but almost none of them actually would like to go into the government.
10:22 so what part of the government you know one question is do you actually have to create a new regulator to deal with this or can you give it you know to existing parts of the government. So for example, I think after the mythos release where there was this fear that government computers could be broken into very easily, it was really the security agencies that got really spooked by this and they're the ones that, you know, pushed the Trump administration to you know, not get that model released so quickly. So that was the defense department, I don't know, the FBI, other agencies like that. who who should be doing this regulation.
11:08 >> So, again, we there's no way around the need for talent related to AI in all areas of say the federal government right now because it's going to be penetrating, you know, whether it's the Department of Commerce or whether it's the EPA or whether it's the FCC, right? There are go every area of government is going to have to tech up. I agree with you that this is you know one of the big challenges. usually end with that in a discussion of AI governance saying look you could write the perfect law but if you don't actually have people there to enforce it you're going to you're going to be swimming upstream and that by the way is a global problem you know for all of its legislation in Europe on the say the European AI act which is the most comprehensive regulation out there if they don't have people in the government to actually monitor what is basically a regulation of the entire economic system in Europe they're not going to be able to enforce it. so I agree with you on that. And so where if you were to develop a more comprehensive AI regulatory scheme in the US, where would you locate it? I think you when it comes to the national security issues, you are going to have to have all of the the various folks in the national security firmament being involved in this. There's no question about that on on those question. So the so the Pentagon and FBI and etc. but I think that if we are going to regulate the technology in general then you know starting some kind of new agency makes a lot of sense but it depends what we want it to do because we're not going to have it regulate every application of AI in the economy. but whether it's it's what was the original US AI Safety Institute which has now been recast as Casey and I think it even has another name this week because we're now calling it super intelligence instead of artificial intelligence.
13:10 that that you're I you if we're going to have a regime of pre-eployment testing where the AI labs have to get some stamp of approval from the federal government then you need technical capacity I think in a new agency whether it's under the department of commerce as it currently is or some place else that has the technical talent and can be nimble enough to evaluate these models but that is just at the at the kind of model layer and and on pre-eployment testing when it comes to enforcement and sort of monitoring all these dangers, it's going to be a whole of government approach. You're going to have to have in each one of these sectors, I mean, take like the energy sector just as an example where where now, you know, the the effect of AI on on energy is going to be so important both on the prices and then and and the requirements of it. And so it's it's going to be all over the federal government. even if we want to like have some specialist in a special agency.
14:09 >> Yeah. I guess part of the problem is that AI penetrates into every part of the economy and so for example child safety you know is an issue for you know certain part of the economy but you know you don't necessarily want the national security agencies to be making decisions on what you know what's not good what's not good for children. So I the way I think about it, you know, I really do think that you need a specialized regulator. I'll just give you a historical precedent for why that's the case. The first American regulator in the late 19th century was the Interstate Commerce Commission, which was put put in place to regulate this new technology, the railroads.
14:56 when trucking came along, people said, well, you know, trucks are just another way of moving things from one place to another, so we'll give this to the IC. And that turned out to be really unworkable because the economics of trucking is completely different from the economics of railroads. So, by the time they got to airplanes, they didn't try to fold it into some master regulator. They created, you know, the FAA and the CAB and a lot of very specialized things. I would think that given how fastm moving and how complex this technology is, you really do need to get people that are really focused on this. The usual pattern I think in the US government is to outsource this thing. But you can't do that. You know, you can't outsource regulation to the people that need to be regulated. And so you really are going to need to somehow develop that that capacity internally.
15:57 >> I agree with that that we need to develop the capacity, but let let's figure out what this group is going to do because as you said, you're going to end up having all areas of federal regulation affected by AI. And so even if you were to develop the the sort of ninja AI agency within the federal government and staff it with what what would probably be thousands of people, you are still going to have to have kind of an updating of the regulatory environment in all these other areas that touch AI. and so that's why I focused on on what I think most people can think about when they're thinking about AI safety just at the model layer, which is that all right, you have, you know, and and the UK has actually been probably the most outfront on this.
16:40 Their AI safety institute has been the leader around the world. So much so that they're kind of developing the standard by which the the model developers are are sort of abiding. And so they'll they'll submit it there and people take a lot of direction from them. we also shouldn't forget the states right because the states are are being active here. It's a problem. I think everybody agrees that a patchwork quil of AI regulation is is is not not just second best but it's very third fourth fifth best right but it it's you know we can't wait because we need to get this right in the next 16 months because if we reach recursive self-improvement and the models are you know off and running I think it's going to be too late for us to come up with the right regulatory regime.
17:28 >> Yeah. So let's talk about the international side of this. You've mentioned China already in any discussion of AI regulation. You always get to a point you say well we ought to slow down. We ought to regulate you know be careful in moving forward. And then somebody says but what about China? China is not slowing down. and if we slow down, they're simply going to beat us to super intelligence or AGI or whatever the goal is. How do you think about that?
18:01 >> So, I I tend to think that the race analogy is the wrong one. That this is not a race with a finish line. and that even if they are six months behind that then you know wherever you put that marker in terms of super intelligence they will get there. and it may very well be that the benefits of recursive self-improvement or the the impact of recursive self-improvement is so significant that it it's absolutely critical that for national security reasons that the United States get there first. because I if well I said there's no finish line if the the shape of the curve is such that once you reach RSI that then your capabilities just you know say for offensive cyber attacks or or some other national security concern that then it's it's important that we get there. It's not clear to me that that that that you can't sort of develop the technology and and develop some of these technological advances that would be important to protecting our our lead here while also making sure for example you don't have agents escaping sandboxes right I mean I think there is a way to continue investing in the research here while also doing it in a safe way But you know the China argument I it it's not just about losing to China. It's about if if the technology is going to be out there and in particular since China has been developing open models right which can be adapted around the world by by anyone that as those approach you know greater and greater power and capabilities they pose a unique regulatory problem. So even if anthropic and open AI and and Google and X and Meta end up slowing a little bit and even if it doesn't the US position in terms of AI dominance if you have super capable open models coming from China then those pose you know unique safety concerns that a US regulatory regime that's focusing on the proprietary closed models is not going to touch.
20:16 >> Yeah. I wonder whether you could actually make a distinction between Chinese models that come from overseas and domestic ones and you know somehow ban the the external ones. I mean they're all connected on the internet. So >> yeah. Well, that's exactly right. And and you know, we're going to get to a point where where the open models that are hosted on places like Hugging Face, I mean, we're probably there already where especially when you get to some of these SP smaller specialized models, they're distilled from Claude, you know, plus Kimmy plus these, you know, that there's these new open models that are kind of just an amalgam of all the stuff that's out there. And so it's going to becoming increasingly difficult to attribute say harm to the actual developer of the model especially when you're talking about these kind of bespoke applications let alone as you were saying you know look it's software so it's basically going to be out in the cloud somewhere so if you're a bad actor and there aren't as many controls in China as there would be here then you're going to have access to those models. I think one thing that people don't realize is that the Chinese themselves are not libertarians who are just saying let the technology rip as fast as possible. You know, we want to be first.
21:30 I mean, you have some Americans that believe that, but that's not the way the Chinese Communist Party acts. And as I understand it, they actually do more regulation of their AI than we do. I mean, so for example, they actually have a government agency that looks at frontier models and has to get give its permission for them to be released. so they're already doing things that we would presumably, you know, want to agree with them that we all should be doing, you know, in the in the interest of general AI safety.
22:02 >> I agree. And I I I look I do think there is a space for some kind of quasi Geneva convention on AI that would at least deal with the existential risks on you know chemical, biological, radiological, nuclear and that maybe cyber. it it's not going to be easy to enforce. This is not like nuclear weapons where you can maybe monitor you know the warheads or the plutonium or something like that. but it does take a hell of a lot of energy to to train these models and it's not as just if any that anyone can develop something at the frontier and so if there were sort of greater oversight and monitoring by an international body on the training of these models at least just on the seaburn and cyber risks then I think that would you know be some kind of place for agreement between the US and China but you know we haven't seen it thus far and And certainly the administration is not pushing to have an agreement that would slow everybody down at once.
23:04 >> Yeah. Yeah. Let's talk about the politics of this. So, a lot of very interesting things are happening both domestically and internationally. in the United States, you now have really big anti- AI wings of both the Republican and Democratic parties. Right now, it's all focused on data centers. a lot of that seems to be driven by some form of nimiism or as I understand it, a lot of it is the feeling that a lot of local officials have been making these deals with the big companies and getting paid off, you know, without actually consulting their local communities. I would think that by the time you get to the 2028 election, AI is going to be one of the big issues that's going to be contested in this. Do you agree with that?
23:54 >> I do. I do. And that that's sort of where I started and I think that you're even seeing it in 2026 in particular areas as seen through the data center issue right and so whether it's for state legislature whether it's for congress there are certain parts of the country where the data center issue is front and center now what is the data center issue about there is the nimiism issue that you suggested as well as concerns about say energy rates and water and all of that but I tend to think that the outcry against data centers is just a way to focus people's diverse anxieties related to AI at something which is literally concrete right so that you can this is an area where you can you know a physical thing that you can protest against in whether you are concerned about AI's effect on your kids or whether it's on the economy or whether it is on the the stuff related to data center the noise the the energy, the water. I I I tend to think that the concerns about data centers are actually politically navigable that that that just the building of the buildout of the data centers, but because I think it represents a sort of broad concerns that people have about AI, I think that it's going to be a political flash point. I'll say that and I'm I'm writing an article on this right now and I as I think you know I'm I'm the head of the task force of the American Political Science Association on AI and we've got a a book coming out on on politics and AI. I tend to think that there is there there is a kind of residual tech lash that is still affecting people's attitudes toward the AI companies. That is a tech lash that started say 2016 with Facebook and everything that happened thereafter or Cambridge Analytica whatever. so there are people who are distrusted tech companies to begin with and then you lop on to that the fact that they're all saying that they're existential risk that this you could lead to 10 to 20% chance of the end of humanity plus the widespread job loss fears that I think some people you know really are concerned about and then other concerns related to to kids or or to the kinds of things that we've seen with social media deep fakes all of that. and so there's all of these different concerns that are now filtering in into the politics. You are right, which is that the there is only about a 10% partisan gap on on most questions that you ask with respect to AI and politics. and so that's that makes it quite different than than most other areas of American political life. But we will see whether the elite signals that are coming out particularly from the administration are going to lead this to have a more familiar partisan structure in public opinion.
26:44 >> Yeah. I think you're absolutely right that the data center issue is not the most important that it really does symbolize these more diffuse fears of what the economic and job impact of AI is going to be. I don't think we've really seen how that's going to play out yet. actually the unemployment rate is still quite low because the data center buildout itself is creating a huge amount of economic activity. So people are busy doing that.
27:16 But does seem to me that at a certain point you know a lot of occupations are going to be replaced by these machines and I don't think that we've even begun to think about how the government responds to you know this kind of technology job loss and you know do you compensate people do you restrict it do you you know universal basic income I mean there are all these ideas floating around that I don't think are very plausible agree with that.
27:47 >> Yeah. I mean I think because ultimately it comes down to redistribution, right? If any of these proposals are going to be somehow the benefits of AI or the money of AI will acrue to whether it's the companies or rich individuals or whatever and then or the government and then the government's going to redistri redistribute it back in some fashion whether it's through creation of jobs or whether it's through actual universal basic income. And look, this is not a political system that is terribly hospitable to to that, right? Just just you know, the widespread cutting of checks to everyone based off of of you know, tech profits. and so I think there's there's there's a lot to be concerned about there. I I don't see the the economic effects being as significant in the short term, but just wait to see if there is a correction if there's a slowdown in the data center buildout or or just if you know if we relive 2008 in a kind of in an with an AI cast to it, right? and and there's like a massive or a 1999 kind of crash, you can imagine what the reaction is going to be toward the AI companies then because they are as you said they're propping up not just the US economy right now but the data center that are built out is propping up the world economy because of the literally trillions of dollars that are now being invested.
29:15 >> Yeah. And you've also got this distributional issue where the profits from you know this kind of activity are going to a smaller and smaller circle of people and the general benefits are really not at all clear. So this is an issue that I've been particularly agitated about you know that and it really has to do with what's the function of work in a society because I don't think that work that you are paid for is simply a means of getting the material prerequisites to living a life you know so you can have housing food shelter you know so forth it really has to do with you know human dignity it has to do with the fact that the society thinks that what you are doing is valuable enough that you're going to be paid a salary for it. And if you take that away and substitute some kind of make work, you know, activities, you know, some of these people say, well, you know, you can pursue hobbies, but do you think that millions of people are going to be satisfied, you know, doing hobbies, instead of actually being paid for, you know, for useful labor? So I think that there's a potential you know real social rupture that may emerge as a you know as a as a result of this. okay how about broadening the aperture here? we've been talking about the US and China and not about any other part of the world. But we do have Europe, we have the global south, we have a lot of countries that are watching this, some of them kind of in horror. how do you think this is going to affect you know international relations on that level when these two big superpowers are the only ones that seem to be moving forward on this technology?
31:08 >> Yeah. So I I've been spending a lot of my time in the last year or two in places like Brazil and Japan and certainly in Europe thinking about what you so-called middle power strategy with with respect to AI. Let me just say first on on your previous point about dignity of work. you know, we've got this project here in the law school as part of the AI initiative doing some work with the Vatican on on on some of the issues that they are dealing with with AI. And that's one of the things that the Pope has said, right, is that it's not just about the ability to earn a living, but it's the dignity of work and that if if AI takes that away, then we've got real challenges. Now on the on the middle power strategy or you know rest of the world strategy I mean different countries I think are have different opportunities here and different capabilities. I think the jury is still out as to whether the so-called Brussels effect is going to apply to AI.
32:07 Right? So the Brussels effect was the idea that Europe regulates while the you know US innovates but that u the Europe sets the regulatory standard for the world. That was probably true with respect to their privacy law GDPR which then became the kind of standard that lamentable or not that that then governed the the rest of the world. but when the US is unable to act you there the Europe is a huge market and so it can it can set the standard.
32:36 It's sort of what California has done of course with like auto emissions and and may even do with respect to AI and so I think a lot does ride on Europe's position as the primary regulator of AI right now because they are the ones who have the most comprehensive law and so will they be the de facto governors? I think I I'm a little skeptical that that is going to be the result here because I think this is different than say privacy legislation and others but it's still possible that they are going to be become that body and and all of the AI companies with the exception of meta and sort of X which signed on to half of it signed on to the codes of practice that came out of the EU. And so I think the EU is still going to be out front as a regulator, but I do think in terms innovator that Europe is it's in a bad spot when it comes to trying to develop the technology and to more importantly to diffuse the technology. And so that is usually my message when I'm talking to these kinds of companies or countries which is that look, you're not going to develop your own foundation model. you're just not going to be able to compete with the US companies or with China. and so you got to think about the different layers of the AI stack. So some places like Brazil, which has abundant energy and sort of I've learned about how much renewable energy the over 95% of their power grid is now renewable.
34:06 they're in a good position to be leaders in you know if they need want to building a data centers and all of that. In addition, places like Japan and and I think also in Europe, if they make it easier for the technology to diffuse into the economy, you could see native app development, you know, and and real competition at the application layer even if they can't do the foundation layer. I think that you know it's still going to be the case that the US and and maybe China are and to some extent South Korea which is important here are going to be leaders on chip on the design of chips but of course Taiwan is still the most critical country in the world when it comes to the AI supply chain because TSMC has just figured out things that nobody else can in terms of the building of of the semiconductors. and so I think each one of these countries has has a different role to play. And I and I think that the competition between the US and China is not just over the power of the technology, but it's also the reliability of the technology to be distributed to all these other countries. I I have I've been surprised in my travels to hear how significant the administration's decision with respect to mythos and fable was in the spring in terms of casting doubt on the reliability of American AI models to be there sort of and be there reliably.
35:37 And so the fact that the president could just pull the plug on on on one of the models that other countries were using for cyber security and other things did send a chill down the spine of many of these places where they were going to deploy this. so much so that for example in in Brazil they're building two giant supercomputing facilities. One is Nvidia based and and sort of American technology and one is Chinese technology. And so they're trying to hedge their bets. And so that is what you you do see around the world is that the these countries are trying to to play both sides for the moment.
36:11 and we'll see if that that continues. And by the way, this is why the Chinese strategy of open models is going to be so critical for those middle powers, right? Because they're going to be able to adapt them to their domestic circumstances. >> Yeah. At a much lower cost. we're already getting quite a few questions in the chat. So I would encourage anyone that does have a question to put it in there. We're going to finish in a couple of minutes now. yeah. So maybe one question is related to those open-source models. You know, China destroyed the western telecoms industry.
36:50 you know, companies like Lucent and Nortell by coming up, you know, Huawei basically came up with a technology that was not as good as the leading, you know, switches that were being made by the big telecom companies, but they were a lot cheaper. and so many companies said, "Okay, we don't need, you know, to pay this premium for the best switch in the world. The Chinese switches are adequate." And it seems to me there's a danger right now of the same something similar happening in the AI world where these Chinese open- source models are being adopted by a lot of people including a lot of people in the United States because they're adequate. You know, you don't need the most sophisticated frontier model to do the kinds of work that most companies, you know, want AI to do. so I do think that I mean that's both a good thing and a bad thing. It's not good for the big companies in the United States. it may be a way for this technology to spread you know to other parts of the world that really can't afford to be at the frontier.
37:54 >> Yes. Well, I also think it's sort of unfortunate that the open closed debate correlates with the geopolitical debate of China versus the US. Now, it's not completely the case, right? because Meta and Nvidia and and others are developing open models and so I think the hope there is that maybe they will develop models that are as capable as Kimmy and and and the other Chinese open models and I think there's some cause for you know hope there but you are right if you look geopolitically and you look at at the adoption in the last year of the Chinese open models it's been quite striking you know because of the cost of the of the those models, whether in the US or or around the world, you're seeing the Chinese models getting very popular for the reasons that you say, which is that they're good enough for most things. You don't need to you know, drive a Ferrari to the supermarket, right? And and and that is you as powerful and and impressive as as the American frontier models are, you don't just don't need them for most tasks that you might need to use AI for. And so especially given the adaptability of the open models to particular kind of economic or industry circumstances or domestic you know political circumstances that you would expect there to be a preference you know in a lot of sectors for to to have an open model that you can then you know tailor for your particular needs.
39:22 >> Yeah. By the way, I'll say one thing on this, which is is a little bit deceptive because even though the Chinese models are open, you're going to need some massive infrastructure, right, just to for inference to power these models anyway. And so it's not as if you can you can just build the everything yourself in your garage with an open model to then to then run at the scale of which we want to, you know, use these models.
39:49 >> Yeah. Okay. Well, that's too bad. I I like doing things in my garage. >> Well, when you were talking about hobbies before, Frank, I I hope everyone knows about I mean, because anyone who follows your Instagram would know that I think you'd be quite successful in a hobby based economy, you know, whether it's woodworking or anything else. You'll do really well. >> Okay, good. So, I think we will wrap up this part of the session, this conversation we've had. Annette, do you want to take over in terms of going through some of the questions that are in the chat that have been posed by some of the people listening to this?
40:26 >> Absolutely. Happy to. thank you and and Nate for so for such an insightful discussion so far. as as Frank said, we'll open up to questions right now. I encourage folks to submit yours in the Q&A box that you see on your council. but before we move on to that, I just want to remind folks that today's webinar is is is sponsored by the AI power and and and new global strategy program which we'll have running on campus u in March of 2027. If you want more information, please look at your console and click the links there. All right, we will move on. we're getting some questions here, but I'll start with one that we got earlier on. okay. How much of the issue around pacing the frontier model stems from the current profit above all at all cost mentality that seems to be driving US businesses overall?
41:21 Well, I think that you you can phrase it that way, but you could also just say that that the AI companies seem to have a winner take all mentality that this is similar though in many ways it's different but similar to like Google search if to be the search engine to be the model and so therefore their their interest in pushing the capabilities is you know from their standpoint almost existential that because there is a chance that only one of them would win. Now, I happen to disagree with that. I think there's going to be room for for several of these companies. but it is, you know, a sense that that they need some outside force like the government in order to discipline them because otherwise the competitive pressures of of the market are going to lead them to just keep pushing the capabilities and maybe not spend on safety. Now, I think there's something to be said by, you know, the arguments of Jensen Wong and and David Saxs and others, which is to say, look, I mean, you have this in other industries and we have, you know, people don't release unsafe products. And so, you shouldn't just blame competition to say, look, regulate me before I kill again. and and so, you know, I think there's something to that, but at the same time, because of the antitrust laws and because of the absence of federal regulation, we don't have the ability for these companies to coordinate on safety.
42:49 Thanks for that. the next question that I have here is which regul regulatory risk are investors in compute and data infrastructure underpricing most today? Export controls, liabil liability for model harms or data rules? >> Okay, that's a very CNBC kind of question there. Let me let me think if I can if I can answer it. well which which are they under underpricing? look, I tend to think that the most that at some point in the next year, we will have some kind of roll back that then will have all these kind of knock-on effects because of the of the crazy circular financing between the the whether it's in, you know, Nvidia and OpenAI and Microsoft, but all the money, it's very hard to track and that does lead to a level of complexity opacity that I think is not great for the sustainability of of the market. Having said all that, I I I do think that, you know, the long term for these companies looks good. It's just a matter of of, you know, what the pace looks like.
43:56 And so I would think that a slowdown in data center buildout or is the thing that might then might lead maybe into sort of public perception that we are you know that that that there's a correction. But, you know, you don't you don't see that right now. They're just trying to they're trying to find some place, you know, any place that they can build a data center, they're they're trying to do it.
44:29 All right. another question we have here is not having AI safeguards and AI regulations actively fueling AI fears and affecting AI adoption, slowing slowing progress in the US. So, I don't think that what whatever regulation might get passed, whether at the federal level or whether by states, I don't think it's going to speak to most people's fears about AI. I still I think it's the right thing to do.
44:59 just because you want to regulate things. yeah, but partly because you want to chart a secure AI future. I think it's good for the companies to have settled legal rules. But I don't think law is going to solve the public perception problem for the same reason that if we were regulating social media, I don't think people would then suddenly say, "Oh, well, I I I love these tech companies." that previously I was saying we're hurting my kids or destroying democracy. and so I don't I don't think that regulation is going to solve the anxiety or the public opinion problem especially if in the coming years the way we think about AI is principally through the lens of the labor force impacts and so if you believe that AI is destroying your job right it doesn't make a difference how how much you're regulating data centers or model development it still has that effect and so I think the politics of this are complex and unsettled right now and I think that the the regulatory debate is part of that part of that political discussion, but it's not all of it.
46:09 >> All right, thank you for that. okay. What are some possible implications on financial market side of things? >> Well, you're already seeing it. I mean, all the all the big financial firms are trying to integrate AI into all of what they do. And so you know whether it's trying to predict the trajectory of individual companies the market itself I mean all of the banks are using AI and so I think that that you would expect that that would be a sector where you'd see it rolled out very quickly. in terms of like regulation though of of the financial industry. I mean, one of the things as as Frank and I were talking before about getting a independent regulatory body for AI up and running, which is something, by the way, Cass Sunstein has a proposal on this, but also Demisabas has put out a a early on a proposal to say that we need to regulate AI in the same way we do regulate financial institutions. So something like FINRA should be proposed for the AI companies. Now that's going to be a kind of public private partnership idea and standard setting and and all of that but without with some kind of accredited third party observation. and so I think that in many ways while while there's going to be important regulation regarding the diffusion of AI in the financial sector I think we can also learn a little bit from the financial sector when it comes to regulating AI.
47:45 >> Okay. how should policymakers address application regulations without also squatch squashing muchneeded AI research? >> Well, I don't I don't think those those motives are necessarily intention. I mean, I think that if if the questioner is actually talking about the applications, I mean, so much of this is downstream of the of the development of the main AI tools, right, of the of the models. And so I I think that we really need to be encouraging diffusion at the application layer. And this is what I was trying to say before about these countries outside the US and China where I do think on the development of applications that they're going to be that is where they need to really innovate.
48:34 you know, I I I think that there's the research questions and maybe what the questioner is asking is how can you do research at the major labs while also pushing the frontier or or or or pacing the frontier. And I do think that you know one of the outcomes of this hugging face incident is in part that you know these were unreleased models that or or un you know deployed models that ended up causing harm. And so these companies really need to lock down their sandboxes to make sure that in the research that they're doing on the models that it's done you know without internet access. Enthropic actually had an announcement about that today. and so that's the way you you try to protect research. You make sure that it's it's done in an environment as as you would with biological, you know, gene editing or any other kinds of things that's segregated from the outside.
49:40 >> Okay. I have another question here. What is the worst case scenario if the US does not come up with the proper and/or unified regulation in the next say 18 months? >> Let me I I don't want to be too da here on this I tend to be a little more I'd say in the normal technology camp let me say this that even if the US comes up with the perfect regulation, however we define that, it's not clear to me it's going to stop some of the doomsday scenarios, right? And so it is going to be very challenging to deal with this loss of control problem especially if we sort of posit and and Frank and I didn't really even get into this but it's almost hard to even conceive what does opensource agentic super intelligence look like right where you have agents that are off of open models that are essentially you know out there and so that that is you know that's going to pose a fundamental challenge I think that if the you know, doesn't regulate and we are the biggest sort of game in town. I think you'll see a series of small level I don't want to call them catastrophes, but incidents, you know, maybe something with the electrical grid, maybe something with stock markets, those kinds of things.
50:57 Hopefully, I I I tend not to be in the kind of existential doomer risk camp that this is that that we're going to have a terminator scenario that leads to, you know, human extinction. But I think that, you know, in the same way that we're going to see incredible benefits, possibly cures for cancer and other diseases, incredible, you know, creativity, other kinds of, great benefits from AI, you're going to have some of these periodic, you know, harms, that that happen.
51:23 this is what happens with new technologies, and this is a particularly powerful one. I mean, as many have said, right, this is like the industrial revolution, but just maybe 10 times as significant and 10 times as fast and you know, it's going to take a long time for us to to sort of come to grips with it. >> Yeah. Truly don't know, but but I love the optimistic view. okay, we talked a lot about we talked a lot about jobs. so there's a question here.
51:52 what human possibly regulatory related jobs do you see emerging in 10 years time that don't exist today? >> Well, I still think that there are existing jobs that you know are going to be retoled, you know, for the AI era. We think a lot about this in the law school. I'm actually teaching a class now on the use of legal AI tools to the first year law students. And you know what we're seeing with law firms, law firms don't seem to be reducing hiring, but what they what they end up doing is that the senior partners are just more capable now because they can use AI to do a lot of the stuff that the junior folks were doing. And that that is, you know, something that I think we should be legitimately concerned about on the labor side, which is that the young people who are coming out especially if they're not AI trained are going to have a difficult time. but as to what what kinds of jobs there might be in the future that that we don't see now, it's it's almost hard for me to describe except to say I can see in every industry AI skills being, you know, a a new kind of job. And so, just to give you an example, I I'm spending a lot of time with these image generation or these video generation tools right now, just experimenting with them. I've developed an AI avatar of myself that can give an oral exam to my students that so they can actually have a conversation. And so this is something I never would have done, you know, 10 years ago. I the idea of I mean the technology wasn't there, but even the the idea of of experimenting with that task is not something a professor would have done. And so I think there are going to be all kinds of things like that where where in particular industries there going to be firms that are going to be hiring people who have these AI skills to perform tasks that they didn't they didn't do before.
53:43 >> I think also one new job category is actually managing AIs. >> Yeah. >> Right. Because most agentic AIs you can't just let them run and expect that they're going to do you know the right thing. So people, you know, it's and one way to think about the future AIs is that they're like you know, they're like workers. They're they're going to do stuff that human beings used to do, but all workers need to be managed. And so having the technical knowledge to manage a machine is going to be an important capacity that you know is going to exist in the future.
54:22 >> That's right. And I think there's going to be there going to be lots of roles that are defined like that, right? That that there's going to be sort of AI management is implied in almost any senior role at a company. >> Yeah. okay. I think we have time for about one more and I'm going to leave us with sort of a fun one that I see here. what's your take on Bostonramm's be careful what you wish for view in his book Deep Utopia?
54:52 I'm I'm going to leave that one to Frank. Frank, do you have do you have thoughts on that? >> Well, I don't really believe in any of the big doomer scenarios. I think that it's actually hard to do the kind of damage that a lot of them are expecting. And I suspect that the degree of human control is actually going to be something that will be developed. A lot of it is going to require AI. you're going to have AIs that are going to be managing other AIs and defending from AIS. and I'm not sure that that's a process that will simply get you know, get out of control. I really do think that people ought to focus not on these we're all going to die scenarios, but really on the near-term very serious harms. you know, like Chase Bank, you know, a million of its customers get their checking accounts emptied out, you know, this sort of thing. and that's pretty bad. You know, it's going to lead to a real crisis in the United States.
55:56 and that seems to me much more likely than any of the, you know, the pdoom kinds of scenarios that people are positing. that I agree with and and and that was what I was trying to emphasize before, which is that yeah, I could see, you know, a series of kind of real AI breakthroughs and then accompanied by these AI harms and these AI incidents, whether you're talking about the electrical grid or stock market or or or something else.
56:22 >> Great. All right. Well, thank you again, Frank and Nate, for this talk on such an important topic. to everyone in the audience with us live today, thank you all for your questions and participation. I want you I want to remind you that today's session was recorded and a link to the recording will be made available to all of you within a week. I hope everyone has a wonderful rest of your day and we'll see you all next