The AI Daily Brief: Artificial Intelligence News and Analysis - OpenAI Investigated by FTC for ChatGPT Harms

Episode Date: July 14, 2023

The Federal Trade Commission (FTC) is looking into whether OpenAI's ChatGPT has caused consumer harms in terms of data collection and safety practices as well as disparaging information. A look at what the implications are for the larger state of AI regulation. Before that on the Brief the Screen Actor's Guild has joined the Writers Guild of America in their strike, and AI is at the very center of the battle; Hugging Face is raising at a $4B valuation, more controversy at Stability AI, and Twitter is suing companies for scraping its data. Today's Sponsor: Supermanage - AI for 1-on-1's - https://supermanage.ai/breakdown 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, the FTC is investigating open AI. Before that on the brief, SAG has joined the Writers Guild in a big strike that could be monumental in how labor responds to the advancement of artificial intelligence. The AI breakdown is a daily podcast and video about the most important news and discussions in AI. Go to Breakdown.network to find our YouTube channels, our newsletter, and of course our Discord. Welcome back to the AI breakdown brief, all the AI headline news you need in five-ish minutes or less. We start today with a story that could very easily be the subject of an entire full

episode of the AI breakdown, which is, of course, that the Screen Actors Guild has joined the Writers Guild of America in their strike. Now, this is the first time that these two unions have been on strike at the same time since 1960, when Ronald Reagan was the head of SAG. Now, of course, if you've been following the strike at all, it's not just about artificial intelligence. There are many issues of contention, but at the same time, it's hard to deny that AI has emerged as the embodiment and representation of everything that this fight is about. That only ratcheted up yesterday when the Screen Actors Guild formally joined the strike, given how much they identified an AI proposal from studio executives as pushing them over the edge.

The Alliance of Motion Picture and Television Producers released a statement that said, AMPTP member companies entered into contract negotiations with SAG AFRA with the goal of forging a new mutually beneficial contract. The AMPTP presented a deal that offered historic pay and residual increases, substantially higher caps on pension and health contributions, audition protections, short and series option periods, and a groundbreaking AI proposal that protects actors' digital likenesses for SAG after members. A strike is certainly not the outcome we hope for as studios cannot operate without the performers that bring our TV shows and films to life. The union has regrettably chosen a path that will lead to financial hardship for countless

thousands of people who depend on the industry. Now, to say that SAG members were gobsmacked by the description of this AI proposal as groundbreaking is not saying it strongly enough. One of the lead negotiators for SAG Duncan Crabtree Ireland had this to say about it. Groundbreaking AI proposal that they gave us yesterday, and that groundbreaking AI proposal, they proposed that our background performers should be able to be scanned, get paid for one day's pay, and their company should own that scan, their image, their likeness, and should be able to use it for the rest of eternity in any project they want with no consent and no compensation. So if you think that's a groundbreaking proposal, I suggest you think again.

Now, in response to this, a spokesperson for AMPTP basically said that that's not exactly true. That statement read, the claim made today by SAG-AFTRA leadership that the digital replicas of background actors may be used in perpetuity with no consent or compensation is false. In fact, the current proposal only permits a company to use the digital replica of a background actor in the motion picture for which the background actor is employed. Any other use requires the background actor's consent in bargaining for the use subject to a minimum payment. Now, frankly, it seems here like the AMPTP is quibbling over details when SAG is getting the message loud and clear that there only needed a very little bit and then AI can take it from there. Listen to this clip from the speech given yesterday by SAG after President Fran Dresher, probably best known for her role as the nanny in the 90s, and it's pretty clear just how front and center this AI issue really is.

They stand on the wrong side of history at this very moment. We stand in solidarity. in unprecedented unity, our union and our sister unions and the unions around the world are standing by us as well as other labor unions. Because at some point the jig is up, you cannot keep being dwindled and marginalized and disrespected and dishonored. The entire business model has been changed by streaming, digital, AI. This is a moment of history. That is a moment of truth.

If we don't stand tall right now, we are all going to be in trouble. We are all going to be in jeopardy of being replaced by machines. What we have here is the first live action battle. between a labor movement on the one hand and a management class on the other who now has access to super advanced AI. The battles around the transition to a world where AI exists are going to be immense and numerous. And I anticipate that part of the reaction won't just be people re-skilling and retraining and changing everything that they do, but actively organizing in a way that we haven't seen for decades to try to exert some control around how much ability corporate leadership has in various industries to transition people out and AI. in. Holding aside any determination of right or wrong or justice or not, this will, I believe,

set a template for other labor organization in other areas that will have an impact on how AI is rolled out. Next up, staying in the realm of battles, Twitter has filed a lawsuit against four companies claiming that they scraped large amounts of data from the service without permission or payment. The lawsuit says, quote, these requests have severely burdened X-Corps servers and degraded the user experience for millions of X-Corps customers. Now, in an enough itself this wouldn't be all that interesting, but in the context of this week, Elon announcing XAI and all the legal threats he has made over the last couple months around Twitter data, it really does seem like the picture is becoming clear.

Twitter is not just a service that he loves. It is a trove of data that can train their particular AI models that they are working on, and it looks like they're going to avail themselves of the legal system to make sure that no one else can do that without their permission. Moving on, another gnarly update in the saga of Stability AI. This is a company that has been beleaguered by not so generous press reports and accusations from early founders, and there is now another one. An original co-founder of the company, Cyrus Hodes, has claimed that current CEO, Amad Mostok, tricked him into selling his 15% stake in the company by convincing him it was worthless, even at the same time as he was going out and having meetings with venture capitalists that would ultimately lead to a $100 million venture financing round. Now, the lawsuit also has other allegations against Imod, including that he embezzled external investors' funds for his personal use.

Now, I, of course, have no idea what the truth here is. It'll be for the courts to decide. But I do agree with Josh Wolfe when he writes, Every tech movement in its first phase of hype cycle has fast and loose early chaos that self-corrects and demands tightened scrutiny and cost of capital for future companies in phase two. And I have to say, just because they've been beset by scandal, stability AI has certainly not stopped pushing new products out. Their latest is something that they call stable doodle which takes sketches and turns them into real pictures that then can be modified using natural language inputs. There are a ton of creative implications for this, prototyping, design, graphic novels, gaming,

designing worlds, you name it, this I think is a really natural flow. And of course, on top of any sort of business or professional uses, this is just going to be incredibly fun to play with. And last up today, Forbes is reporting that Hugging Face, the GitHub for machine learning, is raising a new $200 million venture round at a $4 billion valuation. That's up from $100 million raised on $2 billion last year, so not bad for about 12 months of work. Now, obviously, if you are invested in the open source part of the space,

Hugging Face is a huge asset. So I, for one, I'm certainly excited to see them have more resources to build. Anyways, guys, that is going to do it for today's AI breakdown brief. If you're listening to this, go check out the YouTube. If you're watching this, go check out the podcast. And until next time, peace. Before we get into the main AI breakdown, I want to tell you about today's sponsor, Supermanage. If you work in a professional setting, you probably have some version of a one-on-one meeting,

either with the people that work for you or the people that you work with. Unfortunately, all too often, those one-on-one meetings become glorified catch-up calls. Don't you wish you could jump right to the stuff that really matters? That's where SuperManage comes in. Supermanage AI magically distills your team's public Slack channels into a real-time brief on any employee, any time. Catch up on contributions, work in progress, challenges they're facing, sentiment, everything you need to show up ready for a truly meaningful conversation. And it's completely free. Visit supermanage.a.i forward slash breakdown today to start making

the most of your one-on-ones. And thanks again to Supermanage for sponsoring the AI breakdown. Big news out of Washington this week, the FTC is investigating open AI around the potential harms of chat GPD. Welcome back to the AI breakdown. Open AI has been pretty busy this week. We saw a six-year deal with Shutterstock, which will give OpenAI access to train on their images, as well as integrate their tools into Shutterstock's platform. And they also have a deal with the AP. Still, the bigger news was that the Federal Trade Commission has opened an investigation into OpenAI around the potential harms of ChatGPT. The news was first reported by the Washington Post and focuses on a few different areas of potential harm. One of those is whether ChatGPT has appropriate data.

collection practices. A second question is whether there's been harm to consumers around false information, hallucinations, things like that that come from Chad GPT. And additionally, the FTC is also asking about OpenAI security practices. Now, all of this was included in a 20-page letter that was sent to OpenAI this week. And that letter is trying to get a ton of information from the company. They want to know about how they train their models, how they treat personal data and more. And the question ultimately is whether OpenAI, quote, engaged in unfair or deceptive practices relating to risks of harm to consumers. Now, OpenAI has talked quite openly about their belief that the AI sector needs more regulation. Sam Altman has even testified before Congress about exactly that. This, however,

represents the first time that they're potentially on the bad end of a regulatory action. And it seems he didn't super love how the whole thing was handled. Altman tweeted, it is very disappointing to see the FTC's request start with a leak and does not help build trust. That said, it's super important to us that our technology is safe and pro-consumer, and we are confident we follow the law. Of course, work with the FTC. We built GPT4 on top of years of safety research and spent six plus months after we finished initial training, making it safer and more aligned before releasing it. We protect user privacy and design our systems to learn about the world, not private individuals. We're transparent about the limitations of our technology, especially when we fall short,

and our capped profit structure means we aren't incentivized to make unlimited returns. Now, for those of you watching, what I have on the screen now is the specific FTC document that was sent to Open AI, and that was later leaked to the Washington Post. To give a sense of the questions, we have full legal addresses, identify subsidiaries, parent companies, etc. The requests ask for information about the team, how many people there are, who owns big percentages of the company, what their earnings look like. They want to know the web of relationships, whether it's partnerships or something else, the corporate governance model, all the different products and public statements thereof, and then they get into the meat of the request, which has to do

with the products that have actually made it to the public. The FTC letter requests describe in detail each large language model product with each version of each product listed separately, and they want a ton of information about each of those different products. There is a lot in here about how they've marketed products to customers, asking specifically for their biggest enterprise customers in terms of who they are, how they've identified them, how they've pitched them, etc. Question 15 is a thorny one. It says, describe in detail the data you have used during or prior to the applicable time period to train or otherwise develop each large language model described in response to question nine. They want to know how they obtain the data, all sources of the data, a list of websites

data was derived from, all policies around identifying and vetting data, how much the data varies across languages, and more. In total, there are 49 questions here, along with a request for 17 different types of documents. Now, going back to the Washington Post article, they paint the FTC as fairly aggressive as relates to consumer protection in AI. They point to an article they wrote on April 25th called federal regulators call AI Discrimination a new civil rights frontier, in which they quote FTC Chair Lena Khan, who said there is no AI exemption to the laws on the books. Now, right now the FTC is a highly politicized organization

in Washington. The Washington Post wrote, news of the probe broke shortly before FTC Chair Lena Khan faced a combative hearing Thursday before the House Judiciary Committee, where Republican lawmakers analyzed her enforcement record and accused her of mismanaging the agency. The Post also pointed out that on Tuesday, a federal judge had rejected the FTC's attempt to block Microsoft $69 billion acquisition of Activision. Some of the representatives at the hearing asked specifically about this potential enforcement action. Dan Bishop, Republican from North Carolina, asked Khan what legal authority empowered her in the FTC to make these demands of Open AI.

And indeed, how much authority the FTC has in this case is a big question. Adam Kovosovic, the founder and CEO of the Chamber of Progress, said that while it's clear that the FTC had oversight of data security and misrepresentation, it's not clear whether they had the authority to quote police defamation or the contents of chat GPT's results. Now, as you might imagine, there are a lot of hot takes on this on Twitter. Some fall in line with the Tech Oversight Project who tweets, rushing AI products to market without protecting consumers, ensuring safety, and promoting competition is reckless and irresponsible.

We applaud the FTC for taking steps to protect the American people. Senator Ed Markey basically said that this would be good in helping them get the information they needed to actually regulate the industry. He tweeted, can't combat what we don't comprehend. This action from the FTC will help us better understand the risks posed by AI technology like ChatGPT, but Congress must go farther to stop current and future harm from this emerging technology. One thing you'll note is the language here, we can't combat what we don't comprehend. He might not have meant it like this, but the inescapable conclusion is that Chat

GPT is something not that is an unbelievable example of American innovation, but which is something that needs to be combated. However, on the flip side, there was a lot of questioning around the authority that the FTC did or didn't have as related to this particular issue. Representative Dan Bishop built on his comments to the Washington Post, tweeting just before this hearing, it was reported that the FTC is investigating open AI. They're demanding information on if chat GPG generates false, misleading, or disparaging statements. Is the FTC now in the business of regulating speech? How are they claiming that authority? Now, David Krum followed a similar line, but suggested a different approach that might be better in line with the FTC's mandate. He said, in the civil subpoena to the

company made public Thursday, the FTC says the investigation of chat GPT focuses on whether open AI has engaged in unfair or deceptive practices relating to risks of harm to consumers, including reputational harm. One question asks the company to quote describe in detail the extent to which you have taken steps to address or mitigate risks that your large language model products could generate statements about real individuals that are false, misleading, or disparaging. Krum continues, it might be more fruitful to pursue an investigation into potential plagiarism and copyright violations. That seems to be more in line with the FTC's regulatory authority than policing the internet for defamation. And indeed, a lot of people felt that this was

way, way more political than they were comfortable with. Spellcraft AI founder Lewis writes, I have a lot of problems with OpenAI, but this investigation is plainly politically motivated BS. David Marcus, who was formerly at Meta running their Libra project and who is now the founder of a lightning company on Bitcoin rights, oh, FFS. We cannot become the country that is anti-innovation, innovation, entrepreneurship, and, yes, capitalism, and our value system is what made us into the greatest nation. Foster and encourage it, don't dissuade and kill it. Now, let's one be tempted to see this exclusively along partisan lines. Adam Kovacevich, who was quoted in that Washington Post piece, lists himself as a Democrat and technological optimist. He said, I've read the FTC's subpoena to

OpenAI. It's basically seeking a complete peak under the hood of how chat GPT works. It covers data security and deception where FTC has clear authority, but also copyright and defamation where it doesn't. And at the end of the day, a lot of lay folks felt similar to Ellie Dorado here who tweeted, are there any legal scholars who argued that subpoena power has gone too far? This demand letter from the FTC to Open AI is basically, please give us all the information that we need to bring an enforcement action against you. Now, what I think is going on has to do with where in the particular cycle we are. We do not yet have comprehensive AI legislation. We don't know if regulatory authority is going to be divided among different offices that seem to touch some part of it,

or if there's going to be some entirely new framework, a new department, a new division, new commission, that is created specifically to handle AI. In that type of regulatory vacuum, the thing that happens is bureaucrats and unelected officials race to fill that space with their thesis about how their agency has authority over some aspect of the new thing that's happening. Anyone who's been any time around Bitcoin and crypto over the last four years and especially the last six months has seen this happen to the extreme. And ultimately, the only organization that can solve and address this is Congress. The House and the Senate have to get together and make decisions about how new frontier technologies are going to be regulated, or else it is inevitable that the regulatory agencies will simply do

whatever they're going to do. Now, right now, we have battles around this being fought in the Supreme Court with something called the Major Questions Doctrine. The Major Questions Doctrine, which has come up in a recent Supreme Court decision, basically limits the authority of unelected officials to very specific mandates that are handed to them by Congress. And the working thesis among many in these frontier technology areas is that if they are significant enough, which certainly it would seem that AI qualifies for, the regulatory agencies can't just assume authority. They have to wait to be told that they have that authority by Congress itself. Innovation policy analysts at RSI, Adam Thayer writes, this latest FTC nastygram to a tech company reflects the short-term reality of AI

regulation in the U.S. We'll witness a lot of job-owning and regulation by intimidation through implicit threats of undefined action. By Nastygram, I mean a letter sent by a policymaker or agency that asks probing questions backed by the implicit threats of undefined future regulatory action. It's a long-standing agency practice, but one the FTC is using more regularly to influence firm decisions. The FCC was a real pioneer of the regulation by intimidation and Nastygram strategy. We called it regulation by raised eyebrow in the old days. Job-owning is the more popular phrase today, but that term does not fully capture what is going on when agencies use broader threats. The most crucial part of the regulation by intimidation playbook is the idea of voluntary

concessions, which are not voluntary at all, of course. But agencies know if they can extract such concessions without formal regulation, they can skirt constitutional challenges. Now, Adam has a whole blog post about this called the Biden administration's plan to regulate AI without waiting for Congress, and I will, of course, link to that in the notes of this document. Anyways, guys, I'm not sure ultimately how impactful this particular action will be, but I do think it's reflective of the state of AI regulation. It's a reminder of the urgency of Congress getting together to figure out what it thinks,

because in the absence of that clarity, the agencies are just going to flood in. Now, to the extent that you want, counterbalancing power thrown at AI companies, no matter what the legal precedent, this could be seen as a good thing. And I don't think that the issues raised by the FTC are necessarily things that shouldn't be addressed. However, it's hard to argue that this process, with agencies each following their own agenda, in an uncoordinated fashion, is ultimately going to be good either for consumer protection or for innovation. Anyways, guys, that is where we will wrap for today. If you're enjoying the AI breakdown,

please go share it with one or two other people that you think would find value in it. I'd love to grow the community organically, and that would help a huge amount. Until next time, guys, peace.