S1: Welcome in San Diego. It's Jade Hindman on this episode of KPBS Midday Edition. What could effective regulation look like for AI? Well , one of the leading scholars on global policy , David Vitter , joins us. This is KPBS Midday Edition. Connecting our communities through conversation. Recent events in the world of artificial intelligence have left many within and outside of the industry fearful , fearful of the technology , behaving in unexpected , even destructive ways outside of human control. Just yesterday , the Wall Street Journal reported OpenAI , the maker of ChatGPT , says it will not be releasing its latest model due to safety concerns. And with those fears have come growing calls for more rules to rein in the industry. But what might effective regulation actually look like for AI. Here to talk more is a voice we often turn to. With these big questions of our time. David Victor is a professor of innovation and public policy at UC San Diego's School of Global Policy and Strategy. David , welcome back to midday Jade.
S2: It's always great to be here with you.
S1: Great to have you here. Um , listen , before we turn to you about AI , you know , as we as you know , we always talk to you about a range of topics to better understand policy developments. Um , and that includes environmental policy. We just got done talking about the Trump administration's move to rollback fuel efficiency standards for new cars in the coming years. How big of an impact do you expect that move to have on emissions in this country?
S2: I think in the short term , it's not going to have a huge impact. It takes a long time to develop a new model of a car. The auto industry is always looking down the road. They don't know what's going to happen after the Trump administration. The Trump folks are only in power for another two years. If if somebody from that ilk gets elected back into the white House , then maybe it'll last longer. But there's just a huge amount of inertia in the system , and I think everyone knows we need to be pointed in the direction of greater efficiency , greater cleanliness and so on within within reason. So I don't expect a massive impact right now because folks are handicapping how long these rules will actually last. Hmm.
S1: Hmm. Interesting. All right. Um , so now back to to artificial intelligence. Um , in a recent piece you wrote for the Los Angeles Times , you say the AI industry might want to look how look to how another industry with safety concerns and how they approach regulation. And that's the nuclear power industry. And you start with taking us back to Three Mile Island , the accident that happened in 1979. How does that moment mirror where AI is today?
S2: Well , there are no perfect mirrors , but one of the things that just strikes me about AI industry today is that it is similar to the nuclear power industry in many respects , it offers enormous promise in the nuclear power industry , offered the possibility of of inexpensive power , huge volumes , relatively clean , on and on and on. But it depended on public confidence. And in this country , Three Mile Island , really , more than anything else , shattered that confidence. It happened at the same month. The Three Mile Island , the accident of Three Mile Island happened. The China Syndrome movie was also playing in theaters , so that didn't help the industry as well. But I think what I wrote in the piece for the Los Angeles Times is that what the industry realized is that the whole , the viability of the whole industry was no better than its worst performer. They were all hostages of each other , if you like. And and because of that , they all had an interest in policing each other. And so they set up a mechanism and actually , frankly , informal , non-binding mechanism , at least initially , that went out and did peer review inspections of all the commercial reactors in North America. It's now the model for a global version of this. And so it's not perfect. Some of the risks around AI are much bigger. Some of them are not just about how you operate the AI systems , but how you invent them. There are examples from other parts of history around invention of risky technologies. But but the bigger idea that I wanted to draw people's attention to was that this is not our first rodeo. We've had other industries that that have needed to to , frankly , police each other in order to keep public confidence and then also , frankly , to keep risks low. Hmm.
S1: Hmm. You say there are two main risks that can come from regulation. What are those?
S3: Well , I see.
S2: Two right now. And and I think one of the things we need to have in the as AI evolves rapidly is modesty. We don't really know how to think about these risks. And so some of these challenges are in the realm of the unknowable. And it's one of the reasons why we need a very flexible , but also a very capable system for looking at a variety of different risks and also , frankly , looking inside the companies at what they're doing and what their practices are. This is why the news about some of the companies ignoring warnings from their staff about risks and so on , That's that's particularly alarming , not just because the risks are there , but because we're not getting a window into what's really going on with the technologies. But broadly , there are two kinds of risks. One risk is , is the risk associated with inventing something that could go , that could go off and do dangerous things. Think about genetic engineering is like that. You you invent using genetic engineering , new strains , uh , going back to the 1970s and the innovations in recombinant DNA. And if you do that improperly , then things can escape the lab. And we've seen this with the lab leak hypothesis around Covid 19. That's a kind of invention related risk. And then there's operational related risk. You've got a technology and how you use the technology. So for example , how intelligence agencies use AI technologies or how , um , cyber security mechanisms use AI to understand their vulnerabilities and then might accidentally along the way , create additional vulnerabilities for the operation of the technology , can create risks as well. And the experience with nuclear power is more of the latter type. and I think we have to look to other industries to to identify successful ways to , to regulate or govern the , the invention related risks.
S1: I mean , but can an industry , particularly one with real dangers here regulate itself? Is that even possible?
S2: Well , almost all the examples in history of successful self-regulation happened because governments sitting around and and ready to do worse or ready to step in in various ways. The electric power industry , for example , built a mechanism to largely regulate itself to improve reliability of the grid. Why? Because they had a giant blackout on the northeastern United States , and they knew they were going to face more onerous and inflexible government regulation if they didn't get their act together. And to me , this is the really big risk today. The signals right now out of the United States federal government are that they don't want to regulate AI. Now , the signals are always changing in Washington. And and the political winds are shifting. So maybe maybe attitudes in Washington will change. This to me is a huge , huge risk for the industry and frankly , for for all of us is this is an inadequate regulation or inadequate threat of a government in regulation. And so that that to me is a big concern. That said , the industry in some sense , the industry is going to have to do an even better job of stepping up and forcing itself to a higher standard because there will be changes in government. They will face massive exposures. They may already face massive liability related exposures in Woods Piece and Wall Street Journal a couple of days about the growing liability related threats. Some of that we're seeing in the news around OpenAI and the evidence of escapes there. And so they've got a lot of exposure that they've got to find a way to contain. And it's that motivation. It's kind of existential motivation for firms that's going to it's going to drive them. It's not going to drive off firms. Already we've seen big variations. You know , some of the firms seem more excited about this. Anthropic very much in the news perhaps open AI , others less excited perhaps. Microsoft seems almost certainly meta. And so we've got variation inside the industry. And the industry's got to find a way to get itself organized. Otherwise we're just going to have kind of chaos. And then I don't think we're going to have any regulation or self-regulation.
S1: Well , other industries like aviation , for example , have become far safer as a result of strong government regulation. What threads do you see between the rise of commercial aviation last century and AI now? Yeah.
S2: Yeah. Aviation is a terrific example. So aviation is one of those many industries that has been regulated also , frankly , also sponsored very heavily by government. And then the logic in aviation is very similar to the logic in nuclear power. Leading firms , airlines the most exposed their brands have been the most exposed. The airlines realize that the the viability of aviation as a technology overall required. People have confidence that whenever they got in an airplane , anywhere , as long as it's piloted by proper commercial pilots and in the air traffic control system and so on , that that system would be incredibly safe. And so in addition to government regulation , FAA and others , there's also a very extensive self-regulation system where largely self-regulation system , where people report on each other's failures and on their own failures , and they learn from that. And we've got to encourage more of that. I think one one kind of piece of good news in AI is that you've got a lot of firms that have had little skunkworks inside the firms and little organizations in the firms that have been trying to figure out what regulation would look like. And then you have a lot of people who are motivated by the promise of the technology , and they want to help make it safer. And so you've got a lot of folks who frankly , have , you know , given up big salaries and stock options and things like that. They're still making plenty of money , but they're funded principally by philanthropy to try and figure out how to keep the industry. The jargon is aligned or alignment , how to make sure that innovations in AI stay aligned with what humans humans want.
S1: So , you know , I mean , as we mentioned earlier , OpenAI , the maker of ChatGPT , is not going to release its latest model due to those safety concerns. Does that news at all change how you're thinking about regulating AI?
S2: No , I think that's evidence that that particular firm and they've always known in various ways that that firm knows that it's now it is working with technologies that are increasingly powerful. And it has a very powerful self-interest in in notwithstanding all the news reports about escapes from OpenAI. It has a powerful self-interest in , in not releasing something that's going to be out of control or create a lot of damage , not least for their liability risk. Anthropic did something earlier this year with their latest , uh , models. I think what we're not seeing these models go to the shelf permanently. What we're seeing is a slowdown and more introspection and careful testing and process testing. All of that is good news. What we need in addition to that is some more transparency so that without releasing too much intellectual property , all of us are able to see what we learn from those mechanisms of introspection and self testing and peer review , and that some of the people who are at the frontier of the technology can look at different firms and understand what's going on with these different models. That'll create , I think , a lot more confidence , not perfect confidence , but a lot more confidence in the technology and probably will help us set up some figure out where the guardrails are that are really needed so that we don't create things that inadvertently cause a lot of risk to humans and to life and to nature.
S1: All right. Well , so we've been talking about , you know , some ways we can think about AI by comparing it to other industries. But what unique challenges do you see from the rise of AI?
S2: I think the number one is speed. Um , in my lifetime I've been I've lived through now what? This is the second major technological revolution. I was born in the as the electricity revolution was was continuing apace. But but I've seen the IT revolution and now the AI revolution. Each generation has moved faster. The speed of change in the AI revolution has been just extraordinary , so it makes it very hard to know what's going on. I'd add to that the problem of government if the government is AWOL , that's harmful because it doesn't create any credit. It minimizes the credible threat to firms that that that is one of the motivations for them to , um , to , to regulate themselves or , and to work with government to regulate themselves. And then the last thing I'm particularly attentive to is international cooperation. Right now , in many ways , this is an easier regulation problem because the leading firm leading models are almost all in the United States. And so and by a limited number of firms. So you can figure out who to get in the room. They might not agree with you trying to be fair , who can get them wrong. But ultimately we need to have this effect , the behavior of all AI related innovation. That means the United States and China probably also have some of the leading models that are emerging in Europe. They're very far behind. Eventually we'll see some models in other countries. And and so there'll be. There will be an international cooperation problem that emerges over time. We're not ready to tackle this as international cooperation because we don't know how to frame the problem. We don't know what we can really do here at home. And so it's not surprising to me that when Trump and she met last week , although people have been talking about the need for them to get together on AI regulation , they didn't produce anything meaningful in that regard because because the countries don't agree on a lot of stuff , and they certainly don't agree on what to do here. Interesting.
S1: Interesting. Well , I mean , it's been nearly four years since ChatGPT released so quickly , like unleashing this , you know , they unleashed this wave of AI growth and development. But what's what surprised you most about what's changed in that time frame? And in that four years since its release.
S2: What's changed is the familiarity that people have with the technology. They're learning lots of ways to use the technology , I'd say right now , mostly for good , although there are some innovations in particular around creating bioweapons that that give me a lot of pause and a lot of concern. So the speed with which people are learning about this is very interesting , and that portends extraordinary impacts in the economy. One of the reasons it took so long for electricity to have full , big impacts on the economies. It took a century to learn how to use electricity in the most effective ways. The innovations around AI not happening overnight. They're happening quickly in places that are familiar , like coding more slowly in back office functions and so on. But the speed overall is pretty extraordinary.
S1: I've been speaking with David Victor , professor of innovation and public policy at UC San Diego's School of Global Policy and Strategy. He is also senior fellow at the Brookings Institution. David , as always , thank you so much for your insights.
S2: Well thank you. Great to be with you.
S1: That's our show for today. I'm your host , Jade Hindman. Thanks for tuning in to Midday Edition. Be sure to have a great day on purpose , everyone.
