The AI Daily Brief: Artificial Intelligence News and Analysis - The State of AI Regulation in the USA

Episode Date: October 30, 2023

On the eve of the Biden Administration's anticipated executive order on artificial intelligence, NLW surveys the landscape of AI legislative efforts in the US. ABOUT THE AI BREAKDOWN The AI Breakdown helps you understand the most important news and discussions in AI.  Subscribe to The AI Breakdown newsletter: https://theaibreakdown.beehiiv.com/subscribe Subscribe to The AI Breakdown on YouTube: https://www.youtube.com/@TheAIBreakdown Join the community: bit.ly/aibreakdown Learn more: http://breakdown.network/

Today on the AI breakdown, we're looking at the landscape of all the different legislative efforts in the U.S. around AI regulation in advance of tomorrow's anticipated executive order on AI from the White House. The AI breakdown is a daily podcast and video about the most important news and discussions in AI. Go to Breakdown. Not Network for more information about our YouTube channel, our Discord, and our newsletter. Welcome back to the AI breakdown.

Last week, we got news that the White House would be unveiling its long-anticipated executive order on artificial intelligence this Monday. Now, the timing of this announcement is obviously meant to coincide with the AI Safety Summit that's being held in the UK on Wednesday and Thursday of next week. Last week, I did a little bit of the preview of what is expected to be in this EO. The super TLDR version of it is that in many ways, this is the U.S. using its governmental purchasing power as a way to start applying different tests and licensing to the major AI labs. In other words, when the government goes to buy AI services, they will have to be tested in certain ways, red-teamed from certain external sources from the government themselves, in ways that will

obviously, A, give us more information about the safety of these models, and B, which potentially set precedence for the private sector as well. But obviously, an executive order from the White House, as impactful as it might be, is just one small piece of the regulatory tapestry and what is shaping up to be a fairly significant competition around AI policy in the United States. Let's start first today with a look at the actions of Senate Majority Leader Chuck Schumer. He is one of the politicians that has come out most aggressively trying to put himself at the center of the AI conversation, and part of the way that he has done that is through a series of what he's calling AI Insight forums. Now, these are private, closed-door events that bring in

participants from outside Congress and the Beltway in general in order to help members of Congress and the Senate learn about these important issues before they figure out exactly how to regulate them. Schumer has strongly argued that while it's important to get AI regulation on the books quickly, doing it right matters even more. And so the Congress educating itself about these issues is an important intermediate step. The first of these insight forums featured the participation of a murderer's row of AI Lab CEOs, including basically everyone you'd expect, the Zuckerbergs, the Sam Altman's, even Elon Musk. Additionally, it included a number of civil society leaders, labor representation, and other folks outside the tech space as well. The topics for

these off-the-record conversations include AI innovation, copyright and IP, use cases and risk management, national security, guarding against doomsday scenarios, transparency, explainability, and alignment, privacy, and liability, and more. there has been a lot of chagrin about the fact that these are being held behind closed doors, and I do understand that in the sense that we want transparency around what people are saying and how they're lobbying our government around such an important area of policy. The flip side is, if you've ever watched any sort of public hearing, they are not events that are actually optimized for learning anything. They are events that are optimized for politics.

I think having a balance where a diverse set of people get to share their perspectives, and politicians get to actually listen without thinking about how they have to have a talking point after is probably a useful part of this tapestry as well. Now, the second of these forums was held earlier this week with a focus on innovation. The participants included very notably Mark Andresen fresh off of his techno-optimist manifesto, as well as AOL founder Steve Case, Kleiner Perkins chairman, John Doer, and a number of different startup CEOs including Aidan Gomez from Cohere and Patrick Collison from Stripe. representing perspectives outside of tech included Max Tegmark, the president of the Future of Life Institute, and one of the folks who has been most involved organizing, things like the six-month pause letter,

Amanda Ballantine, the AFL-CIO Technology Institute director, was there, as was Derek Johnson, the president and CEO of the NAACP. Now, because this was a closed-door session, we have to rely on very limited reporting to figure out what was going on. The tech policy press has the best summary that I've seen, and so I'm primarily using them as a representation. They noted that, quote, while more than two-thirds of the Senate attended last month's closed-door hearing, the Senate forum was met with much less fanfare. Some senators seen shuffling into the hearing, according to a next star reporter present on the Hill, included Senators Elizabeth Warren, J.D. Vance, and John Hickenlooper.

Now, Tech Policy Press also points out that on the same day that this was happening, two different AI bills were introduced, but we're actually going to talk a little bit more about specific legislation in just a moment. So let's move on to what the discussions actually were at this particular event. The topics discussed, included transformational innovation that pushes the boundaries of medicine, energy, and science, government research and development funding that incentivizes equitable and responsible AI innovation, open source AI models, i.e. balancing national security concerns while recognizing this existing market could be an opportunity for American innovation, making government data sets

available to researchers, and minimizing harms, including job loss, racial and gender biases, and economic displacement. According to attendees that these reporters spoke with after the hearing, the tone of the forum was described as encouraging, thoughtful, civil, and optimistic. It sounded like there was a lot of agreement on the very, very fundamentals, like government having a role in AI regulation. But they also wrote, quote, the greatest point of tension seems to be around the level of AI regulation different stakeholders see as reasonable. This comes as little surprise, given that billionaire attendee Mark Andreessen published a techno-optimist manifesto the week prior that appeared to equate AI regulation with murder. And although tension around the right

path forward on regulation was very visible in the room, as Center for Democracy and Technology President Alexander Reeve Givens observed, it wasn't, quote, in a manner that felt completely insurmountable. Apparently during the hearing, Senator Schumer suggested that the next forums will focus on issues including AI harms and solutions. Now, in the meantime, as Chuck Schumer is hosting these sessions, there are a variety of legislative proposals out there, and they fall broadly into two categories. First, our comprehensive legislative frameworks that are meant to address a broad swath of issues around AI regulation. One of the most notable of those efforts is a framework from Senators Richard Blumenthal, a Democrat from Connecticut, and Republican Senator Josh

Holly that was introduced at the beginning of September. That legislative proposal was actually introduced the same week that the first Chuck Schumer AI Insight Forum was held in Washington, D.C. One of the subtext, actually, and not that subtext, depending on some of the comments that you could read from Josh Holly, was that government shouldn't necessarily just be giving big tech a bigger microphone, and should instead be making more decisions about actually regulating the industry. There were a few key pillars in that framework that are worth noting, given that they might find their way into other future efforts as well. The first was to establish a licensing regime administered by an independent oversight body. Basically, they said that models as advanced

as GPT4 or more advanced would have to register with this oversight body, which would then have the authority to audit those companies and work with other enforcers as well. The Holly Blumenthal framework would ensure legal accountability for harms. One of the big things here is that they want Congress to clarify that Section 230, which is, of course, the famous provision in the 1996 Internet Decency Act that says that platforms are not responsible for what their members do on those platforms shouldn't apply to AI, and that victims of harms perpetrated by AI can take the companies behind those models to court. Other parts of the framework, including defending national security and international cooperation, promoting transparency, and protecting consumers

and kids. Now, that wasn't the only comprehensive legislation or team of senators putting together comprehensive legislation that we've seen recently. A couple weeks later, we got word that John Thune, who was then the Senate's number two Republican, was working on legislation backed by Democrat Amy Klobuchar on what they called a light-touch AI bill. Now, as opposed to defining it in opposition to the Holly Blumenthal approach, instead they defined it in opposition to what they anticipated as the potential, quote, heavy-handed plans that they expected from Senate Majority Leader Chuck Schumer. Thune told reporters, what we're trying to do is mitigate against the riskiest applications of AI. There are mechanisms in there in addition to self-certification

that I think create the safeguards, but without having the heavy hand of regulation that we think could be harmful. The Thune Bill, it sounds like, would do a lot more to place self-enforcement in the hands of AI companies than perhaps some others in Congress and the Senate are comfortable with. Now, still, while there are those big comprehensive legislative packages trying to pick up momentum, there are a heck of a lot more smaller and more specific acts dealing with AI that are attempting to address some specific problem or challenge or opportunity in a way that is potentially easier to get more consensus around because it is inherently more comprehensive. For example, there were two of these announced around the same time as the Second Insight Forum,

including S3050, the Artificial Intelligence Advancement Act, which would establish a bug bounty program and require reports and analysis on data sharing and coordination, artificial intelligence regulation in the financial sector, and AI-enabled military applications. On other bill, Senate 2691, the Shats-Kennedy AI Labeling Act, would require companies to create labels for content that was generated by AI. Senator Brian Shats from Hawaii said, This act puts the onus where it belongs on companies, not consumers. Because people shouldn't have to double and triple check or parse through thick lines of code to find out where something was made by AI.

It should be right there in the open, clearly marked with the label. Other examples of this that we've seen recently is in September. We saw bipartisan legislation called the Protect Elections from Deceptive AI Act, which was legislation designed to ban the use of AI to generate deceptive content falsely depicting federal candidates in an attempt to influence federal elections. Basically, they want to make it officially illegal to create fake AI videos or images of presidential candidates doing bad things. Another recent one came in October and was called the Federal No Fakes Act. Once again, it was released as a discussion draft by a bipartisan group of senators and basically extended

likeness laws officially to artificial intelligence as well as federalizing them. This came hot on the heels of Mr. Beast and Tom Hanks, both warning about AI ads that used their faces and voice to try to make people think that they were endorsing products that they had never had any interaction with. And one of the things that you may be noticing at this point is that when it comes to these specific acts, it feels like there's a little bit more of the common sense sort of rulemaking versus big contentious things. Now, with any legislation, with any regulation, there are going to be disagreements about

implementation and where to draw lines, even on things that seem pretty agreeable at first. But by and large, it's going to be a lot easier to get consensus around things like not allowing presidential candidates to use deep fake videos to malign their opponents than it is to get extremely impactful cross-cutting AI rules. Now, there is one other interesting dimension of this that I just wanted to mention briefly. A couple weeks ago, Politico released a very long comprehensive piece called how a billionaire-backed network of AI advisors took over Washington. The subtitle, a sprawling network spread across Congress, federal agencies, and think tanks is

It's pushing policymakers to put AI apocalypse at the top of the agenda, potentially boxing out other worries and benefiting top AI companies with ties to the network. It's a deep dive effectively on open philanthropy and the broader connections to the effective altruist movement, which has obviously gone through some pretty painful moments in the wake of Sam Bankman-Frieds, seeming egregious fraud. Basically, what this article tries to do is track all of the ways in which a larger set of institutions surrounding open philanthropy and the EA space have placed staffers. inside the offices of numerous politicians, as well as how intermingled they are with some of the

leadership of top AI labs. Now, I think it's a good piece it is absolutely worth reading. And what's more, I think it is always worth asking questions about potential conflicts of interest. It is reasonable, for example, to ask if members of a group called the Horizon Fellows should qualify for protections that allow nonprofits to pay the salaries of Senate and congressional staffers as long as those nonprofits don't have a specific policy agenda. But I think it also reveals something that's kind of interesting about the state of the AI safety discourse. There are two real boogeymen here presented in this political piece. They're not the boogeyman of Mark Andreessen in his techno-optimist manifesto, who is obviously on the far other end of the spectrum and is actually an accelerationist.

But instead, the big concerns here are, one, the concern around AI extinction risk-type issues will box out more near-term and present issues, such as bias and inequity. And two, there is a real focus on the idea of regulatory capture. The regulatory capture one is more difficult to assess from outside. On the one hand, I think you can make a pretty strong argument that it is very unlikely to be the government pulling up the ladder for future AI labs that makes today's leaders ultimately the biggest companies in the space, versus the incredible concentration of talent and capital and computing power that they have and their veritable head start. In other words, regulatory capture is theoretically a big issue, but in practice, how much is it going to really

change? Still, I do think that it is always worth asking questions of regulatory capture, because if nothing else, it will drive us to have the important conversations we need to about questions such as open source and models that clearly come under some threshold of concern. Now, on the second point, this idea that concern around speculative future harms is somehow blocking out attention for more immediate problems, it just seems like a weird argument to me. I think people are able, by and large, to understand that they're two different types and two different categories of potential risks and that they need to be addressed on their own and that they're both meaningful and significant. It's sometimes hard to feel like the folks

who are angry about how much traction the X-risk type conversation have aren't just jealous that the things that they're concerned about in AI aren't the ones that everyone is talking about. It just seems ultimately to me that if you truly believe that these are both categories of risks, It's not an either-or conversation, it's a both-and. Now, from the flip side, from the standpoint of the network of folks who are trying to raise those extinction-risk-type concerns, my question is kind of, what do we expect? If a group is genuinely concerned that the advance of a new technology has a meaningful chance of harming human existence, what do you expect them to do? Of course they're going to lobby.

The reality is that everyone in Washington lobbies. Money points in millions of directions all at once. And so again, I'm actually brought back to the more salient questions of whether there are specific, discrete conflicts of interest that are really worth noting, rather than just the general principle of, boy, the AI safety folks who are focused on human survival seem to be getting a lot of traction with those arguments. Now, I'm not endorsing any of these views. I'm not taking any sides on this. I'm just pointing out this interesting sort of meta-debat in the nature of the conversation around AI in D.C. right now. But, as I said a couple days ago, we are certainly getting to a crescendo moment in the space.

And of course, when we get that executive order tomorrow, I will be sure to let you know what it contains. For now, I appreciate you guys listening as always. And until next time, be safe and take care of each other. Peace.