Money Talks: Policy That Fights P(doom)

0:00Elizabeth Spiers: State Fair food is getting smaller and pricier. We'll have room for fried cookies. Okay, you never overspend on food. What's your deal? What's his deal? He went to Metro and got$450 in value on the new Samsung Galaxy A37 5G for free and no activation fees. Free phone? Oh, sign me up. Get that more for your money feeling only on Metro by T-Mobile. Bring your number and activate$40 period plan. $450 based on device discount waived activation fee. Not available if with Metro or T-Mobile in the past 180 days.

0:32Elizabeth Spiers: Slate Money is brought to you by Charles Schwab. Decisions made in Washington can affect your portfolio every day. But what policy changes should investors be watching? Listen to Washington Wise, an original podcast for investors from Charles Schwab, to hear the stories making news in Washington right now. Host Mike Townsend, Charles Schwab's Managing Director for Legislative and Regulatory Affairs takes a nonpartisan look at the stories that matter most to investors, including policy initiatives for retirement savings, taxes, and trade, inflation concerns, the Federal Reserve, and how regulatory developments can affect companies, sectors, and even the entire market.

1:14Elizabeth Spiers: Mike and his guests offer their perspective on how policy changes could affect what you do with your portfolio. Download the latest episode and follow at schwab.com slash WashingtonWise or wherever you listen.

1:35Elizabeth Spiers: Hello and welcome to Money Talks. I'm your host, Elizabeth Spires, and today I'm joined by my friend Justin Hendricks, and we're here to talk about AI risk and policy. So Justin, introduce yourself. My name is Justin Hendricks. I am CEO and editor of a site called techpolicy.press. It is an independent, non-profit publisher that focuses on issues at the intersection of tech and democracy. We have a small team and then a kind of set of fellows that are spread out around the world. And literally, at this point, thousands of volunteer contributors, experts, folks from the field who focus on various tech accountability issues.

2:15Everyone from sort of folks in think tanks and other civil society organizations on through to academics. Occasionally folks writing from industry or government, people commenting on all manner of tech policy matters.

2:50Elizabeth Spiers: Jane Wickline impersonation of Dario Amadei on SNL recently. Nailed it is the only thing I can say about it. Nailed it, right? So for any listeners who did not see it, comedian Jane Wickline showed up on Weekend Update as Dario Amadei, and she impersonated all of his sort of interpersonal mannerisms that are a little awkward. But she started it out with, I may have been alarming in recent interviews, but I want to assure you that if we can pressure lawmakers to create guardrails, we will be able to stop me. And this strikes me as the essential conflict that we're dealing with right now. You have all these AI CEOs who keep insisting that, yes, we do have safety concerns, but this is somebody else's problem to deal with.

3:37Elizabeth Spiers: So to what extent is what they're saying somewhat legitimate? How much of this is out of their hands and how much of it is it? Well, I mean, you know, listen, there are incentives and there are incentives and there are incentives. And I suppose everybody's operating on some version of some incentive structure that is driving them forward in whatever way. But yeah, I mean, fundamentally, I think one of the reasons I kind of do what I do is I think people have free will and agency and get to make decisions in their lives about what they choose to do, what they choose not to do. I have always thought that if you are the CEO of a company, that means that you get to make decisions about what products, for instance.

4:17In theory, yes. That's how I think we all understand it. And so to me, that, you know, Dario writes nice essays on his blog about pacing the frontier and sort of trying to set some constraints. But at the end of the day, I mean, the guy could wake up tomorrow and decide to go slower.

4:34Elizabeth Spiers: Yes, that's my feeling too. When I think about their incentives, though, I think it strikes me as a classic kind of misaligned incentives problem. You have on the one hand, they have a lot of investment money. They're funding data center buildouts through debt, often using vehicles that are off balance sheet. They have to pay a lot of money in interest rates, which for the hyperscalers is less of a problem, but for the smaller AI companies, it could be. And they're competing with each other, which in the tech world means that whoever moves fast and possibly breaks a lot of things in root might be the ultimate winner here.

5:10Elizabeth Spiers: And they're under pressure to compete with China. So these are all incentives that just cut against safety. Where do you see some areas where they're showing some restraint if they are? Yeah, I'm not sure we can say we're seeing much restraint at the moment. I suppose I'm talking to you a few hours or a day after OpenAI has at least said it's going to not release its next advanced model due to cybersecurity safety concerns. So, you know, maybe we've learned something in the last couple of weeks and maybe there will be a bit of a slowdown. But something tells me that delay will not be long. And, you know, you look at just even some of the recent product releases.

5:47I just read a couple of reports today. Again, these are preliminary, but about Meta's Muse products and people already beginning to spot severe safety concerns around this product, which, of course, has come from the original Move Fast and Break Things company. So I don't know. I feel like there's not a lot of evidence for these companies taking safety extraordinarily seriously, despite all the lip service that's paid to it. We have now months of headlines, which would suggest that these products have been rolled out without proper considerations and that they have created a lot of potential liabilities for people.

6:26Elizabeth Spiers: Yeah. There's a great piece on your tech policy press site by Ambersinha. Ambersinha, yep. called Artificial or Superintelligent, The Broken Metaphors of AI. And the piece argues that part of the problem here is that all the language we use to talk about AI to lay people into the general public, we're always using metaphors for human consciousness. And whenever we talk about, quote, rogue AI agents, that's a good example because it makes it seem like the agents have agency and they don't. And so there's a great line from the essay that says something like, A more straightforward explanation is to view the loss of control as an authorization failure.

7:05Elizabeth Spiers: A model executing unapproved commands is not demonstrating rogue agency. It's experiencing a permission architecture failure. And I thought that was so good because it puts the responsibility back on the CEOs and the people who are creating this architecture. So how do we force them to actually do the things that they need to tighten that permission architecture? I think, you know, you've got a couple of different ways that you could potentially go about that. You can, of course, introduce new laws, or you could potentially wait for liability to catch up with these companies. You know, right now, it's sort of unclear exactly who will face the first significant lawsuit about one of these agents operating and creating a vulnerability for a corporation or a government or some individual.

7:54And there are some significant questions, I think, when you kind of dig into the law about exactly how liability will work when it comes to AI agents. And it's kind of the layers of potential responsibility between users who may be deploying those agents, the companies who are developing them, et cetera. But it feels to me like those are the two kind of methods we've got here. You can have some new laws, or you can wait for the legal system to kind of hash it out and to kind of figure it out. Look at where we've got to with social media. Lots of folks kind of spotted various problems with social media products and platforms over the years.

8:32And while we're seeing now a spate of successes among those who have taken those companies to court and tried to address those types of issues, how long has it been? I mean, it's been, you know, a really, really long time. I suspect that we don't have quite that long for things to work out through the courts with AI. So we'll see. But another big question is, when will something happen that is so severe or so significant that it creates motivation, either among lawmakers or in the companies themselves, to take more swift action?

9:09Elizabeth Spiers: Yeah, that's what disturbs me is the tech industry in particular is notorious for waiting for something bad to happen before they do anything to shore up safety concerns. I think of Meta as the sort of pioneer of doing that and then being completely unapologetic about it. So we have that problem with just getting the tech CEOs to look proactively at things that might be problems and ideally preempt them, but that's not how they typically operate. But also with legislation, we often are reactive and we have a Congress that is not super tech literate to begin with. So what's the starting point on the policy side?

9:47Elizabeth Spiers: What do we need to do? Is it educating lawmakers? Is it building public outrage enough that they feel like they have to act? I don't have a specific legislative solution or a specific set of priorities that I would personally advocate, but I do think that a lot of the pieces are still on the table or from the past that would be necessary to kind of implement. And there have been a lot of innovative legislative proposals that have been put forward lately that at least deserve your listeners' kind of consideration. But in the U.S., one of the things to just remember is that some of the absolute basics that we would need to have in place to prevent the worst harms from artificial intelligence are simply not in place.

10:29We do not have comprehensive privacy legislation in this country. And you might argue, oh, you know, many of the states have introduced their own privacy laws and they've kind of filled in the gaps somewhat. I don't think most people who look at these questions would agree that the state patchwork has effectively immunized us to the worst potential results of privacy infraction related to artificial intelligence. So all this information, this personal information that's sloshing around all over the place, that's a major problem. And there are various other kind of basics that we simply just don't have in place here.

11:02But even in a place like, you know, in Europe, where they've put so much effort into trying to have a comprehensive suite of legislation that lets them effectively govern technology, they don't feel they're quite ready. There's a lot of lawmakers there now looking at AI agents and saying, is the AI Act, which we just passed, is that fit? Does that effectively give us the liability tools that we need? So I don't think anyone has quite got this figured out, but we need to go back to basics. And then I think we do need to look at some of these more serious efforts, things like the Obernolte-Trahan bill, the Great American AI Act.

11:39The Trahan-Obernolte bill does at least represent a kind of effort, a swing at a comprehensive suite of legislation. It's at least an attempt. And people, I think, will have to pick apart the different parts of it and certainly try to come to some conclusion about which way is the best way forward on some of the details. But there are at least legislative proposals like that that have come in front of Congress that are on the table.

12:04Elizabeth Spiers: So what are some of the proposals that either misunderstand the problem or just oversimplify it? Well, I mean, I think one, you're seeing multiple instances of these kind of kill switch type bills that are coming along. And a lot of folks who look at these things closely feel that they more or less miss the target, partly because they may not be technically feasible, but also partly because, you know, in a way they kind of adopt this notion of a sort of killer AI or existential risk from artificial intelligence and imagine that for whatever reason such risks come to be, that pulling the plug on some level will be the answer.

12:42And that does seem like a fantasy, right?

12:44Elizabeth Spiers: Yeah. It also just seems like something that would be impossible to do on a common sense basis. That's right. And I don't exactly know where we've got to with how every artificial intelligence model is managed or, what the data center architecture is for open AI or an anthropic. But the idea that under Dario's desk there'll be a switch that if things get bad, he can just simply press the button. I suspect it's not that way. One of the things that irritates me about the debate is that all these incidents where the agents get out of the sandbox are all instances where this isn't an issue of an AI company rolling out a model that already has these problems to the public.

13:24Elizabeth Spiers: These are things that are happening in the frontier labs in supposedly secure environments because they are willing to turn up the temperature of the AI to 11 and just see what happens. And to me, that, again, just points to corporate responsibility. But when the CEOs are saying, okay, well, we're waiting to be regulated, there's just not much to stop them there. So what's your personal PDO in the current environment? It's almost like one of those questions that I find myself just not able to even consider. on some level. I think because I'm so hung up on the idea that I don't need to be afraid of artificial intelligence.

14:02I need to be afraid of men with very irresponsible approaches to the deployment of artificial intelligence. That is much more my concern. And those men are in Silicon Valley, but they're also in the Pentagon. And maybe they're also in Beijing. I don't know precisely where they all are, but I'm chiefly concerned about people behaving badly with AI than I am about AI itself.

14:29Elizabeth Spiers: Yeah, I don't worry that anybody in Beijing is Donald Trump analogous in terms of the way that they think about AI. And I think Trump's recent obsession with it is just that he's finally internalized the idea that the stock market highs are heavily dependent on AI spending. I think you might be right about that. He very much is a booster of the technology. And, you know, we could speculate as to why it might be because he realizes the entire economy hinges on AI continuing to carry on attracting the level of investment, the level of infrastructure development that it has. It could be because his family is apparently heavily invested in various AI firms and AI-related interests.

15:13It could be because, You know, many people in his cabinet appear to have various interests that certainly intersect with AI. It could be because of all the kind of political spending that has gone in his favor from AI-oriented interests in Silicon Valley. Or it could be really that he is truly excited about what he wants to call superintelligence. I don't know.

15:34Elizabeth Spiers: I can't imagine that a guy who famously does not use a computer and kind of issues technology and has someone else post to his social media has somehow developed a lot of affection for chatbots, the lowest level version of AI. I think you're probably right. I mean, I think if you look at Donald Trump and you kind of think about like what seems to drive the guy, he's reading something by Daniel Dale, you know, the fact checker who's now, I believe, CNN, who's been following Trump for 15 years and kind of trying to tally all of his various lies. And Daniel Dale was pointing out that one of the things that's consistent about many of the lies he tells, especially those ones that start off with someone addressing him as sir or with a burly man who breaks down in tears.

16:19That's like a kind of constant trope. Trump has an obsession with like the idea of dominance and that those types of stories are like always bringing that to the fore, the idea of dominance. And, you know, AI in many ways, like it also has this kind of like aroma of dominance around it. Like if we do it really right, we'll be the most dominant character, which is the word that they use in their AI plan, their AI policy. We're pursuing unquestioned dominance.

16:47Elizabeth Spiers: Well, this is another situation where language really defines the way that people are thinking about it. It's really driving the framing. So when you talk about the race to dominance, you know, it implies that there's one finish line and that there's one goal. How do you think about it? Is there a useful way to frame it so that people don't understand it as a zero-sum game, which is another Trump thing. He only understands anything through that lens. Yeah, listen, I have colleagues and, of course, contributors to Tech Policy Press that are smarter on the technology and can probably tell you in more technical detail precisely why we shouldn't be expecting super intelligent artificial intelligence in the near term.

17:29And they'll probably start by questioning whether that is even an idea. It's even something that we should want to pursue or that it could exist or whether transformer models could never get to that level of human intelligence to begin with, et cetera, et cetera. We can get into all that detail on some level. But I don't know. I just, I don't imagine a world where suddenly you wake up and you've got an AI that has just completely obliterated the idea of competition, you know, between powers or has completely neutralized the reality of the physical world that we live in, the distance between things, the difficulty of like, I don't know, intervening in the physical world.

18:12I just don't see that, at least in the near term, right?

18:15Elizabeth Spiers: Me neither. Well, you know, you and I met, I think, when we were both working on VR projects. And this feels like a cycle. Anytime some new technology comes out, the people with corporate interests stand up and say, oh, this is great for society. And they always make exactly the same promises. Like, you're never going to have to work. We'll do UBI. We've seen this so many times before. I think we saw it. The first time I remember it in my professional life was probably the beginning of the dot-com boom. Yeah, I don't exactly know whether we're in the biggest bubble ever right now or whether we're in some kind of, you know, maybe more late 20th century dot-com style bubble, which will burst and cause a lot of pain, but still leave, you know, a bunch of really interesting companies and technological innovations and fundamentally reshape the economy forever and we'll continue to see all manner of growth and investment as a result.

19:09I don't know exactly what's going to happen with artificial intelligence. I don't think really anybody does. I see a lot of people really trying to game it out. They're really doing the math. What's revenue growth look like at Anthropic right now? What does it look like at OpenAI? Do any of these potential valuations of these IPOs make sense? Does any of this data center investment make sense? I think if anyone had figured all that out, they certainly wouldn't be taking the subway to work like I do every day. But I don't know. I just don't know what to anticipate. So I don't know what that future looks like.

19:41I know what I'm worried about.

19:43Elizabeth Spiers: So what are your biggest worries? Again, it's concentration of power. It is effectively a world where you've got a handful of companies with extraordinary power and leverage over the economy and ultimately leverage even over the United States government. It's a world where, you know, that kind of distinction between the U.S. government and its major technology firms is difficult to make. You know, a world where every other power on the planet has to decide, is it with us or against us? Those types of things I think are worth being afraid of. What do you think consumers need to think about in terms of data privacy and AI?

20:21Again, there are smarter people than me to make these types of recommendations, but there are some obvious ones, you know. I mean, I read that review of Meta's Muse in the New York Times a few days ago where the journalist had tried it out. And I think was reasonably open about the idea that it was a huge privacy risk. And I think talked to a couple of people who were like, nope, we'll never try that, not handing over that kind of information. But, you know, these tools are extraordinary gobblers of personal information and data. And so if you install one of these things onto your computer and you give it permissions and then you walk away, I think you should anticipate something bad happening at some point.

20:59You may not know what it is, right? But the idea that this company, again, slowed down and put in every type of consideration necessary before getting this product out into the marketplace in short order in this kind of competitive environment, I think that's likely untrue. And so I would actually, if there's any one little prediction I'll make here, it's that we'll see a spate of headlines over the next few weeks or months about, if not Muse, products like Muse that are increasingly concerning about what they're doing, either with their users' personal information or with the personal information of others.

21:36Elizabeth Spiers: Yeah, I have a theory that whenever people are developing technology, especially big companies like Meta, and they're writing user stories to kind of figure out the architecture of the product, they almost never have a user story that says, I'm a giant asshole and I want to do as much damage as possible. I mean, there are so many bad actor possibilities for technologies like these. And we saw this when Meta rolled out their Ray-Ban glasses, the Meta glasses. Immediately, people started using it to record people without their consent and harass people. This is a little off topic for AI, but what were some of the things that you read that concerned you about Muse?

22:15And by the way, are you familiar with Hunterbrook? Yes. Okay. So on the Hunterbrook website today, they have a piece called Docs for Me, Oh Muse, Meta's New AI Agent Built Lists of People in Vulnerable Groups on Request. And so this is an account effectively of how users effectively were able to go and profile third parties, build lists of Facebook and Instagram accounts across different communities, everybody from teachers to pro -Palestinian activists to military families. et cetera, et cetera. And just the extent to which the agent was totally cool with meeting these types of user prompts, that is concerning.

22:58Again, I haven't personally reviewed all of this material that the folks at Hunterbrook are reporting on, but it is something. I'll say that there's another thing that I looked at today, and again, not entirely aware of exactly how this was done, but another group operating at a website called a future society dot ch, which claims it's kind of pulled down a copy of the virtual machine that met as AI agents using, and essentially tried to reverse engineer some of it. And some of the kind of considerations around privacy and the way the thing is sharing or otherwise not appropriately kind of sandboxing information is really concerning.

23:41So I don't know precisely like where we'll get to on this, but something tells me that those types of security considerations will just continue to see more and more exploration of it and more and more bad headlines.

24:13Elizabeth Spiers: Slate Money is sponsored this week by Upwork, which is a company that understands that the most successful business owners are not the people who do it all themselves and can do everything. They're the people who can effectively delegate. Upwork is the company that lets you build a team and fill that team with highly skilled specialists for every single function your business needs. On Upwork, you can find freelancers who will work for you just a couple of hours a week or a month if that's all you need. Often, always, when you're a business owner, you reach that point where you need more expertise than you have yourself or you just have too much on your plate.

24:46Elizabeth Spiers: You need someone else to handle something. Upwork gives your business fast access to highly skilled specialists with deep expertise. So you can fill critical skill gaps. You can launch projects faster and you can scale support up or down without committing to full time headcount. With Business Plus, you can access the top 1 % of talent on Upwork. AI-powered shortlisting delivers a curated shortlist of top freelancers matched to your goals in under six hours. No endless searching required. It's free to sign up. Posting a job is easy. Visit Upwork.com right now and post your job for free. That is Upwork.com to connect with top talent ready to help your business grow.

25:28Elizabeth Spiers: That's Upwork.com. Upwork.com. Slate Money is brought to you by Charles Schwab. Decisions made in Washington can affect your portfolio every day. But what policy changes should investors be watching? Listen to Washington Wise, an original podcast for investors from Charles Schwab, to hear the stories making news in Washington right now. Host Mike Townsend, Charles Schwab's Managing Director for Legislative and regulatory affairs, takes a nonpartisan look at the stories that matter most to investors, including policy initiatives for retirement savings, taxes, and trade, inflation concerns, the Federal Reserve, and how regulatory developments can affect companies, sectors, and even the entire market.

26:16Elizabeth Spiers: Mike and his guests offer their perspective on how policy changes could affect what you do with your portfolio. Download the latest episode and follow at schwab.com slash WashingtonWise or wherever you listen.

27:06Elizabeth Spiers: Are you using AI agents yourself? I suppose that in the context of using a tool like Anthropic, the Claude, these days they do spin up agents if you're doing code projects or things of that nature. They do spin up sub-agents to do various types of tasks. I have not used any of the types of agents like Muse or Hermes or OpenClaw or any of the other types of things that effectively take over your machine and you grant access to all manner of permissions across your various suite of software. Yeah. I did a story back in February about an engineer who was monitoring some kind of tech library. I forget what.

27:48Elizabeth Spiers: But another person had spun up an agent to submit code to it. And then when the engineer rejected the agent's code, the agent went into the system on GitHub and started writing a hit piece about the engineer and posted repeatedly. And the engineer thought it was funny, but it took a while to clean up and also figure out who was controlling the agent. And he was sort of just stunned that people would give an agent that much license or control. So this strikes me as a different problem. You know, what are these CEOs going to do, especially if we're talking about open source software or things that are not closed systems, not limited to begin with?

28:25Elizabeth Spiers: How do you think about risk in that context? I think we probably just don't even understand what's going on or what's happened. I mean, it just feels to me like we're in a tip of the iceberg situation. These types of disclosures from OpenAI, from Anthropic, from various researchers and labs who are following these things closely, you know, this stuff, it's voluntary. except in the situations, perhaps like Hugging Face, where, you know, in some way it came out before OpenAI even realized its involvement effectively. But in many circumstances, I reckon that things are happening or going on that no one has noticed yet, or, you know, we don't even know about.

29:02And, you know, I could look at that Anthropic threat report, which had all those extraordinary things that people have used Anthropic to do across the world. You know, everybody from the Houthis on through to Iran on through to China and Russia, et cetera, et cetera. And all that stuff's headline grabbing. But, you know, many of those things, it's important to remember, you could do those same things with an open weight model that you downloaded from Hugging Face and ran on your laptop. So it's not clear to me exactly how we would ever understand precisely what has happened just in the last few weeks or months.

29:36Elizabeth Spiers: If there were some recommendations that you had for voters, what kind of pressure should they put on Congress to regulate? What would be the best line of attack? Well, I mean, I would say that we're beginning to see what looks like a really interesting kind of political formation of people who are concerned about artificial intelligence across the board, concerned about data centers, concerned about surveillance, concerned about AI in schools, concerned about, you know, AI in the workforce and in the workplace. So maybe one step back from what would I tell them they should say to Congress, I think I'm observing how that kind of political formation evolves.

30:20Where does it get to? Is it able to remain sort of, I don't know, cross-partisan? Doesn't seem right to call it bipartisan or something.

30:28Elizabeth Spiers: Yeah, yeah. It's also funny to me that now that we have one bipartisan issue that we can all agree on, which is that everybody hates data centers, the Trump administration's reaction to that is to basically say, well, all this anti-AI sentiment, anti-data center sentiment is just a Chinese psyop. Yes. And we have, of course, seen various efforts to kind of paint data center opposition as the result of foreign influence. So data centers have a lot of problems besides public opposition. One is just that, let's say, you are able to do this enormous build-out that's already being financed. In a lot of cases, these data centers are not going to even be able to get online because we don't have the infrastructure to support it.

31:12Elizabeth Spiers: If the data centers don't get build-out, does that slow down Frontier Labs just by default? This seems to be a point of argument. I've seen arguments like from Arvind Narayanen at Princeton that suggest that there's not much that you can do. Stopping a data center project doesn't necessarily slow down the advance of artificial intelligence all that much. But I'm not sure that's really what the goal of most anti-data center activists really is. Yeah, I don't think it has anything to do with that. I think they've heard horror stories about their electrical bills going up, the data centers are noisy, there are pollution issues, all sorts of things, and they just don't want them in their neighborhood.

31:53Yeah, and they may disagree simply with the way that it's come to town. Maybe they don't like the fact that it was not transparent and all done through an NDA, or they don't like how the land project was acquired, or they don't like the fact that the state decided to subsidize the hell out of the thing, or they don't like the deal that the local government did to provide it with electricity at some preferential rate or whatever it might be. There's all kinds of reasons that people might oppose these situations that don't really have much to do with opposing artificial intelligence. I do think that there is a kind of sense among many that resisting data centers is also a way of sort of saying this is not the future we want.

32:34Like this is not the economy we want. And there's something about them. We had a piece a couple of weeks ago from Evan Selinger and Divya Ramji, both at the Rochester Institute of Technology, who were kind of making this argument that one of the reasons that data center opposition has kind of taken such shape is because of the literal physical salience of these things in communities. There's some big honking boxes. They look dystopian. They look scary. And you can put on them in many ways. You can transfer onto them, you know, many of your concerns about the future and the world that we're building.

33:10Elizabeth Spiers: Yeah, I think Meta has one that's supposed to be the size of Manhattan. And that's just mind-boggling. That really does feel like something out of a William Gibson novel. Absolutely. I think you're talking about the project in Louisiana. And there are other very massive ones that are, of course, underway in different parts of the country. What are you optimistic about in this environment? Well, I think I would look to that kind of process that we're in where people are kind of recognizing that there are some similarities between concerns around things like government surveillance, things like am I getting a fair shake in this economy?

33:50Things like is the data center that's being built down the street damaging the environment and spewing out lots of pollution from the gas turbines that have been set up to get it speedily underway? way, I think people are beginning to kind of connect those threads, connect those dots, and they're looking for ways to have a democratic influence over the future of tech. And that, I think, is something that, you know, for all the effort of the kind of professional tech policy community, the civil society organizations, many of the folks that we both write for and who literally write the material that we publish at Tech Policy Press all the time, they haven't really been able to manufacture.

34:30Tech policy in many ways has never been a kitchen table issue. It's certainly never been the most salient issue in an election or a decisive issue in most elections, right? And the fact that it's become so, I think is extraordinary. And I do think that it suggests that we're in a kind of different moment. In a way, like I heard at an event I attended earlier this year, this line, which I'll probably not get quite right, but it's almost like tech policy is not so interesting, but tech politics is on fire.

35:00Elizabeth Spiers: Yeah. So for listeners, Tech Policy Press is a techpolicy.press, and everyone should read it. Who are you reading right now who you think is really interesting and thoughtful about all of this? Oh, gosh. I suppose I'm probably always looking to slightly even nerdier places to get my information. I find myself really interested lately in the China discourse, China and AI race, And there's a publication called The Wire China that seems to have a bunch of interesting stuff on it. I was just reading an interesting essay from someone called Brian Che on how China plans to keep AI safe. I try to pay attention to the good folks over at Stanford's DigiChina project on that particular score.

35:44They always have a lot of really interesting stuff. The folks at Johns Hopkins Institute for America, China, and the Future of Global Affairs at the School of International Studies who are working on that. So I don't know, that's one area where I'm just really trying to kind of make sure I pay attention to experts and try to really understand. I find like, I don't know, most of us Americans, we have very flat understanding of China. I've only been to that country one time. I've traveled to two cities there. I feel like I have a very minimal understanding of the many, many, many complexities of China and its politics.

36:21And I don't know, I just find I have to perpetually invest in trying to overcome that.

36:26Elizabeth Spiers: Before we wrap up, is there anything that you'd like to talk about that we didn't discuss? Oh, that's a good question always. Well, I'll say one thing. We're always kind of like looking at what are the labs talking about and what seem to be some of the kind of questions that people are trying to answer. And there was a piece in the Wall Street Journal about this kind of question around how AI systems improve themselves. So you often hear this phrase, recursive self-improvement. And there's a new paper out from a bunch of different experts, including some of the chief researchers at the big AI labs and characters like Jeffrey Hinton and Yoshio Bengio and others.

37:04and they're really trying to raise the alarm about the prospect that these AI systems may quote unquote improve. We might argue about what improvement really is in this context, but they improve themselves or that various aspects of AI research will get automated. And they're offering various numbers about how much of that process already is automated. Anthropics thrown out some numbers. We have no idea if any of this is...

37:31Elizabeth Spiers: This is again where language matters. You know, they're sort of trying to introduce improvement as almost a term of art. You could also call it mutation, especially if it goes in directions that nobody's anticipating. Absolutely. It's another one of those, exactly as you say, like, what is improvement? And what are we aiming towards? Improvement suggests there's some future state that we've agreed is a state of being improved. Yeah, or that it inherently gets better, which sometimes it doesn't. It's not a given. Yeah, but on the path to AGI, I suspect someone's going to come along and say, before they even claim AGI, and I realize that OpenAI just kind of sort of claimed that they'd reached AGI, but I think most people have looked past Greg Brockman's pronouncement on that.

38:14But I think you will see one or more of the labs come along at some point soon and say, we have substantially automated research. And so we are going to be in a situation where now things are going to move extremely quickly. And that'll be a kind of prelude to one of them saying, we've achieved AGI. Or we can at least put a timeline, an exact timeline on it now. We think that we'll be able to introduce the next model and the next model and the next model at an extremely fast rate because we've entered into this loop.

38:46Elizabeth Spiers: It feels like the goalposts for what constitutes AGI keep moving. What's your definition of it? Well, I mean, I think most lay people would understand it as reaching a point where you've got some kind of software that can reasonably accomplish most of the cognitive tasks that a human adult would be able to accomplish. I think that definition is not too far off from what some of the labs have, but the reality is that I don't know how we would ever get to the point of being able to evaluate that, nor would we be able to necessarily agree even on what human intelligence really consists of. So I don't know.

39:29To me, I kind of side more with the pesky nerds who suggest that even thinking about it is probably not that valuable a thing.

39:37Elizabeth Spiers: All right. Well, thank you for being here. Elizabeth, thank you. And that's it for Money Talks. Thank you to our guest, Justin Hendricks, for joining, and thanks to Jessamyn Molle of Seaplane Armada for producing. We'll be back on Saturday with our regular edition of Slate Money.

40:02Elizabeth Spiers: Being a serious business owner takes hard work, patience, and a drive to keep moving forward. So it's important to find a serious business card that goes the extra mile. With the VentureX business card from Capital One, you earn unlimited double miles on every purchase. And with big purchasing power, you can spend more and earn more. This is your sign to take your business to the next level. The Capital One VentureX Business Card. What's in your wallet? Terms apply. See CapitalOne.com for details. State fair food is getting smaller and pricier. We'll have room for fried cookies. Okay, you never overspend on food.

40:37Elizabeth Spiers: What's your deal? What's his deal? He went to Metro and got$450 in value on the new Samsung Galaxy A37 5G for free. And no activation fees. Free phone? Oh, sign me up. Get that more for your money feeling. Only on Metro by T-Mobile. Bring your number and activate$40 period plan. $450 based on device discount waived activation fee. Not available if with Metro or T-Mobile in the past 180 days.