Loading summary
Transcript30 lines
- [00:01]
GMC Advertisement
At gmc, ignorance is the furthest thing from bliss. Bliss is research, testing, testing the testing until it results in not just one truck, but a whole lineup. The 2025 GMC Sierra lineup, featuring the Sierra 1500, heavy duty, and EV. Because true bliss is removing every shadow from every doubt. We are professional grade. Visit gmc.com to learn more.
- [00:36]
Stephen Overlea
Hey, welcome back to Politico Tech. I'm your host, Stephen Overlea. The House is still trying to pass President Donald Trump's big, beautiful reconciliation bill, including a provision that would fundamentally alter the way that AI is regulated. The bill blocks states from enforcing AI regulation for 10 years, a whole decade, which practically speaking, means no new AI laws in the US at all. Since that's where all the action has been happening, the provision will have to clear the Senate and get Trump's signature. And it's already divisive. State attorneys general, civil rights groups, and AI safety hawks have all come out against it. Industry, on the other hand, couldn't be more thrilled. And then there are folks like Sam Hammett. He's the chief economist at the foundation for American Innovation, a center right think tank that generally opposes tech regulation. But Sam says the proposed moratorium is broad and long and may not actually be the best idea. I called him up before the late night action in the House to talk through it. Here's our conversation. Sam, welcome to Politico Tech.
- [02:02]
Sam Hammett
Thank you for having me.
- [02:03]
Stephen Overlea
So, to me, a decade with no state AI laws sounds kind of like a lifetime. I mean, especially with technology that progresses as quickly as AI is progressing. What do you make of this proposal?
- [02:17]
Sam Hammett
Well, I tend to agree. I'm really of two minds. First of all, there's been over a thousand bills introduced in state legislatures to regulate AI in various ways, some on similar topics, but not always mutually compatible. And so there is, I think, a genuine, valid worry about a patchwork forming, particularly because AI is a type of technology that is really deterritorialized. It doesn't really matter where the system is produced, whether it's even in the United States. It crosses borders, it crosses state boundaries. So if every single state has a different requirement, that becomes a huge burden and a burden that only large companies could absorb. That being said, this moratorium is quite broad in its scope, and frankly, we don't even know all the things that we cover to date. The kinds of bills that you see coming out of state legislatures deal with bias and discrimination, which I think are understandably hard to define. But then you also have things like whistleblower protections or basic Reporting and transparency requirements. What new kinds of regulation will we want or will we need that we haven't even conceived of yet? And can we count on Congress to move quickly?
- [03:26]
Stephen Overlea
And so where do you stand on that? I mean, do you think this moratorium would be a good thing? Can we count on Congress to actually take action if some of these issues arise?
- [03:35]
Sam Hammett
Yeah, I lean towards waiting and seeing. You know, first of all, there's a question whether this is even allowable within reconciliation. Right. The reconciliation process is supposed to be about fiscal and budgetary issues, and this is not. That being said, we're sort of at this inflection point in the progress of AI development where you listen to the leaders of leading AI labs like Dario Amode at Anthropic, or Sam Altman at OpenAI, or the folks over at Google DeepMind, they're all pointing to this year and really the next two to four years as being a major inflection point in the power and capability of their models. The power and capability for the kind of mundane tasks we use them today, but also the ability to autonomously navigate a computer. Or there's a recent company that just posted state of the art benchmarks on a hacker agent, something that can find zero day exploits in any code base. ChatGPT came out roughly two and a half years ago. We don't know what the state of the technology will be in a few years, much less 10 years. And so I think the biggest count against this moratorium is it removes option value. Right. Congress may step up to the plate, but if they don't, then it's, you know, we're removing 50 other legislatures that could want to weigh in.
- [04:48]
Stephen Overlea
Well, that's, you know, realistically, I have to say, Congress tends to operate on a deadline, right? That tends to be when they do most of their work is at the 11th hour. So if Congress has a decade to develop federal standards, do you think there's any reason to expect it would happen sooner than that?
- [05:06]
Sam Hammett
Unlikely. Standards is a hard thing with AI because it's not like automobiles where you have a really well defined objective like the fatality rate in a car crash. These systems are changing on a month to month basis, week to week basis in some cases, and their behavior is not easily generalizable. There are AI models for predicting the weather. There are AI models for doing your taxes. There are AI models that will superimpose your face onto a naked body. And so because of this broad panoply of potential issues that arises out of AI, I do think we need more Eyes on the issue and more options on the table. My biggest criticism though, and I can honestly argue both sides, because I think there is this sense in which we're in a race with China, that we don't want to do anything to slow AI down. But I worry that it's sort of missing the forest for the trees, because the bigger barriers to the diffusion of AI systems in the real economy is not AI specific laws or AI specific legislation, but all the legacy legislation we have dealing with every other sector of the economy, from healthcare, finance, every area where AI is going to make an economic impact, it's running directly into 20th century laws, and that's going to be the biggest barrier and inhibitor. And so I'd almost rather a moratorium or a jubilee on all that older legacy regulation if you really cared about winning the AI race.
- [06:32]
Stephen Overlea
Well, the argument that I hear often in favor, I guess, of this idea of a moratorium is what you alluded to earlier, which is all of these state laws popping up, creating kind of a patchwork of regulation that companies have to grapple with. And I guess to sort of be the contrarian in some ways, because we've seen that same dynamic around privacy, for instance. We've seen it around kids online safety. I guess I question if the patchwork of state regulation serves kind of a point. Right. They are like the laboratories for democracy. They're testing out these different ideas to see what works before a federal standard is established. Is there an argument to be made there that annoying even if it is, that patchwork serves a purpose?
- [07:14]
Sam Hammett
Yeah, honestly, yes. My biggest issue with the moratorium is not the idea of a moratorium per se, but just its indefinite scope. Right. It's putting a moratorium on ideas that we don't even have yet. And so it would be better if Congress came forward with a broad, streamlined framework for the sort of basics, the things that there's broad consensus on, like the leading frontier AI labs should have some duty of care with the kinds of systems they put out into the wild. There should be some basic protections for whistleblowers, especially for the leading companies. Basic transparency and explainability. There are some basic things. Deepfakes is another example. One of the only laws that's passed this Congress was related to deepfakes. So there are some areas of agreement. And in the last Congress there was this big effort around the Insight forums to try to get Congress on board with some broad framework. If we had something like that that defined clear rules of the road, especially for the leading AI companies and for the most catastrophic, severe kinds of risks, then the moratorium would sit a lot more comfortably with me because it would really apply to all the sub severe risks. The things that really can jam up the works or are hard to define, like issues around bias, discrimination or ancillary reporting requirements that don't really help anyone.
- [08:33]
Stephen Overlea
Do you think there's much appetite under the Trump administration to develop those kinds of rules? I mean, obviously the White House is in the process now of thinking, thinking through what its AI frameworks and strategy will be. But there has been a pretty deregulatory bent so far to this administration. So I wonder what you sort of see for the next four years on that front.
- [08:56]
Sam Hammett
Yeah, again, AI is such a broad umbrella term where there's a sense in this administration that we want to move faster. And I'm very on board with that. I think the thing that's going to limit our adoption of AI again, our legacy regulations, our ability to build the requisite energy and transmission and data center infrastructure, and they're all in on that kind of stuff. But then there are all these other issues that pop up around AI, like child sexual exploitive materials, the ability to impersonate someone and defraud them, where it's actually in some cases been folks on the right that have led the charge against those particular uses. So it's not something that you can paint with such a broad brush, including on the ideological dimension. In fact, some of the most sort of hardcore accelerationists, if you want to use that word, have been on the Democratic side.
- [09:47]
Stephen Overlea
One, as you said, this moratorium, at least the way it's written, is quite broad in its scope. I mean, the other question that has been raised to me is like whether this is even legal given, you know, the state's rights and the ability of states to kind of decide for themselves how to oppose regulations and laws for their people. What's the likelihood that this whole moratorium idea just gets sunk by some of these constitutional or procedural questions?
- [10:19]
Sam Hammett
Oh, quite likely. I'm not an expert on interstate commerce or those elements of constitutional law, but they're trying to do this through the reconciliation process. And I think that's where it's going to hit its first major roadblock, which is that reconciliation is primarily a budgetary process. It's about taxing and spending. And while folks sometimes get creative in how they interpret their various amendments, this one seems to fail the sniff test and will probably get bird ruled, but you never know. So I think there's a bigger issue with engaging with the states and educating state lawmakers on if you are going to pass AI regulation to understand the technology in the first place. Because one of the biggest valid critiques I think that we've seen come out from some of these state laws are forms of regulation that are simply unenforceable or that don't quite understand the nature of the technology and the kind of traction that you can even have in the regulatory context. Foreign.
- [11:29]
Jack Blanchard
Hi, I'm Jack Blanchard, Playbook's managing editor and author.
- [11:33]
Stephen Overlea
And I'm Dasha Burns, Politico, White House Bureau Chief and Chief Playbook correspondent.
- [11:37]
Sam Hammett
And we are the new hosts of the Playbook podcast.
- [11:41]
Jack Blanchard
Every morning, Dasha and I dig into the top story of the day and give you some analysis and insight into what's really going on in D.C. and the world. So join us every weekday morning. Listen to the Playbook podcast wherever you get your podcasts.
- [11:55]
Stephen Overlea
Foreign Are you surprised that we're even talking about this idea of a moratorium? I mean, I'm frankly a little bit surprised. I don't think this has really come up before, to my knowledge, in, in these tech policy debates. And I obviously there's been talk of preemption, but this idea of the federal government just stepping in and saying states, you can't pass laws on this issue for the next decade. I'm, I'm frankly a little bit surprised that we're here.
- [12:24]
Sam Hammett
Yeah. On the ground in D.C. there are many just in the last year or two, many more lobbyists, for lack of a better word, advocates. Folks on the go faster side of the debate, I'm honestly on that side of the debate. I think we do need to go faster. But I also consider myself more of a techno realist where I believe the CEOs when they say this is the most transformative potential technology in history and there are all kinds of unforeseen things that could come out of this and we shouldn't try to pigeonhole this into a sort of simple, formulaic, whether partisan or ideological lens. I'm not totally surprised. There are many voices whispering in the ears of lawmakers saying we need this to be China. And I think everyone needs to take a step back and understand if China wins in this race, whatever winning means, it won't be because we had Colorado passing a deep fake law. It will be because they've built fake, phenomenally more energy than us and have pushed out the technology into sectors like healthcare, education, manufacturing, while our legacy institutions are slow to adapt.
- [13:29]
Stephen Overlea
I mean, if this does pass and granted, as we've discussed, that is A big if. What kind of precedent do you see this setting for tech policy?
- [13:39]
Sam Hammett
Well, in the background of this, there's a broader discussion about what the US's role should be internationally in promulgating and setting precedent for global AI standards. And I think there's a sense that I agree with that you can't simply do nothing. That if nature abhors a vacuum, and if the US isn't out there setting some baseline standards or some broader governance philosophy beyond just laissez faire, that that gap will be filled either by China, by the European Commission, or by other actors. And so, as we just coming out off of this golf trip, think about pushing American technology into other countries and trying to get them within our ecosystem, a big part of that should be setting some standards. And so I think it is kind of incumbent on lawmakers and the administration that if we do go forward with moratorium, it does pass legal muster, that there be some framework and it doesn't have to be prescriptive, it doesn't have to slow things down, but it should set a standard partly so that other standards don't fill its place.
- [14:46]
Stephen Overlea
I think that's such a good point, because even if you don't have this patchwork of state AI regulation, you're going to have a patchwork of global AI regulation. And that's the level on which these big companies ultimately operate. They're already contending with the EU AI act, for instance, regulations in other countries. And so the US creating a vacuum, which has been the case in other tech policy debates too, just means other countries are going to set standards that American companies then have to follow.
- [15:14]
Sam Hammett
Yeah, exactly. We're seeing this right now with the EU Code of conduct, which from people I talk to, even member countries in the EU don't, aren't quite excited about it, but it has so much momentum, they can't really stop it. And so companies like Anthropic and OpenAI are adapting their, their models and engaging with the European process to conform with that code of conduct. And that goes to show there is some truth to the first mover advantage in setting these standards.
- [15:44]
Stephen Overlea
As you've been talking, obviously, I sense you're trying to thread this nuanced argument here where you need light regulation, but you still need some regulation that we need to move fast, but the wheels can't come off. I mean, what do you actually want to see then from Congress or from the Trump administration when it comes to regulating AI?
- [16:06]
Sam Hammett
Well, if we start with the states, I think there's valid concern about states regulating say anything to do with the Internet really, not just AI because Internet is cross border, but when you have these large companies located in your state, I think it's completely reasonable to, for them to have things like basic, know your customer rules, basic consumer product safety kind of standards, potentially liability because that's literally within, within their, their borders. And so that's the biggest thing that I worry about with this moratorium is it's going to sort of cut the knees off of, of a bigger policy debate when it comes to the, the federal response and the Trump administration. You know, for the most part, I think they've been doing a decent job. So I think going forward, it comes back down to actually engaging in the global governance discussions. They can't just sit on the laurels because this is a technology that's going to affect every aspect of the economy of American life. And if they go all in on doing nothing at all, paradoxically, I think they'll live to regret it.
- [17:15]
Stephen Overlea
Well, listen, Sam, appreciate you being here on Politico Tech.
- [17:18]
Sam Hammett
Oh, thank you for having me.
- [17:23]
Stephen Overlea
That's all for today's Politico Tech. If you enjoy Politico Tech, please subscribe and recommend it to a friend or colleague. And for more tech news, subscribe to our newsletters, Digital Future Daily and Morning Tech. Music in our show comes from the mysterious Breakmaster Cylinder. Our producer is Normal Malaikul. I'm Stephen Overle. See you back here next week.