Artificial Intelligence’s Antitrust Concerns

about Bloomberg Intelligence for those unfamiliar. We're the investment research platform on the Bloomberg terminal, with five hundred analysts and strategists working across the globe and focused on all major markets. Our coverage includes over two thousand equities and credits, and we have outlooks on more than ninety industries and one hundred market indices, currencies and commodities. Today's topic Artificial intelligence AI. It seems to be everywhere, from simple use of chat

GPT to grow prevalence of AI across business platforms. It's hard to sit through a single news show, much less a whole day without hearing something about AI. Some say it's the next big thing. Some say it's the next version of the space race where the US and other Western nations need to show their dominance. But one thing

seems to be certain. AI and all of its surrounding issues have become ubiquitous as of recent news reports regarding antitrust implications and investigations around the technology by enforcement agencies across the globe, and all of that begs the question who's controlling the development of AI and what exactly does that mean From the perspective of competition in this developing sector, are smaller AI actors really able to get off the ground.

With all of that, I'd like to welcome my guest, Sarah Myers West, Co executive director of the AI Now Institute, the leading policy research and advocacy organization on the subject of AI. HI. Sarah, thanks so much for joining me today.

Thank you for having me absolutely so. I could spend a half hour just telling you all about Sarah's skill set, but what I'll say is Sarah spent the last fifteen years or so focused on the role of technology companies and the growth of their political influence, and her current

role is as co executive director. Sarah focuses on addressing the market issues and infrastructure that guides the role of tech in society and holds a sector to task in serving the needs of the public, including a relevant to our chat today antitrust issues. Her research is featured in leading academic journals and a wide range of media outlets, and she's a go to for governments on questions surrounding tech and AI, having testified before Congress and advised the

White House, European Commissioned CFPB, and other agencies. Also incredibly important for today's discussion. Sarah recently completed a term at the FTC as a senior advisor on AI. She worked on competition and consumer protection issues, and last, but certainly not least, Sarah's also working on a book, Tracing Code, which we'll examine the origins of data capitalism and commercial

commercial surveillance. Sarah, There's so much more I could say about you, but there's obviously a lot to talk about in this space, so let's get to it. We're recording this on July nineteenth, twenty twenty four. Obviously some really big news today considering the Microsoft out it just wondering if you have anything you'd like to lead off with on that subject.

small handful of players. In this instance, it was Microsoft and CrowdStrike, where the two companies kind of behind this, you know, massive technological outage that uh, you know we're all experiencing today.

I think have really captured everybody's in genations today. And you know, the definition of what AI is in the present moment, is in many ways constituted by it's defined by an existing concentration of power within the tech industry. So if you wind back the clock maybe twenty years, you see these tech giants Google, Amazon emerging on the scene and developing out business models that involve the collection of massive amounts of data and the development of huge

computational infrastructures needed to process that data. And once you had those two big pieces in place, these companies began to develop new techniques of trying to process that data to make sense through analysis. Often it's you know, sort of basic statistics, but sort of combining massive data sets massive computational infrastructures. And those are two of the core elements that you know, make up what we call AI today.

The third is the field of research and the the big tech firms also began to dominate the leading conferences in artificial intelligence. The latter their corporate labs became the place that computer science graduates wanted to go get hired. They were offering the highest paying salaries. There became the most prestigious places to go and work. And so between the sort of business elements of what kind of AI was being developed within these firms as well as defining

what constitutes the cutting edge in the research space. Big tech firms I think have really significantly shaped our understanding of AI, what it can be used for or what is most of interest over the span of at least

a decade and probably more. And so getting back to our subject today of thinking about antitrust, it really brings to question, you know, if we were to intervene in the AI market in a more structural way, not only sort of like tackling the the you know, concentration in big tech, but they're shaping hand in shaping research and

development as well. I really wonder how that changes what kind of AI we're building and what vision it serves, because to my thinking right now, it seems somewhat stale compared to what you know, we were being sold a year ago from companies like open Ai. Like I find it kind of hard to get all that excited about a chatbot and more excited about the vision of what could be possible if we were to to break up big tech.

influence where AI goes. It just so is so dependent on the data that it's taking in, if you will, And that really nothing nothing regarding AI is inevitable or has to be inevitable. And I think with those two kind of points in mind, you know, what do you really see at the most pressing antitrust considerations right now from maybe a company standpoint or problems that have really emerged recently that you think really are right for a challenge.

and see. Maybe we're going to see startups starting to bubble up, and you know, we'll have new entrants that are going to be competitive with Google and Microsoft. But now I think we're seeing a recognition that this heavy concentration in these firms and their control over the data, their control over the infrastructures are presenting real barriers to innovation and competition in the market. Now there's a long

way to go on actually meaningfully tackling that issue. You know, the DOJ just I think maybe three weeks ago, announced that it's going to be looking into in Nvidia, which is the maker of the chips that are powering a lot of AI development. The French authorities are a little bit further along. They conducted a ra aid on in videos headquarters last year. But I think that infrastructure is only one part of the problem. These firms also control crucial paths to market, and that I think is still

an undercovered antitrust issue. It's why we see promising startups like Mistral striking deals with dominant firms like Microsoft, because these new entrants need help connecting to their customers. There just isn't really a clear business proposition for a lot of new startups. And we've seen firms like Goldman Sachs and Sequoia, firms that have themselves made big investments in this market issue increasingly skeptical reports about the potential of

new players to compete. So I think addressing those two pieces, the role of the big firms in dominating the resources that you need to build AI, the infrastructure that you need to build AI, as well as their role in shaping the market it through their dominance of the ecosystem of the path to you know, to profit is key.

you know, it's really fascinating. I think, you know, from a standpoint of what the FTC or DOJ can do in the context of traditional anti trust enforcement, right, you see these situations where perhaps making a large investment in a smaller AI startup, if you will, or hiring most of that that startups you know, labor folks or skilled labor who you know has the brain power behind AI as a way to kind of get around that traditional concept of a heart Scott Redina review or something like that.

You know, where it would be a traditional full skill out acquisition. It raises a lot of questions about what can be done from an enforcement perspective, and you know, we'll get to that a little bit later on, you know, in terms of legislation and you know what what might

reshape the way things are proceeding right now. But you know, I think another issue that you know, I'm hearing a lot about from my side here too, is you know this concept of this AI you know space race, you will, right, the US has to compete with other global entities that are also developing AI. And you know, no surprise, I think in political rhetoric we hear you know, China comes up a lot, and this purported need for the US

to compete with China on the subject of AI. How do you view that And do you kind of think it's this red herring if you will, in some ways, as you know, you know, as a reason not to really reiin in AI from an anti trust perspective.

a new anti trust rules for the industry. And you saw prominent people in the Defense establishment, any of whom we're also on the payroll of Eric Schmid adjacent firms out there saying that if you do anything to restrain big tech firms, it's going to diminish our competitiveness with China. And I think the Biden administration, to its credit, clap back by issuing this Executive Order on Competition that said so, actually,

national monopolies are not in the national interests. What we want is a really vibrant and competitive sector, and that's what's going to set us up for a success. But I don't think that other parts of the government have fully lived up to that vision. I mean, certainly, as you know, Congress hasn't really meaningfully moved, and in part that's due to fears about the US falling behind on

innovation if it regulates the market. But I think that's misguided, if anything, regulation of a sector that's gotten this bloated would really foster more innovation and would benefit a wider public.

the hill. There's there's no question out that. And you know, I think, you know, building upon the rhetoric around you know, China and this just this idea of an arms race, you know, I think there's this other ick factor, if you will, sometimes AI where people think about how that might implicate things like biometrics, surveillance, right, how does that roll out in the future, and what industries is that does that become an issue and if you have any

thoughts on about that aspect of AI too. I think that's something folks are super interested in right now as well.

in their use places like in cars. For example. You know, there are if you're a delivery driver, one of the ways that drivers are being monitored by their employers is through the use of biometrics in car can that purport to be able to evaluate you know, whether they're falling asleep or whether they're you know, they're signaling aggression. And that use case of what we call emotion recognition is

one of the areas that I'm especially worried about. Emotion recognition systems have largely been disregarded by the scientific community as pseudoscientific. You can't sort of with any statistical validity take you know, the measurement of somebody's facial expression and be able to evaluate what their interstate looks like. But that certainly doesn't stop a company from selling a system that claims to be able to make those kinds of assessments.

And we're seeing the rollout of emotion recognition in a lot of workplace surveillance contacts. You know, mentioned drivers, call center workers are having their voice print measured and they're getting feedback about, you know, whether or not they're sounding pleasant enough to their customers, and all of these other places where often they're used in ways that have quite

significant consequences for the people that it's used on. So I think that biometrics are we remain an undercovered area and one that we really need much more policy scrutiny of.

If it continues to roll out the way it does. Sounds like so many, so many flags I think on the play with that one. But you know, kind of turning back a bit to actual enforcement or the regulatory oversight here with anti trust. You know, so we see all these these reports of investigations going on. I feel like several times a week I'm seeing something about investigation happening either in the US or abroad, you know. I know you mentioned the French investigation of Navidia, for example.

But what do you see as that kind of problems right now with the available methods for oversight, you know, and the way that regulators are going about employing their oversight or investigations of these entities.

They're also in an unenviable position of trying to play whack a mole after the fact with companies that are so big that they're very comfortable sort of experimenting out in the wild and just seeing, you know, how these technologies are going to work, and you know, have people

give them feedback on where they fail. And so like the the enforcement agencies and investigative journalists and independent researchers are all like playing whack a mole, trying to find and surface harms long after they've occurred, and then you know, attempting to use the law to remediate harm. But in some cases it's it's way too late. So I think that that's you know, far from an ideal mode of regulating a sector that has such significant consequences for the public.

And there's you know, limited incentives for companies that have

grown this large to really comply. You know, for a lot of these firms, they just allocate a budget line for the finds that they're going to get hit, sure, sure, and so we need stronger incentives for compliance and stronger incentives for firms to be you know, engaging in basic levels of you know, testing and validation of their systems even to make sure that they work as intended before they roll them out, let alone, make sure that they're not causing harm.

I draw a circle and say, hey, you've got you know, five stores, and that's really concentrated relevant market here. The merger is not good for this particular reason, right, But obviously AI, you know, it's a different scenario. What kind of thing should folks be looking looking at around merger activity? Around AI? Should we actually start seeing you know, real acquisitions in that space.

while at the time was one of the largest acquisitions in tech history, the Facebook's acquisition of WhatsApp. So, you know, the use of data obtained through acquisition can lead to competitive harms. It can also lead to privacy harms. I mean, certainly for the users whose activity was being monitored. There were similar concerns raised.

FTC was evaluating that acquisition. Uh FTC commissioners Bedoya and Slaughter issued an opinion saying, you know this, this merger is going to go through, but we do want to raise some concerns about the new pools of data that Amazon's now going to have access to, because they're now going to have access to sensitive health data that may not you know, violate privacy rules, but nevertheless, could let Amazon make inferences about people in ways that will will

be highly consequential. So that kind of activity, I think is one thing that deserves scrutiny. And also what you raised earlier around you know, these sort of non traditional paths to not acquiring companies but making these unusual business arrangements through provision of compute credits or things that don't reach a Heartscott Regina threshold sort of and runs around traditional acquisition are the other that we're seeing a lot of in the sector.

Where where what's happening, or what's alleged to be happening, is that you know, competitors are acting in kind of this hub and spoke you know, conspiracy where they're sharing their private information with a third party but also getting access to their competitor competitors information when they by virtue

of them doing that. But you could talk a little bit more about those lawsuits and you know what, what other kinds of suits maybe do you expect that we might see coming up regarding you know, AI and its uses in those contexts.

wages that they'll pay to workers. Viena Duball has really documented really well the use of algorithmic wage discrimination in the gigwork sector, and what that means is that a lot of people are being squeezed on both sides that that the firms, you know, on the one hand, are pushing prices as high as they can get people to pay and trying to target them to pay as much as possible, and then on the other side, paying them as little as they possibly can, trying to you know,

utilize consumer surveillance data on both ends, and the firms that have the ability to take advantage of information asymmetries are the ones that benefit. So, like you've said that, we've seen these cases focusing specifically on the use of algorithmic methods to set rents, and this is problematic for many reasons. You know, we already have a real affordable housing problem in this country that's exacerbated by the existence of monopolies in housing markets and cities around the country.

And what this what this does is it essentially extends cartel like behavior into markets where it otherwise couldn't extend. It would be really difficult for rental companies in Houston and Chicago to coordinate with one another on setting prices, but through systems like you know, those provided by real Page, it creates sort of this hub and scope, hub and spoke model across all of the firms that are relying on this one company. There are a couple of bills

that also try and tackle this issue. Senator Klobuchar has one. Senator wide, it has one, but like you've mentioned, nothing's really moving through Congress right now. So what that means is these lawsuits are really the front lines of trying to tackle these problems in the absence of regulatory movement, just doing the best that they can with the law we already have on the books right right.

can with existing law. But you know, to your point about this legislation pending in Congress, you know there's no shortage of anti trust legislation that that seems to be pending on the Hill right now. I know you testified regarding these bills before Senator Klobashar's subcommittee last December, but you know, beyond just those those bills you just mentioned, what do you view as the most pressing legislation that kind of needs to get over the finish line right now in DC?

law is AI policy. What we saw in the EU after chat GPT was released is the swiftest action that regulators took took place under the GDPR, the EU's privacy law. We saw, you know, the EU be able to clap back on Uber and Lift's use of algorithmic pricing methods and algorithmic wage setting also under the GDPR. So I

think that's a really good place to start. The other place, also, non conventional but I think reforming our labor law and whistleblower protections because so much of what we know about AI, which is often a very opaque sector, and many of the early policy wines that we've seen have come from movement by workers, whether that's organizing or whistle blowing, or you know, codifying restrictions on how AI can be used

through collective bargaining. So I think protecting workers is the other place where I think we can do a lot more.

those lines. You know, some of that legislation does seem to have some bipartisan support, you know, And I think you might have alluded to this a little bit earlier, you know, when we first started speaking too. But you know, what are your view? What's your view? I'm perhaps withstalling this legislation and if you could talk a little bit about, you know, just the influence of tech and D see today.

I think that's also something that you know a lot of folks might not be considering as part of what's happening with the legislative angle as well.

So big tech lobbying is very prominent and it's gaining traction as you know, the industry tries to push the narrative that AI is too complicated for a government to regulate and that we need to trust them to determine next steps instead.

We're also seeing new constituencies emerge, and I think that we've seen this particularly come to the four with the elections that there's you know, on the one hand, this existential risk lobby, which was very effective in shaping a tech agenda focused on like this sort of doom and gloom vision for AI, and then on the other hand, this effective accelerationism lobby, which is, you know, promoting unrestrained development.

That's the Andres and Horowitz and y Combinator and sort of they've established their their presence in shaping sure Trump's AI agenda, and you know, with those two camps, I think it's it's really important to make sure that the scope of the agenda, like the vision for what we hope to see with this technology, doesn't just get narrowed to that range of interest from Silicon Valley, because I think that there's a lot more that we could build and see, but that the industry's version has sort of

benefited only that narrow set of actors. And that's where I hope that stronger anti trust enforcement will give us fresh perspectives that we can really sort of like revitalize this industry and make sure it's serving all of us.