The AI Daily Brief: Artificial Intelligence News and Analysis - Nancy Pelosi Urges SB1047 Veto

Episode Date: August 20, 2024

Former Speaker of the House Nancy Pelosi has voiced opposition to California’s controversial AI Bill SB1047. This episode examines the intense debate, covering Pelosi’s position, key arguments from both sides, and the potential impact on the tech industry and AI safety. Will this bill lead to significant changes in AI policy, or is a veto on the horizon? Get the latest updates and insights on this critical issue. Concerned about being spied on? Tired of censored responses? AI Daily Brief listeners receive a 20% discount on Venice Pro. Visit ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠https://venice.ai/nlw ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠and enter the discount code NLWDAILYBRIEF. Learn how to use AI with the world's biggest library of fun and useful tutorials: https://besuper.ai/ Use code 'podcast' for 50% off your first month. The AI Daily Brief helps you understand the most important news and discussions in AI. Subscribe to the podcast version of The AI Daily Brief wherever you listen: https://pod.link/1680633614 Subscribe to the newsletter: https://aidailybrief.beehiiv.com/ Join our Discord: https://bit.ly/aibreakdown

Today on the AI Daily Brief, former speaker of the House Nancy Pelosi has come out against controversial California AI Bill SB 1047. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. To join the conversation, follow the Discord link in our show notes. Hello, friends, quick note before we dive in, once again, and I think this tends to happen with these big policy discussions with lots of open letters, the main part of the episode ran long, and so we will not have a headline section today.

We will be back with our normal format, presumably tomorrow. But for now, we are just going to get into. the latest in SB 1047. Now, if you have not had a chance to watch it yet, I highly suggest you go back to my show from about a week and a half ago called SB 1047, the world's most important and problematic AI policy debate. I go fairly in depth there on the background of the issue and the points that the various sides have been making. And there have been points of plenty. The story of SB 1047 recently has just been open letter after open letter after open letter. On August 7th, for example, we got a set of professors arguing in favor of SB 1047, calling it extremely

light touch, indeed using the phrase that bare minimum for effective regulation of AI. But if the godfathers were for it, the godmother, quote-unquote, came out against it. Faye-Lee came out and said that while it was well-intended, she believed that it would actually harm the USAI ecosystem. Startups and venture capitalists have both published their own letters, including very notable ones from A16Z and Y Combinator. And where I had left off was that Congresswoman Zoe Lefgren, the ranking member of the House Committee on Science, Space and Technology, had also come out against the bill. In her letter to Bill sponsor, Scott Weiner, she wrote, I firmly support AI governance to guard against demonstrable risks to public safety, but this bill would

fall short of these goals, creating unnecessary risks for both the public and California's economy. As we started to get into in that previous show, part of the reason that the debate has been so intense is that this is the first time that there has been a specific instantiation in potential policy of the debate around AI safety. Brian Chow in Reason Magazine expressed one side of that in a piece he wrote titled California lawmakers face backlash over doomsday-driven AI bill. In that piece, he claimed that, quote, lawmakers were swayed by one-sided narratives from the open philanthropy-funded center for AI safety, which claims that mitigating the risk of extinctions from AI should be a global priority, until recently policymakers remained largely unaware of just how disconnected these

narratives were for the broader AI community. Now, I'm not using this particular show to debate the merits of AIX risk or anything like that, but it is clear that this bill and where it decides to put its focus is aligned philosophically with the AI safety movement. Indeed, that was at the core of Lofgren's criticism that there was basically too much in this bill about future theoretical risks and not enough about clear and present and current risks. And that brings us to the current where the most significant update came a few days ago when Nancy Pelosi and a group of other California representatives weighed into the friccacy in opposition to SB 1047. For a highly nonpartisan take, I turned to the New York Post, which characterized this as Pelosi blasting the bill as ill-informed.

There were actually two parts of this. The first was a letter from a group of Congresspeople to California Governor Gavin Newsom, and separately was a statement from former Speaker of the House Nancy Pelosi adding her support to that letter. So let's see what the letter actually says. It was signed by Zoe Lofgren, Anna Eschu, Rokana, Scott Peters, Toi Cardenas, Ami Berra, Nanette Beraghan, and J. Louis Correa. The letter begins, Dear Governor Newsom, it is somewhat unusual for us as sitting members of Congress to provide views on state legislation. However, we have serious concerns about SB 1047, and we felt compelled to make those concerns known to California state policymakers. They then go through

their credentials and say, based on our experience, we are concerned that SB 1047 creates unnecessary risks for California's economy, with very little public safety benefit, and because of this, if the bill were to pass the State Assembly in its current form, we would support you vetoing the measure. So what are their arguments? Many of the points of the letter echo things that were in the previous letter from Zoe Lofgren. They point out, for example, that the, quote, bill requires firms to adhere to voluntary guidance issued by industry and the National Institute of Standards and Technology, which does not yet exist. For example, they write, even though we do not yet have the standardized evaluations necessary for a developer to confirm with confidence that an AI

system could cause critical harm, the bill bases its liability provisions upon such hypothetical guidance. They also point out that the approach to regulating based on compute thresholds will, quote, almost certainly be obsolete by the time the law would go into effect. Such premature requirements, they say, based on underdeveloped science, call into question from the outset the efficacy of the bill in achieving its goals of protecting public safety. We should not seek, they right, to cement our current understanding of AI safety science into law, but should instead provide ample flexibility, agility, and public consultation to allow the law to grow as our understanding grows. In its current form, SB 1047 falls short. The letter also calls out the focus of the bill

on the more existential risks as opposed to more current problems. They write, SB 1047 is skewed towards addressing extreme misuse scenarios and hypothetical existential risks, while largely ignoring demonstrable AI risks like misinformation, discrimination, non-consensual deepfakes, environmental impacts, and workforce displacement. Unlike other letters, they go even farther arguing against the notion that AI somehow represents a new threat. They write, there is little scientific evidence of harm of mass casualties or harmful weapons created from advanced models. They use the example of the threat of AI models creating nuclear weapons, but write that the quote, production or acquisition of the fissile material is the primary impediment to the creation of such a weapon, because the technical

details of nuclear weapons have been known since at least 1945 and can almost certainly be acquired on dark corners of the internet. Now, this is not to say that they don't think any of these issues are worth dealing with. They write, understanding, measuring, and monitoring the risk inherent in AI systems as they evolve will be key. Especially the marginal risk of AI systems causing mass casualty events is contemplated by SB 1047. There is also a pressing need to develop the scientific tools, tests, and standards to enable the effective evaluation and assurance of AI systems.

They also, as has been the case with almost all of these letters, say that they support things like the whistleblower protection provisions. This letter also goes into some detail about the other AI-related bills, also under consideration in California, that are focused on more of what they consider the demonstrable harms. writing that, quote, these bills have a firmer evidentiary basis than SB 1047. Now, this counter response to the focus on AI safety concerns is the first and primary critique that they have, getting top billing in the letter, but secondarily, they also say that they're concerned

around the unintended consequences from SB 1047's treatment of open source models. They write, not only is it unreasonable to expect developers to completely control what end users do with their products, but it is difficult, if not impossible, to certify certain outcomes without undermining the rights of end users, including their privacy rights. As such, the natural response from developers will be to stop releasing fully open AI models and instead implement limited release models like APIs. They basically argue that the provisions, including things like the Kill Switch, would, quote, decimate the ecosystems that spring up around AI models.

One notable thing that goes farther than previous letters that we've seen is that they're open to the possibility that at some point open sourcing models does become too dangerous. They write, it may be the case that the risks posed by open sourcing models with potentially dangerous capabilities justify this precaution, but current evidence suggests otherwise. After seeking comment from the community and looking at the risks, the national telecommunications and information administration released a report last month saying government should not restrict access to open source models with widely available model weights at this time, but instead should actively monitor the ecosystem should risks evolve. Basically, the argument here is that at current

capabilities, the benefits of open sourcing outweigh the risks, but of course that could change. Of course, there is an economic dimension to this for this particular group of representatives, and they're not circumspect about that. They write, given that most of the discoveries that led us to this moment were achieved through openness, SB 1047 could have a pernicious impact on U.S. competitiveness and governance in AI, especially in California. In short, they conclude, we are very concerned about the effect this legislation could have on the innovation economy of California without any clear benefit for the public or a sound evidentiary basis. High-tech innovation is the economic engine that drives California's prosperity. There is a real

risk that companies will decide to incorporate in other jurisdictions or simply not release models in California. This is not entirely speculative. As an example, they point to meta deciding to not release its multimodal systems in Europe due to the EU-AI Act. Thus they write, while we are confident of the good intentions of the bill's proponents, we are equally confident that this bill would not be good for our state, for the startup community, for scientific development, or even for protection against possible harm associated with AI development. Because of those reasons if the bill were to pass the legislature in its current form, we would support you vetoing the measure. Today's episode is brought to you by Venice. The leading AI companies store your

entire conversation history and attach it to your identity forever. That's every question you ask, every answer you receive, every image you generate, every thought you share with the machine it's all being spied on. If you trust all the company's hackers and NSA board members that will ever have access to your AI conversations, then rejoice, for you are well served. For the rest of us, Venice is an alternative. Venice is a powerful AI app for text, image, and code generation that respects you as a sovereign individual and believes privacy and free speech are not only human rights, but necessary for civilizational advancement. permissionless and uncensored, you can try it for free without an account. AIA Daily Brief listeners

receive a 20% discount on Venice Pro. Visit venice.a.i slash NLW and enter the discount code NLW Daily Brief. That's NLW Daily Brief. All one word. Today's episode is brought to you by Super Intelligence, the platform that helps teams maximize AI. Super is of course the platform that we've been building that pairs fun, fast, practical video tutorials with step-by-step instructions to get you actually using AI, and from there unlocks information about hundreds of use cases that show you how people are actually getting value out of AI right now. Now, we have just launched Super for Teams. This is a new add-on experience that allows teams to share more information about what AI they're using, how it's working, and how to get more value out of it. Whether your company is 25 or 2,500,

superintelligent is going to be the best platform for unlocking information about how to get the most out of right now. If you'd like to learn more about these super intelligent teams offering, go to B-super.a-I slash partner and send us a note so that our team can get right back to you. Once again, that's B-super.com. A.I. slash partner. Now, if you thought that State Senator Scott Weiner would be deterred by this, you would be wrong. In his own press release, he wrote, I am aware of Speaker-Amerita Nancy Pelosi's statement opposing SB 1047. While I have enormous respect for the Speaker-Amerita, I respectfully and strongly disagree with her statement. I don't know that it's the most effective technique, but he takes quite a cagey tone saying, I encourage the Speaker Emerita to have a

conversation with the AI luminaries who are quite, quote unquote, informed, and who are supporting SB 1047, pointing, of course, to Benji O'Hinton, Lawrence Lessig and Stuart Russell. Weiner reiterates in his press release, his belief that it's a, quote, straightforward common sense light touch bill, that it only requires AI developers to do what they've already committed to, and he says, really all the opposition is just big tech. He says that untrue narratives have continued to percolate around about SB 1047. Wiener argues that some in the tech space have taken a get-off-my-law approach to policymakers, telling us we should do nothing while offering no constructive policy ideas other than to just let technology companies do whatever

they want to do without even minimal regulation in the public interest. And this is the point at which it becomes clear that this is not just about AI legislation, but about a reckoning with tech policy in general. Weena writes, the aggressive anti-regulatory approach by the tech sector has led to almost no federal protections for public health and safety relative to technology. And then he goes on to list places that he believes that California has had to, quote, step in to protect our residents. This theme that this is all just a well-funded campaign by the technology industry to have no regulation at all is something that has particularly infuriated the AI safety community. The AI safety memes account, which I will say for those of you who are unfamiliar, has a very clear perspective,

which of course you can guess from the fact that they're the AI safety memes account, but is generally thoughtful and not prone to histrionics, writes, the A16Z dark money campaign is getting uglier. Recently they've gone full corporate villain mode, throwing around cash and just openly lying about what SB 1047 actually says. Not just saying misleading things lying, repeatedly on the record over and over. They will say anything to kill this bill. They shriek endlessly about their outgroups being cults while blocking anybody who disagrees with them. The actual cultiest thing you do to create the false appearance of consensus. It's not just A16Z, of course, but most of the opposition is from a handful of billionaires in big tech. Despite the fact that the big AI

corporations are all lined up against this bill, they're actively pushing an absurd conspiracy theory that actually it's the tiny AI safety non-profits doing all the lobbying. I think for most of you viewers slash listeners who sit fairly comfortably in the middle of, on the one hand, the accelerationists and on the other hand, the most concerned subsection of the AI safety group, there's a lot that you can understand about the narratives being pushed by either side of this and having the contextual grains of salt to understand how to contextualize whatever they're saying while still looking at the legitimate points underneath. While I saw numerous arguments from the AI safety side of the aisle, arguing that this letter from these congressional representatives

simply reflects lobbying money being successful, my guess is that probably a lot of the listeners to this show are fairly sympathetic to, one, concerns around preemptive regulation for areas of hypothetical if serious risk, and two, sympathetic to concerns around unintended consequences around open source. The debate, as it always is with politics, is around the details. There is no such thing as perfectly knowable information, and that's only extra true in the case of this new and incredibly fast-moving technology. Now, the latest update today, POMS once again from Zoe Lofgren and Anna Eschew, this time to Robert Rivas, the Speaker of the California State Assembly. Basically, this letter is pointing out that in the wake of the letter that they published last

Thursday, the bill was amended substantially during markup. They write, we would like to acknowledge the efforts that the author of the bill in the State Assembly made to improve the underlying proposal with additional flexibility and clarity. Unfortunately, they write, there are still substantial problems with the underlying construct of the bill. It is our view that the bill in its current form should not be approved by the legislature. Most of the rest of the letter is just a rehash of those things that they had said in the previous letter. They also shared, however, the memo that the staff of the science committee had sent to ranking member Lofgren in their review of the updated bill. Though staff write, overall SB 1047 is considerably better than it was

before. They weakened or clarified many of the key regulations. However, the problematic core concerns remain. There is little evidentiary basis for the bill. The bill would negatively affect open source development by applying liability to downstream use. It uses arbitrary thresholds not backed in science, and catastrophic risk activities like nuclear biological deterrence should be conducted at a federal level. Ultimately, they write, the changes seem to be aimed at placating various concerns from specific parties rather than addressing the core challenges of the bill. They point specifically to the example of Anthropic, having had their specific changes adopted, suggesting that Anthropic, quote, likely no longer is opposed. Some of the major amendments,

they point out, one area is around clarifying harms. They write, one amendment scopes harm to only those caused or materially enabled by a developer. This is a positive addition they write as the initial lack of a strong standard was problematic. The bill also exempts harm from information publicly accessible by an ordinary person. One of the things they don't like, they write the amendment adds a $10 million cost threshold for fine-tuning an AI model. Similar to the 10 to the 26 computing threshold, this is another arbitrary threshold with little to know scientific backing that will do little to constrain the scope of the law as models and the companies that build them improve over time. That said, they do acknowledge that, quote, some researchers believe this is an improvement on the

previous language. Overall, they still come back to this idea that the fundamental architecture of this bill is about the wrong thing. And to me, having reviewed this and continue to watch it, I think this is simply the inextricable challenge of this particular piece of legislation. Now, as I'm sure you've noticed, this has been an extremely tense conversation. But to the extent that you are looking for a silver lining of this, this is exactly the type of regulatory process that tends to lead to massive political education and ultimately better policy. The AI safety folks, for example, may not like that a lot of their beliefs are not being affirmed by these sitting Congresspeople. They may have concerns as well that they're being influenced

by money from big tech. But when you read these letters and the staff memos going into them, it's undeniable that these questions are being taken seriously. I simply tend to believe that questions being taken seriously tends to lead to better policy than things just slipping through. Now, of course, this is all a little bit of a sideshow to a broader national level conversation around what federal policy should be, and perhaps one of the legacies of this process will be that it will create context for Congress to actually try to bring these issues into the federal discussion. Of course, there is a little chance of that happening before the elections, but I don't think that the impact of this bill, whatever happens from here, will simply be limited

to whatever happens in California. Anyways, those are the updates from here on SB 1047 that will, of course, keep you informed as things change. For now, though, that is going to do it for today's AI Daily Brief. Until next time, peace. Thank you.