AI Law in the Laboratory of Democracy with Kevin Frazier

Greetings fellow AI optimists. I'm Sophie Singletarian. I'm excited to welcome you all to today's podcast installment of Working Intelligence Making Americans Irreplaceable. This is a campaign to highlight the story of AI on the ground in America. And today I'm joined by Kevin Frazier, the AI Innovation and Law Fellow at the University of Texas Law School. Kevin clerked for a Montana Chief Justice, got his law degree from UC Berkeley, a master's from the Harvard Kennedy School, and today lives in Texas. So he is truly an American man through and through. He's got robust lived experience across the US and a very nuanced understanding of both state and federal law. And there's truly no one better than Kevin to discuss the burgeoning field of AI law and regulation. So grateful to have you on today, Kevin. Thanks for the opportunity, Sophie. And I think I'd argue that you're the Irreplaceable one. I really enjoy this podcast and it's a good to get to join you. Glad to have you on and figured we could start at the beginning here. You have had this fascinating path across the country, Berkeley Law, Harvard Kennedy School, you clerked in Montana, teaching, and now you're in Austin. How did you get here and what drew you to this intersection of AI law and innovation policy? Yeah, I wish there was one clean through line, but I think the best explanation is that I realized pretty early on after working for the state of Oregon government and then switching to the entire opposite side of the spectrum and working for Google that there is a huge gap in knowledge and understanding between state governments and the federal government and then the technology side of the equation. And both actors really want to do great stuff. Anyone you talk to in tech says we want to improve the world, right? We want to make the world a better place. Anyone you talk to in government will say the exact same thing. Why do you work for the government? I want to help the public. I want to inspire people. And yet for whatever reason, it seems like these two parties are usually adults. And so by virtue of having this experience pretty early on in my professional career, I realized what we need are translators. People who can help explain what's all the great stuff that tech is up to, both the garage engineers who are building the future, as well as the big tech companies that are solving problems on a global scale. And how can we translate what they're doing to local governments all the way up to the president of the United States? And I try to play a very small role in making sure that policymakers understand the true benefits and potential of tech so that we don't inhibit it and limit it in a way that misses out on all of the opportunities, all of the gains we could make more ubiquitous for Americans across the country, many of whom I've gotten to meet by virtue of living everywhere from Portland, Oregon to Miami, Florida and lots of places in between. Well, it's a fascinating nexus, certainly at the intersection of AI and innovation tech and law. This is a novel area of American jurisprudence, this AI law and regulation space, paint us a picture of the legal landscape kind of as it stands. Yeah, so that's basically asking me to paint you a Jackson Pollock painting, which is to say it's all over. It's splattered. There's crazy colors everywhere. Unfortunately, we're seeing the sort of patchwork approach to technology develop that we saw, for example, define the privacy fight. Whether it's the local level or the state level, there's a sort of rush to regulate AI. And I understand it, folks feel like we didn't get social media right. They look at their kids. They look at reports of mental health issues. They look at their own screen time. And they think, by golly, something didn't go right. And the inclination now is to say, we need to clamp down as quickly as possible on AI to make sure we don't get it, quote, unquote, wrong again. And so that's showing up in laws like bands on AI therapy tools in Illinois. That looks like limitations on what AI tools can be used for in California. And then there are more than a thousand AI specific related bills pending before state legislatures right now in 2025. And that's just insane to think about all these different approaches, all these different definitions of AI. All while we're still trying to figure out what are the best use cases? How does AI even work? How can we truly maximize its benefits? And so the state of AI regulation is kind of a mess. Fortunately, we are lucky to have leaders at the federal level, like representative Overnalty from California and representative Liu from California, who are really smart on this topic and have acknowledged on several occasions that if we want to make sure the US remains a leader in AI innovation and can push the AI frontier, then we're going to need federal leadership in this regard. And I think that momentum is building. It's been quite the fight, but Congress is increasingly aware of the fact that if we're going to keep pace with China, if we're going to realize the full benefits of AI, then we need a national approach to making sure we can push that frontier. And before we dive in here on the national versus state approach, you mentioned privacy law, which is kind of the sacrosanct pillar of American law, although you've argued that traditional privacy law might actually be antithetical to AI development, which is a pretty provocative state of being and in stance here. Walk us through what new paradigms for this collection and storage of data related to AI might look like that balances innovation and these legitimate privacy concerns. Yeah, so I am very attentive to the fact that people want to make sure that their data is being used in a way that aligns with their interests. That's something I think everyone would go to bed comfortable with admitting and saying we want to make sure our information is being used in a responsible fashion and in a way that keeps our goals and aspirations in mind. The problem is the current privacy paradigm does not do that. The notice and consent approach to privacy law is I have to be a little blunt here. Kind of a farce, right? So if I'm going to admit that when I see a little thing pop up, a little banner pop up on my browser asking if I want to decline all cookies or accept all cookies, I blindly click it when I see a and what does that even mean? I do it to whose cookies for what reason, you know, what is it? I love cookies. I love cookies. Most Americans aren't steeped in the weeds of these privacy questions. And when we ask them, for example, to pour through terms of service agreements and click through all these different arrangements, it's what some legal scholars have referred to as processed theater. It's the idea that by virtue of having more steps, that's somehow going to be more protective of that individual's interests. And that's really not working. We see that our data is being sold to data brokers and third parties without recognition. We see that most Americans aren't being stewards of their information in a way that I think a lot of legislators thought these laws would work out in practice. So we need to admit that. We need to admit that this isn't the way we should regard our privacy and regard our data in the 21st century. And what's critically important to point out is that our data now in the AI era has new value. So in the same way that we didn't regulate uranium very strictly if at all in the 1930s, right? Who cares? It's uranium. It's just something in the dirt. Who knows what we're going to do with it? Well, when new technology comes around like nuclear energy, we change the approach to how we regard those inputs from a governance perspective. And now data today is one of the most important assets for our national well-being that we have. We know that good data, good quality data, and high vast quantities of data can be transformative with respect to the development of reliable AI tools. And so we need to do a lot of education around data privacy and data maximization for the American public because we won't get AI companions that are reliable until we have training data about how to make AI companion tools more responsive to the needs of child users, for example. And we won't have better AI tools for a medical context until we have access to more medical data. And so what I really want us to begin to think about is how can we treat data as a national strategic asset. And that's the exact approach China's taking, and it's helping them catch up in the AI space, and I'd argue that it's an approach we need to emulate here, while of course being attentive to things like surveillance concerns, as well as leakage to bad actors. And you mentioned China there. I mean, the CCP has pretty much full reign over the data in China. What's the blocker in the US to access to this data that you just outlined would help in the development of these elements? Yeah, Sophie, I think this podcast isn't meant to last three hours. So I can't go through the entirety of all the barriers, but just to list off a few acronyms to give people a sense of the number of outdated privacy laws that are still on the books. There's FERPA, which limits access to students' education records. We're not going to have good AI tutors until we have good data on how students learn. There's COPA, which limits the information available when children are on the internet. And I'm not saying I want to harvest all of the kids' data and sell it to some bad actor or anything like that, but we can't develop more responsible tools unless we know what a bad tool might look like, for example. So FERPA, COPA, there's HIPAA. HIPAA requires the deanonymization of 19 different fields of information before it can be disclosed and used in certain contexts. That information is oftentimes what's going to be pivotal to making more accurate medical diagnoses. Then we have the FCRA at the state level, we have the CCPA, we have the Colorado Privacy Act, the Virginia Privacy Act, the Washington My Health My Data Act. I told you, Sophie, I didn't want to go on for hours and hours and hours, but all of these laws weren't designed with the fact that AI can unleash human flourishing if we have more access to high quality and vast quantities of data. And so these privacy laws, while very well intentioned at the outset, have become barriers to a better reality for a lot of Americans via better AI tools. And Kevin, you've mentioned a number of states and laws in the books. I think you said it's something like in the thousands of laws that are, I guess, currently being debated or drafted. Federalism, though, is pretty fundamental to our well-functioning democracies. We think of states as these laboratories of democracy and enable citizens to participate more locally. With this also comes a bunch of different state laws. I was reading just recently, California passed SB 53, which is mandatory reporting of certain safety incidents and the publishing of these so-called safety frameworks and other seemingly burdensome measures. Question here is not a matter of whether AI will be regulated, but by which be it the federal government or the state government? It seems inevitable states are already well on their way. What do you think would do a better job of AI regulation and kind of talk us through what that would look like if at the federal level? Yeah, so I wouldn't be living up to my role as a law professor if I didn't force us to go back to the 18th century and talk about the articles of confederation. So, Sophie, it's critical that we know that we've been here before. We've tried this thing where states have pretty much free reign to try to dictate what are national concerns. Now, regardless of who you talk to on the AI question, whether they are really behind states regulating or really supportive of the federal government regulating, the reason they're so concerned about this issue is because they realize AI is both a national security concern and a matter of our economic stability and prosperity down the road. Well, those are the very sorts of tasks that the founding fathers wanted the federal government alone to dictate. That's the whole reason we have the commerce clause. The idea was to make sure that Congress alone was going to set the rules for national markets. And yet, here we are again, with states trying to regulate what should be a national question. Again, I understand the temptation of folks in California and in New York to address what some regard as very near term, very significant risks. And it's important that we're attentive to those and measuring them as best as possible. But the problem is we've got a constitution that has clear lines about the roles each actor is supposed to play. And while I support states acting as laboratories of democracy, there are certain attributes of experimentation that aren't being adhered to. So I'm not a scientist full disclosure. But the typical experiment only lasts for a finite period has a clear hypothesis about what the interventions going to be and what the measurable outcome may be and is contained with an environment like within a laboratory. Well, with bills like SB 53 in California and SB 205 in Colorado, these are experiments run a mock, right? What happens in California right now is not stain in California. And what happens in Colorado is not stain in Colorado. So this is an experimentation, right? These are states projecting their legislation into other states. And that just goes against a constitution that was designed to make sure that everyone had accountability over their elected officials. And you and I, Sophie, we don't get to vote on what Governor Newsom does, for example. I wish I did. No say, though. I'm a proud Floridian. There we go. There we go. I'll stay there. Some have called for a moratorium on AI development, which seems absolutely ludicrous, not to opine here. I know I'm asking you this, but tell us a bit about your take on these pause proposals. And would they ever make sense? Why do them? China's not pausing their development. I mean, what's the logic there? I think the logic is a lack of imagination and a failure to appreciate that there are so many AI use cases that people aren't paying attention to. So if you go and you look, for example, at open AI's summary of how folks are using chat GPT, only 1.9% of interactions with chat GPT are for quote unquote relationships. 1.9% out of 800 million users going to it for these sorts of relationship dynamics. And yet that's all we read about in the headlines. That's what we see dominating political conversations. We're not paying enough attention to the 98% of other uses, a vast proportion of which are doing things like expediting drug development, accelerating our ability to predict where storms are going to occur and help warn individuals to get to safer locations faster, helping like we're doing in Austin at Alpha Schools, helping improve education by providing everyone with an AI tutor that is responsive to their learning styles and needs. It's those use cases that really show the flaws with thinking that a moratorium is aligned with what's best for society. Because I'd argue if we look back to well before you or I were born and well before we were even having these kinds of debates, we see that there is a very strong correlation between technological progress and societal progress. And so to call for a moratorium on a technology that can extend lives and improve lives to me just misunderstands that fundamental relationship between mankind's development and the development of technology. You mentioned just a couple of the novel ways that AI is being used, drug development among them, and also this notion of AI serving as a companion. There are of course, a multitude of ways. I didn't know that it was that small. You said 1.9%. 1.9. And that's just as relationships generally, right? That's not even the fraction of users who are miners using it very specifically for problematic relationships. Right. And so, you know, concurrent with the development of AI, we have this very real kind of anxious social media generation, this youth mental health crisis, and really loneliness crisis at large, which we are right to worry about. You've thought about how AI could actually be a tool to help instead of harm these folks. Talk a bit about, you know, what your vision for AI is as a companion and as part of the solution to help these lonely people. Yeah. So, Sophie, I think it's important to ground this in my own experience with mental illness. I was anorexic as an 11-year-old, and you don't meet a lot of anorexic 11-year-old boys. It's not exactly a cool thing that you announce to your classmates. It's not something you advertise or anything like that. There's a real stigma with announcing you have any sort of mental health illness, especially something that's typically associated with females, regardless of incidence rates. That's just the kind of popular perception. And so, I'm always thinking about what tool could have made my experience easier, and can make the experience of other boys, other girls who are going through similar crises. And for me, this is the part of the mental health conversation we're not having, which is to say for a number of our teens and for a number of our kids who are dealing with mental health crises. They're not going to tell their parents about it. They're not going to tell their teachers about it. And so, if the choice is providing them with absolutely no care, which is going to occur as long as that stigma still exists, or finding ways to improve their access and providing them with the education and resources necessary to learn about responsible uses of AI tools. I very much want to live in that world in which we're not assuming that everyone has access to a great therapist or a great guidance counselor, for example. So, I really see that when we start to ask the question of compared to what, we really learn about how AI can be transformative. And the fact of the matter is when we ask compared to what in the mental health crisis, most people cannot afford a therapist. Many Americans live in rural communities that don't have therapists trained for a particular mental health issue, especially a mental health issue that a child's dealing with. And so, I just want to encourage legislators to see this as an opportunity and to think about how can we encourage the adoption of these tools in a way that aligns with our expectations and the needs of our kids, which are growing every day, unfortunately. I couldn't be more with you. And anecdotally, my mom is a therapist. And if there's anything she says, there is a shortage of therapists and outsized demand, unfortunately. And as you mentioned, too, multiple barriers to therapy, whether it be cost or you know, rural areas, as you said. So, I see this as an avenue for offering people help that may otherwise be unable to get it. So, with kids though, so we've talked about kids here, there's also this notion that AI companions could really help seniors. Many of them are also lonely. What do you, what's your take on that? What's the plausibility of its role out there? Yeah. And I don't want to sound again like a broken record, but it's compared to what? We know that a lot of senior residents are oftentimes in homes that are distant from their loved ones. They may not have as many friends who are around anymore. And so if the question is, for example, in the context of my grandma, she had one bridge game every other week. And that was awesome. And I'm glad she had that community, but slowly the bridge table got smaller and smaller. And suddenly the game started occurring less and less frequently. And so if there's some tool available that can help her talk about the Portland timbers who she loved and get the latest stats and learn about the players and even interact with a player's avatar, things we'll see in the near future. All of that to me is vastly superior to her, for example, sitting on the couch and watching a big series of Jeopardy. I mean, if you're binging Jeopardy, I think we can do better than that. And I think we can come up with tools that are superior. And this is just again a matter of us all asking to expand our imagination, expand our creativity and not come from a techno pessimist perspective. We can build these tools. We just have to make sure we do it in a way that acknowledges some of the risks and some of the limitations. And then working diligently to make sure that the tools are reliable and accurate and aligned with user expectations. Well, I appreciate all of that. And I know similarly to law and AI, this podcast has been kind of a patchwork of topics. I want to go as we wrap up here back to China. Thank you. I want to go back to China, not the country, but the topic. And talk about, you know, I wonder if you could outline how America's legal infrastructure, which has often criticized as slow and fragmented and confusing at the at the state visa, be the federal level. How does the legal infrastructure in our country actually become a competitive advantage against China in the AI race? Yeah. So the best part about the American legal system is our commitment to the rule of law. When we are champions of clear rules and clear guidance and efficient court systems, that's when we thrive. And that's when our legal system is something that people can count on for clarity and people can make strategic. And as we're seeing today, massive investments in new technology. But the problem is we're seeing increasingly ambiguity and vague terms or vague our legal system. So right now, for example, we still don't know what the status of copyright law is with respect to training AI models on copyrighted data. That is very much hindering AI innovation. And that's a real problem. The rule of law demands clarity and not clarity over the course of four years or five years or whenever the Supreme Court decides to take on this issue, we need to be acting clear and in a faster fashion to make sure folks understand the rule of road, excuse me, to make sure folks understand the rules of the road and can act accordingly. Whereas China, you can have a decision come down from on high in the flash of a minute and all of a sudden the legal paradigm shifts. You can see, for example, that they'll just decide not to enforce certain laws. And that sort of Helter Skelter mentality to the rules of the road can really hinder AI innovation. And so if we get back to that adherence to the rule of law and trying to specify clarity as quickly as possible, then I think we're going to see that we can continue to lead this AI race. Well, thank you, Kevin. I agree that we are in the lead and can maintain our lead, especially with people like you thinking thoughtfully about these topics. So thank you. And as we wrap up here, we like to end with what we call the bright side blitz. If you rapid-fire questions, if sounds good deal. Let's do it. I love a good blitz. Awesome. Okay. AGI within five years or beyond. Definitely within five years. Bigger threat to innovation, over-regulation or under-regulation. Over-regulation. That's an easy one. I think that we're seeing too many small firms have to deal with way too many regulatory regimes and they just don't have the resources to be able to navigate that labyrinth. Most overrated AI concern. Wow. Again, I don't know. This is an unfair blitz question because I want more time. Most overrated hallucinations. Well, I hallucinate in speech every day. So the model is at least better than me. Okay. And then last question here for you. What are you most optimistic about the U.S. about America beyond AI? I still fundamentally believe that this is the best place to have an idea. You can go anywhere else in the world. You might find better coffee. You might find better desserts. Maybe even a better steak, although I doubt that. But this is still the best place to have a really good idea, to find collaborators, to find capital and to literally build the future. And that's always what's going to define us. And we need to make sure that continues to be the case. Well, what a wonderful way to wrap us up here. Thank you so much, Kevin, for your optimistic, optimistic outlook on AI in terms of regulation and law in general goings on in America. So thank you so much for joining us. Thanks for the opportunity. And thanks for this space. Awesome. Take care.