The AI Daily Brief: Artificial Intelligence News and Analysis - The Faux Outrage Around OpenAI's EU Lobbying

Episode Date: June 20, 2023

Today TIME published a "scoop" about OpenAI's lobbying efforts in 2022 around the EU AI Act. NLW argues that the story is not only not some smoking gun, but is the type of nothingburger that makes people drop out of otherwise important conversations because they're feeling like the media is trying to manipulate their feelings. Before that a new research study reveals some unexpectedly positive attitudes towards AI. Also on The Brief, AI adds $150B to the net worth of the world's richest. 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 talking about the faux outrage around open AI's EU lobbying. Before that on the brief, a very interesting survey about people's attitudes towards AI. The breakdown is a daily podcast and video about the most important news and stories in AI. Like, subscribe and share, and go to breakdown.network for more information. Welcome back to the AI breakdown brief, all the AI headline news you need in five or seven or whatever number of minutes it takes. Today we kick off with a study from Cap Gemini Research. new report is called why consumers love generative AI. And there are a couple big banner headlines that are really quite surprising. First of all, they show that boomers followed by Gen X have the highest

level of awareness of AI tools. Millennials come in the lowest in this survey. The study also suggests that they're really trusting when it comes to AI. 73% of consumers trust content written by generative AI. 53% of consumers trust generative AI assisted financial planning. 67% of consumers believe they could benefit from receiving medical advice from generative AI, and 66% of consumers would seek advice from a generative AI tool on personal interactions or relationships. Now, of course, watching this space every day doesn't make me an authority, but I would put these all at significantly higher by a factor of maybe double than I would have anticipated. The study also suggests that people don't really have a sense of risk.

49% of consumers they surveyed were unconcerned about fake news stories created by AI. only 34% were worried about fishing attacks that used AI to deceive them, and only 33% were worried about things like non-recognition and non-payment of artists whose work is used in training generative AI models. Now then there's a whole segment about how AI could be a tool for consumer search and for product search, and that I all buy. These are completely low stakes in obvious ways that this technology is going to impact our existing online experience, so no real skepticism there.

So I tried to look a little bit deeper into how this could possibly have come up with these results. Again, they just feel really off to me. But Cap Gemini is a respected source for this. It's not some random company or something like that. Now, the only thing that I could find that might explain this, at least a little bit, is that the survey took place primarily in April 2023, but between October 22 and April 23. This might explain why people aren't particularly concerned about the risks. Obviously, as you can see from this show, the AI safety conversation has accelerated dramatically

over the last two months with high-profile defections of people like Jeffrey Hinton from Google. And alongside that, there has been more concerned generally about AI risk, not just the existential or extinction risk, but around the short-term risks that exist already. Still, overall, super interesting and definitely at least one data point that we need to consider as we think about what the world thinks of AI. Speaking of what people think about AI, it has been very expensive to be an AI skeptic for the last few months. Wall Street Journal has reported that this year, short sellers have lost $120 billion in U.S. markets. And of course, the primary driver of those markets has been AI and AI enthusiasm. The tension all year has been, on the one hand, really uncertain macroeconomic conditions, including a Fed that's still tightening, banking crises, tightening credit, worsening

liquidity conditions, all the normal things that would make a market worse. But those things, of course, have come up against the unbelievable strength of stocks like meta and Microsoft and Nvidia. In the first half of June alone, short sellers have lost $72 billion. Meanwhile, over in the world's wealthiest sector, that group has seen their net worths go up by more than 150 billion thanks to AI this year. Meta is up 134% on the year, increasing Zuckerberg's wealth by over 57 billion. Oracle is up 55% giving Larry Ellison a net bump of 47 billion and making him pass Bill Gates for the first time ever. Gates himself has gotten $24 billion richer, as has Nvidia founder Jensen Huang.

Good work if you can get it. Now, meanwhile, a company that's on the other end of its spectrum is 11 Labs. They're a voice cloning and voice synthesizer product and have just raised a $19 million series A. Now, you'll remember last week we talked about their new speech classifier. Speech classifier is a safety technology that allows people to upload an audio clip and find out if 11 Labs models were used to generate any portion of that audio. Speech classifier has come online and 11 Labs has also announced a new feature suite called Projects. It's a production workflow that allows people to create and edit long-form spoken content directly in the platform. Speaking of AI Creation, Vimeo has joined the AI battle.

They've just announced a new suite of products that include one, a prompting tool that can generate scripts in seconds based on a few key words as well as a description of tonality. And then they've also added a descript style editing suite where you can edit the transcript and that actually edits the video as well. It can automatically pull out pauses and fillers. And generally, if you've used a script, you know just how game-changing this type of video. video editing is. Now, when it comes to the question of whether AI will be massively job destroying or will instead usher in a new golden era in which humans can use their times for higher order pursuits, Airbnb CEO Brian Chesky believes the latter. In a conversation on Jason Kalakanis' this weekend startups, Chesky said that he believes that AI will propel millions of new

startups. He said that he thinks that AI will allow engineers at Airbnb to be 30% more productive in the next year, and that the net impact was going to be a lot more creation. Chesky said, I think anyone can do what only a software engineer allowed you to do five years ago. It's going to be awesome for many people. It's going to be wildly disruptive for others. Lastly, today, Infigen, which stands presumably for infinite generator, it's an open source procedurally generated photorealistic dataset for 3D vision. So this is not something that's created by AI. It's just procedurally generated by math. The way that they positioned it is as an automatic generator for 3D scenes.

As Nvidia's Dr. Jim Fan puts it, every little detail is randomized and customizable. even the wrinkles on a flower pedal, and it features ground truth automatic annotations, optical flow, 3D scene flow, depth, surface normals, penoptic segmentation, occlusion boundaries. Now, Dr. Jim points out that this could be incredibly valuable for the training of future AI models. He says when we run out of good training data in reality, simulation is the next goldmine. So even though this isn't AI itself, it could have implications for how future AI is developed. And of course, it's part of this trend of 3D world creation that is just getting so, so much easier every single day. Anyways, guys, that is it for today's AI breakdown brief.

If you're enjoying, please like, subscribe, and share, and I'll be back soon for the main AI breakdown. Today we're talking about lobbying, AI safety, and full outrage that could absolutely undermine everything that anyone in AI safety is trying to achieve. Welcome back to the AI breakdown, aka a live NLW aneurysm today. I woke up this morning to a scoop from Time magazine. Billy Perigo, the investigative journalist behind said scoop, writes, flashing red light emoji scoop open ai lobbied the EU to weaken forthcoming AI regulation even as in public it calls for stronger AI guardrails documents obtained by time show so what we have here is the promise of a smoking gun of hypocrisy a confirmation of the sort of regulatory capture that many have suspected on the part of sam altman and open AI the idea or the concern is that the reason that open AI or companies like them are advocating for regulations or rules is that they really just want to pull the ladder up behind them and make it so that new competitors can't compete. Now, by extension, they also want the rules to be things that they can actually comply with.

And so, of course, they have an incentive to, as Times puts it, water down those regulations. Now, in question is a white paper on the European Union's Artificial Intelligence Act from September of last year. And it's worth starting with a little bit of context. The EU started the process for the EU AI Act in 2021. This was pre-generative AI. It was designed to be a risk-based approach, in which there were different remediation, recommendations, policies, mechanics, etc., for different AI based on what it was actually being used for. In other words, the EU theoretically recognized that there was a difference between generating an image and using AI to profile future potential criminals and arrest them in advance. Now, importantly, mid-2020 is when Dali, GBT3, and Stable Diffusion all came out,

showing the promise of generative AI that hadn't really been considered in the first year of drafting this act, and so the European Parliament quickly worked to graft on a set of new provisions for generative AI. Now, of course, when you graft on new provisions that have been considered for less time, there is naturally going to be a bit more calibration required. This is inherent in any sort of policy process. There is precedent for how to deal with this in Europe with the MECA or Markets and Crypto Assets Regulation Act. In Mika, which was started during the ICO boom, they made the determination to not include too much about NFTs and defy, even though those things have been around for a couple years now, in the rules that were passed just this year.

The reality was that those issues were different than the context that had been negotiated, and so they needed to be saved for Mika too. Now, of course, in the AI Act, they weren't going to ignore generative AI completely. But they did need to race to catch up, and so as part of their consultation with the EU, OpenAI was asked to present some suggestions, and in September 22 did in the form of this white paper that Time has now published. Before we discuss the way the time frame this, let's just look at what the actual paper says. The first section is all about what classifies as a high-risk system. Again, the EU was making different rules for how risky something was, and so OpenAI was, of course, concerned with whether its tools would be labeled high-risk or not. Basically, what they argue in this section is that the way that the rules were written at the time,

Their general purpose models such as GPT3 might be considered high-risk because they could theoretically be used for high-risk purposes. In contrast, however, OpenAI writes, By itself, GPT3 is not a high-risk system, but possesses capabilities that can potentially be employed in high-risk use cases. Accordingly, we have dedicated significant resources to determining guidelines, best practices, and limitations for uses of our services. We currently outline a set of high-stakes applications in fields such as law, medicine, politics,

finance, and civil services, where applications proposed to be built using our services are subject to additional scrutiny that requires clear identification and management of risks. We consider and continue to review on an ongoing basis the different ways that our systems may be misused, and we employ many protective measures designed to avoid and counter such misuse. The current framing may inadvertently incentivize an avoidance of active consideration of ways that a general purpose AI system may be misused, so that providers do not have, quote, sufficient reasons to consider misuse and can avoid additional requirements. The fundamental nature and value of general purpose AI systems are that they can be used for many application areas.

We do not think it would meet the goals of safe and beneficial AI to inadvertently encourage providers to turn a blind eye to potential risks. We suggest reframing the language to incentivize rather than penalize providers that consider and address system misuse, especially if they take actions that indicate they are actively identifying and mitigating risks. The second section is about a provision that would designate a huge portion of content generation systems as high risk, because they generate, quote, text content that would falsely appear to a person to be human-generated and authentic, or, quote, audio and video content that appreciably resembles existing natural persons. OpenAI again points out that they have a number of systems in place to allow them to

verify synthetic origins of images, have rules around people purposely using their tools to mislead, and suggest and said that the AI Act can, quote, sufficiently require and ensure that providers put into place reasonably appropriate mitigations around disinformation and deepfakes, such as watermarking content or maintaining the capability to confirm if a given piece of content was generated by their system. TLDR, the way that at that time the provision was written in the AI Act, would make content generation AI a priori high risk, versus trying to put in place systems that prevent high risk uses for the underlying technology. The third section is about what they call new conformity assessments for substantial modifications.

Basically the idea that if OpenAI's model, which was previously approved, gets some big update it needs to be approved again. On the face of it, the principle isn't disagreed with, but what they talk about is the fact that this gets very murky very quickly. In other words, what constitutes a substantial modification? Software in the modern world is deployed extremely iteratively. Small updates are pushed constantly rather than there being one big update per month or anything like that. Now, even with that, OpenAI's only suggestion was to create an exemption for updates and modifications that are made for safety or risk mitigation reasons. with, by the way, the ability to be rolled back should there be compliance concerns thereafter.

The fourth and final section is really just about debating what uses are high risk or not, which is of course what the European Parliament was doing at the time, and in some ways this is just Open AI participating in that conversation. The examples they give include job seeking, basically saying, yes, it's high risk if AI is being used exclusively as the way that someone is selecting who should be hired. But there's also plenty of uses of AI which could help modernize and significantly speed up and improve the job-seeking process, such as people using generative AI to write better job descriptions. They make a similar argument around vocational training. The piece concludes, given the continued

advancement of AI systems capabilities, we expect that currently unknown high-risk use cases will continue to emerge, making it important to ensure that the AIA remains agile in capturing ongoing developments, quickly capturing new high-risk AI systems and removing those which have proven themselves sufficiently low risk must be low friction. So that is the actual piece that this whole scoop was based on. But let's now talk about how Time Magazine decided to frame their piece. First of all, the headline, exclusive, OpenAI lobbied the EU to water down AI regulation. It's hard to ignore the politics of choosing the phrase water down to describe this, but let's continue. The lead reads the CEO of OpenAI Sam Altman has spent the last month touring

world capitals where at talks to sold out crowds and in meetings with heads of governments, he has repeatedly spoken of the need for global AI regulation. But behind the scenes, OpenAI has lobbied for significant elements of the most comprehensive AI legislation in the world, the EU's AI Act, to be watered down in ways that would reduce the regulatory burden on the company. Now, I don't know about you, but that sounds a little bit different than what we just read. Now, stunningly, to time, it appears that the EU might have actually thought some of the ideas and suggestions were good. Time rights, in several cases, OpenAI proposed amendments were later made to the final text of the EU law. Indeed, time writes, OpenAI's lobbying efforts appear to have been a success.

The final draft of the Act approved by the EU lawmakers did not contain wording present in earlier draft suggesting that general purpose AI systems should be considered inherently high risk. Instead, the agreed law called for providers of so-called foundation models or powerful AI systems trained on large quantities of data to comply with a smaller handful of requirements, including preventing the generation of illegal content, disclosing whether a system was trained on copyrighted material, and carrying out risk assessments. Now, of course, the presumption lurking behind that text is that this was a fundamentally worse approach. This was the watering down of which the article speaks. And in case you had any

doubts that that is what time is trying to say, here are the quotes they chose to run as evidence. From Sarah Chander, a senior policy advisor at European Digital Rights, quote, they got what they asked for. The document shows that Open AI, like many big tech companies, have used the argument of utility and public benefit of AI to mask their financial interest in watering down the regulation. Another quote, this one from Daniel Lefer, a senior policy analyst focused on AI at Access Now Brussels office. Quote, what they're saying is basically trust us to self-regulate. It's very confusing because they're talking to politicians saying,

please regulate us. They're boasting about all the safety stuff they do, but as soon as you say, well, let's take you at your word and set that as a regulatory floor. They say no. Now, to read Twitter, a set of people who I can only assume haven't actually read this piece are pushing this exact narrative and tonality from the piece. Fellow tech podcaster Paris Marx writes,

Billy Perigo confirms what many of us suspected. Sam Altman went around the world talking about the need for AI regulation, but in private, OpenAI successfully lobbied to water down the EU's AI Act. It ensured its products would face less stringent rules. Sasha Costanza Chalk writes, Open AI. Our extremely powerful general artificial intelligence might kill us all. Please regulate us. Also Open AI, but on the DL to EU regulators,

what we're doing is not really high risk and it's very important you let us run experiments in the wild. Cheris Papavangelo writes, ah, the typical big tech capitalist discourse of please regulate us, but not really. And then there's Scott Galloway who frankly should know better writing, Shocker. OpenAI CEO Sam Altman giving speeches saying AI regulation is essential. Behind the scenes, OpenAI lobbies for significant elements of the EU's AI Act to be watered down. So there are two possible critiques of Open AI here. The first is that they're lobbying at all.

Well, guess what? They're a company. Companies lobby. In fact, everyone lobbies. Individual citizens lobby, nonprofits lobby, advocacy organizations lobby, corporations lobby. The fundamental fact of reality is that corporations will lobby for what they think are better policies, including better policies for them. The whole point of democracy is that it's strong enough to take lots of different types of lobbying in

and still come away with something that nudges us forward, even if it pisses off everyone along the way. The second possible critique, and this is really the big one, exhibit in all of these tweets, is that OpenAI is being hypocritical, that somehow Sam Altman is speaking out of both sides of his mouth. You see it in all these tweets that, in front, front of the press, they're saying that AI regulation is important, and behind the scenes, they're really trying to water things down. This is why language matters. All of these tweets, shaping all of the opinions that they're shaping, are predicated on the idea that this represents an attempt to water down the legislation. It's right there in the title. However, as someone who does multiple videos and a

podcast every day about this industry, I would venture to say that I've followed this a little bit more closely than most of these folks who are tweeting now. And what I saw was Sam Altman and OpenAI getting castigated by EU politicians for saying that they were worried about overregulation and that the way that the EU AI Act was structured at current. And by the way, this was in May, might not be workable for them and they might have to actually leave the EU. Now, they walked that back, and Sam Altman then went and had what were apparently productive conversations with numerous members of the European Parliament. But there was a whole press cycle, being mad at Sam Altman for saying that OpenAI was worried that the EU might be over-regulating. Kind of undermines the idea that he's saying

one thing in public and another thing in private. Secondly, when it comes to OpenAI's position on things, people seem to be taking the idea that OpenAI is advocating for regulation and assuming that it means that to be legitimate and not hypocritical in that, they have to be lobbying for the type of regulation that these critics want to see, or that the far left in the EU Parliament wants to see. The possibility that these critiques haven't considered is that OpenAI is outside of just advocating for its own interest. lobbying for policies that they think would be better in general for regulating this space.

These critiques don't consider the possibility that badly written regulation can do as much harm as it does good or even more. They don't consider the unintended consequences of loosely written words drafted by people who are just trying to understand something very new, whose end result has enormous power to shape things in directions that warp something for the worse. And it's not like this is abstract. GDPR is well-intentioned legislation that does far less to protect. privacy, then it does to, one, annoy people because they constantly have to, one, opt into their privacy being broken, but it's still broken, and two, radically increase the market share of Facebook and Google when it comes to advertising because they're the only companies that are

willing to pay for GDPR compliance costs. In other words, the squawkers on Twitter are so convinced that regulation is a priori good and so busy being angry at OpenAI for trying to push for the regulations that they think would be better, that they didn't take any time to share how they think the regulations could be better. It's all one big political game of confirming people's priors. So why does this matter? It's just one dumb piece, right? It matters in my mind because when the fourth estate constantly claims scoops and smoking guns that are actually nothing burgers, it undermines their credibility. In the world we're moving into, where we have the serious AI policy conversations that we're going to have, we need our institutions to have

credibility. We need our political institutions to have credibility. We need our media establishment to have credibility. This was a hit piece, pure and simple. It had an agenda going into it. The words written in Times piece don't match the words on the document that they shared. And so it's hard to see it as anything other than politically motivated. Maybe not just at Open AI, but maybe as part of the larger culture war battle that the media has been fighting with big tech for going on five years now. Bad critique doesn't hold company's feet to the fire. Bad critique undermines important critique later on. Bad critique doesn't lead to change, it leads to people tuning out. And so for people who are genuinely concerned about the risks of AGI, they should be frustrated that time is wasting,

its scoops and its exclusives, and its heightened Twitter drama on a non-story for the sake of clicks. But whatever. So I have a podcast so I can go out and tell you what I think about the news, instead of just having to listen to the garbage that is fed to us constantly, passed off his fact, and wrapped up in a bow to get us angry and predictable ways. That's it for today's AI breakdown. If you're enjoying this, please like, subscribe, and share. Check out the podcast version in the newsletter. And until next time, don't believe everything you read.