Is AI harvesting your knowledge on the cheap?

0:00This BBC podcast is supported by ads outside the UK.

0:29CFO. LinkedIn has a word for that. Bull spend. Now you can invest in what looks good to your CFO. LinkedIn ads generates the highest ROAS of all major ad networks. You'll reach the right buyers because you can target by company, industry, job title, and more. So cut the bull spend. Advertise on LinkedIn. The network that works for you. Spend$250 on your first campaign on LinkedIn ads and get a 250 credit for the next one. Just go to linkedin.com slash broadcast. That's linkedin.com slash broadcast. Terms and conditions apply. How did Russell Brand become a Christian influencer? I'm Tristan Redman.

1:16And I'm Asma Khalid. And we're the hosts of the Global Story podcast from the BBC. Brand is facing charges of rape, which he denies. Over the years, he's been a stand-up TV host, author, Hollywood film star, podcaster, and now a new religious incarnation. Listen to The Global Story on BBC.com or wherever you get your podcasts.

1:42In some privacy policies, there was a little note that the car might be collecting information about your sex life. What? Welcome to The Interface, the show that decodes how tech is rewiring your week and your world. I'm Nikki Wolfe. I'm Karen Howe. And I'm Thomas Germain. Today on The Interface, the frightening reality behind the AI industry's race to consume the world's knowledge. Hantavirus starts up the viral fear machine again. And your car is spying on you, and it's so much worse than you think.

2:22So, Nikki, Tom, you know, we've talked before about the hidden world of human labor behind AI. Sounds familiar. We've also talked before about the impact of AI on work. Well, this week I wanted to tell you about an investigation I published with More Perfect Union, which sits at the intersection of these two trends, and it kind of really freaked me out. So AI is coming for your job, but not exactly in the way that you think. So don't listen to what other people say. Listen to what I'm about to say. Listen to Karen. Yeah. I've been saying this for years. So I have been covering the hidden world of human labor behind AI for years, for around eight years now.

3:10And generally speaking, I go to global majority countries to meet with workers in really impoverished contexts, communities, who are being hoovered up by the AI industry to train their AI models. So this kind of work is content moderation. It's cleaning the data that they use to feed into their models. It's also annotating the data. And this category of work is generally called data work. It is literally the work that the AI industry needs to make their technologies exist. the shocking thing that's happening now is that this kind of work has become the number four fastest growing job in the u.s according to a linkedin report from earlier this year in the domestic u.s as well as in the domestic u.s yes like i'm talking about u.s based people being hired up as data workers now.

4:10These are not any kind of workers. These are extremely highly educated individuals. They are people with college degrees, master's degrees, PhD degrees, law degrees, and also people who have been working within different industries as scientists or lawyers or doctors even for years that are now turning to data work as another job. And part of the reason why this is happening, this is coming in the context of AI also being a reason why the labor market has become a lot worse for people. It is slowing down hiring. It is accelerating layoffs across different companies. So whereas the AI industry tells you that AI is coming for all work and no one will have work anymore.

4:59And this is basically just another way of hyping up their technologies to say our models are capable of all work. What's actually happening is this vicious cycle where, yes, companies are using AI to justify layoffs, to justify not hiring workers. And those workers who don't have the full-time employment opportunities are now getting fed back into the AI industry to work for the industry itself as data workers. So it's not taking away your job totally, the industry is actually in shittifying your job and turning it into a gig work version so that they can take and consume your expertise and your knowledge.

5:43And are these long term jobs or is this literally they do that job until they have trained the AI and what it can do and then that job is gone? So, yeah, this is a kind of a complicated question because no, they're not long term jobs in the sense that these are contract-based gigs. Yeah, there's no workers' rights. These aren't pensions. Yeah, exactly. It's not short-term because the job goes away. It's short-term because of the nature of how the industry is structuring the data work. And yeah, these are incredibly awful jobs to work. I interviewed a bunch of data workers for this story. And And one woman was an Ivy League PhD graduate who then literally applied to 200 jobs and didn't get any of them.

6:30So ended up in data work. And she described this extremely piecemeal, anxious process where she's drip fed this data work and an email will just arrive in her inbox and be like, a project is available for whatever,$35 an hour or something like that. Do you want it? And they'll have all these terms like, oh, it's going to be for potentially a four week long project or this is how many hours you should be able to work, so on and so forth. But that never actually happens. When she picked up her first project, it started and basically lasted for two weeks and then immediately like overnight disappeared.

7:09And she was at her university about to attend her graduation when the work just poof went into smoke. And she thought to herself, should I even celebrate my graduation anymore and like get this nice dinner because she couldn't afford it. And this is kind of how this work happens is there. You never know when it's going to arrive. You never know when it's going to go away. And it kind of pits different workers against each other because the people who claim the work fastest are the ones that actually get the opportunity. What exactly are these PhDs doing? What knowledge are they imparting? What's the work?

7:53When you think about everything that an AI model can do, companies want you to believe that they magically learned it themselves. But actually, there are humans involved in teaching literally every single possible task. And so when you think about the fact that ChatGPT can chat, it's because there were people that showed these models, examples of dialogue, person A talks, person B responds, person A talks again, and so on and so forth. And as companies have gone from simply just getting their models to chat about anything to these more sophisticated tasks where they're trying to sell this idea that their models are going to arrive at PhD level intelligence, that is part of the reason why they're now, yeah, Yeah, that's why they're now motivated to hire PhD-level workers to try and teach this knowledge to the models.

8:48So they're training it to sound like PhDs. They're not giving it an actual PhD. They're training the models to have the trappings of PhD-level intelligence. They're giving it PhD vibes. So they're making them more annoying. Yeah. So because, just to make sure I'm getting this right, because the state of the economy has made it harder to get these jobs that require specialized knowledge, people are out of work, and then there's a new kind of work. You have to be an expert to get it, but it's low-paying. It's unreliable. You never know when it's going to come. It disappears overnight. And it's like this new class of, you know, Uber Eats-style workers, except they have PhDs.

9:39Yeah, exactly. And it's not just the state of economy for any reason, but specifically because of AI that the state of the economy is this way. So in corporate investor calls, CEOs are talking about right now, the reason why they're accelerating layoffs and slowing down hiring is because they are experimenting with AI. This goes back to a previous episode where we were talking about how Meta is laying off some of their employees as they're also really pushing to introduce more and more AI into the workforce and how they were making this future bet, essentially, that the technology is going to ultimately work out and going to allow them to not have to rehire these workers back.

10:25And a lot of other companies are making the same bet. It's not necessarily that they are literally seeing dividends with the adoption of AI right now, but they're hoping that these models will improve to the point that they don't have to recall these workers back and they can shrink the workforce in the present day to gain some of those financial efficiencies. And so this is the moment that we're in where it is the very industry that is causing this cascading effect on the economy that is then getting to profit off of the creation of an underclass. Because these people are available for this kind of work.

11:06Yeah. Interesting. Yeah. You would think that this vicious cycle is going to lead to an end state where all of these workers are laid off from full-time jobs. They then train these models to have all of their expertise. And then they just all basically by the end, there's no work left for humans. But the workers were telling me that the way that the data work is organized is so chaotic and so poor that the AI models are actually not very effectively capturing their expertise. They, for example, I talked with one PhD graduate who he was mentioning that he has a PhD in philosophy, but then he was overseeing this team of other PhDs in math, chemistry, biology, whatever.

11:59And none of them were actually evaluating tasks within their domain of expertise. He was being asked as a philosophy PhD to evaluate math stuff. Whether or not these jobs are ultimately replaced, right? Like maybe it's difficult for people with PhD level education to get a job right now. The fact that those people who've dedicated so much time with the expectation that they were going to achieve high incomes, I mean, depending on what the field is, right? Like, those people are taken out of the labor force. And, you know, the economy is built around, you know, different classes of people making different amounts of money.

12:41If there's suddenly this shift, you know, even if it is temporary, the effects could be pretty long term. Yeah. I think this really hits on a central point. I was interviewing MIT Institute professor Darren Asimoglu, who won the Nobel Prize for Economics in 2024. And he was saying that if Silicon Valley gets what it wants and successfully Uberizes most knowledge work, then the kind of inequality that we would experience is something that we would have never seen before, where most workers are sidelined for meaningful work and only a few companies actually employ most workers as well. And this would, I think, fundamentally break the social contract.

13:26Yeah, why would anyone then pay tons of money to get training in college in a PhD program in a law degree when this is what is awaiting for them on the other side? So it would absolutely lead many more people to reconsider whether they should be getting higher education at all. And that in and of itself, whether this moment is temporary or longer term, would then have further consequences down the line. So is there any kind of solution here? So I interviewed this guy named Tim Newman, who heads labor programs at a policy org, policy nonprofit called TechEquity. And one of the things that he points out is that when you think about the full AI development supply chain, workers are a crucial part of every single stage of it, not just data work.

14:17Also, you know, the people mining minerals, the people designing computer chips. What he was pointing out is that a key way to govern the development of AI is actually by strengthening worker rights across the economy in general, strengthening the ability of workers to unionize, to be able to hold their companies and their industry accountable through collective action. And specifically when it comes to data workers, they don't actually have these formal unionization efforts. There are actually efforts on the way in process. I talked with this other researcher named Crystal Kaufman, who was formerly a data worker herself.

15:01And she has been for years actually now organizing data workers to form these more collective groups that can engage in this kind of action. And finally, there is, when it comes to just improving the conditions of the data workers themselves, there is this new interesting bill that has been introduced in California that Tim told me about called the Sweat-Free AI Procurement Act, which borrows from the fashion industry, which also had a lot of labor exploitation in that industry, and then used different mechanisms by which to shore it up. And what this bill does is it would require the state of California to, whenever they use any kind of AI model within government work, to ensure that it was not built on exploitative labor conditions.

16:00And this is a model that could potentially be propagated to institutions, to universities, to other forms of government as a way to create a minimum labor standard across the AI industry.

16:17Where did the obsession of some U.S. administrations with Iranian regime change come from? It may be older than you think. I'm Tristan Redman, host of The Global Story from the BBC. It happened once before in 1953 when the CIA led a coup that tried and succeeded in toppling the Iranian government. And many Iranians haven't forgotten. For more, look for the global story on BBC.com or wherever you listen.

16:49All right, so coming from that story, Karen, I have a story that is going to sound also kind of like a downer, but actually has a hopeful note that we'll get to. So it's okay. So it's okay. Oh, okay, good. Well, I'm not going to worry the whole time then. I want you guys to go into this story with, in your mind, everything is probably going to be all right. Okay. Have you been following the story of this cruise ship virus outbreak? Hantavirus, it's called, right? Am I saying that right? Hantavirus? Yeah, Hantavirus. On a luxury cruise exploration ship traveling from Argentina to Cape Verde, they've had an outbreak of a virus called Hantavirus, which is, it's a deeply unpleasant virus.

17:37It's got a mortality rate higher than 30%. I think it's around 35, 40%, which is horrendous, unbelievably dangerous. It's a virus that originally comes from rats. It's hugely dangerous that people are exposed to it. But in terms of the kind of global pandemic that obviously this has triggered everyone's absolute terror response, it does not spread in the same easy way that the coronavirus did. All scientists have agreed that basically it's very unlikely that this turns into, and very unlikely, extremely unlikely, there are some factors of this virus that make it not the kind of virus that turns into a global pandemic, right?

18:29So the situation on the ship is very serious. three people have died. But then this spun out all of these viral hoax, COVID-26, people were calling it, another pandemic, people were calling it, Chinese bioweapon, people were calling it, it reignited all of these beliefs that people can project onto it. And one of the things that we saw during the coronavirus pandemic was this backlash against guidance of trusted experts. If there is another global pandemic in the style of the coronavirus, what will make it disastrous is this social media content that turns it into a conspiracy theory and that makes people refuse to get vaccinated, that makes people refuse to follow guidelines, that makes people refuse to wear masks, that makes people refuse to self-isolate.

19:32The information ecosystem is damaged and this hantavirus outbreak is showing us that it is still damaged in a way that we really need to address because when the next actual danger happens, and it might not be a pandemic, it might be something else, we have a real information ecosystem problem. I want to read you this quote from Katrine Wallace, epidemiologist. So when a future outbreak with real pandemic potential eventually emerges and one will, millions of people will encounter it inside an information environment already primed to distrust public health guidance before it even arrives. The narratives are pre-written now.

20:17The audience already knows the cues. that I think perfectly sums up what the problem is here. So it's like we're getting a trial run almost, right? This virus breaks out on a boat. It's very scary. It kind of sounds like the kind of thing that could become another global pandemic, but it's not. That isn't going to happen. People are misinterpreting it. And yet, because it sort of sounds that way, people jump straight to, oh, they're lying to us. They're making up an imaginary virus. And pretty soon it will be another COVID and we'll all get locked down under false pretenses. Not going to happen here because this virus, it is real, but it's not going to spread that way.

20:58Right. But once we get there, the gears are already ready to start turning. This is fake. It's not happening. And here's what's going on here. Here's the reason those gears exist. Is the content that gets the most engagement is the most monetizable. And the kind of content that gets the most engagement is fear and anger. They don't want you to know this is one of the most powerful kind of intros to internet content, right? Yeah, it's funny. They're trying to scare you is what we're hearing when, in fact, the opposite is true. Experts are like, no, no, no, you shouldn't be worried about this. It's not going to be a thing.

21:35And people are doing, you end up with kind of ivermectin again, which was this kind of nonsense thing that happened at the beginning of the coronavirus that was a sort of snake oil to cure. coronavirus, people are selling things based on this fear. Because again, if you're afraid, you'll buy stuff. It's a very powerful capitalist imperative. So you had alluded to earlier this idea that part of the reason why we are seeing this frenzy is a bit of a trauma response. There are people who are genuinely afraid that this is in fact going to be the next pandemic, falsely so. and also you mentioned that there are people who are just trying to profit off of this fear.

22:22So when talking about this machine that spins up, like how much of that machine is people actually just trying to manipulate information and cash out and how much of it is average people that are genuinely scared and inadvertently amplifying the conspiracy? It's probably both. Yeah, it's definitely both. I think the number of people who are like rubbing their hands with glee and saying, oh, this is another thing that we can make money off people who are afraid is less than the number of people who have big accounts and are afraid and are making content that just through the process of the platform are monetized, right?

23:11The monetization of content on these kind of platforms is quasi-automatic. I don't know, Tom, if you can speak to exactly how the kind of monetization works. Yeah, I mean, you get big enough on certain social media platforms, essentially, they just start paying you, right? If you have a certain number of followers, you're getting a certain kind of engagement. The rules differ depending on what platform we're talking about. But like the incentive structure, whether you're lying and trying to make people upset is the same, right? You want to make a piece of content that is sensational, that gets a lot of attention.

23:48If you're very worried about something sincerely, you still want it to be sensational and scary and attention grabbing. If you're trying to trick people, you do the same thing. The algorithm treats it the same way. And because these systems are in place, this kind of conspiratorial mindset is ready to spread at any given time because that's what the system is designed to do is to spread this sort of idea. It's really difficult to track because we're now in a short form video algorithmic dominated system. Yeah. So you have to be within these kind of ecosystems in order to see some of this content.

24:29So it's like another filter bubble problem that probably I would assume that makes this harder to address, right? Yeah. Filter bubble is exactly the right term. We're talking about everyone's social media and information ecosystem experience is unique to them and kind of impossible to see into from outside. Like for a little while, I as a reporter, and I don't know if you guys have tried this as well, but made specific accounts with specific interests just to try and see what kind of content gets fed to you. And it is completely different from, you know, people are living in these self-contained information worlds.

25:14So what could you possibly do about this, right? You've got this community of people who are like the thing that they distrust is expertise, right? That like they don't like a bunch of quote unquote experts speaking down to us, telling us what we have to do and what we have to believe. How do you address this? I mean, this is a colossal problem. And it's not just an outbreak like this where we're seeing what kind of problem this is. We're seeing massive resurgence of measles, for example, in a lot of places where it had been essentially eradicated because people are refusing to get vaccines, refusing to get their children vaccinated.

25:57This is a ginormous problem for healthcare because if people don't trust the very concept of vaccines, that by definition is rolling back, you know, a century of extremely brilliant kind of medical advances. How do you convince people when in a lot of other ways, governments have really not earned their trust? It's the incentives online that make this kind of fear and conspiracy, fear-mongering information spreadable in this kind of way. But they're not based on absolutely nothing. It's an information ecosystem problem, and it's a governance problem, and it's a media problem. There's all of these things happening at once.

26:53And they all feed into each other, right? They all feed into it. One makes the other a bigger problem. The people are primed to feel this way. Right. And then social media is ready to pick that up and elevate it and make it bigger and spread it to more and more people. Spreading more distrust, the feedback loop continues. Right. During the COVID-19 pandemic, this was obviously a huge issue. And one interesting thing that we could look to history for is how the social media industry tried to address it. So there was a period where, you know, institutions of power, you know, within business and within government and, you know, other parts of the society were all kind of on the same page.

27:37Like we have to do something about this misinformation. And all the social media platforms, all the big ones anyway, launched these programs for different ways that they were going to address misinformation, health misinformation in particular. So like a lot of accounts that promoted COVID conspiracy theories got banned from YouTube and from Facebook and Twitter, right? Or they did these things where they would add these labels where they would have like fact checking, where you'd post something about COVID that wasn't true. And then there'd be this little thing underneath it that says, actually, experts say that this is a lie, but we're leaving the post up because we don't want to censor people.

28:12It was a huge effort on behalf of the tech industry. And it caused this big backlash, particularly among conservatives and people, regardless of their political ideology, were buying into the conspiracy that big tech was censoring us. And because of that, basically all those efforts have been abandoned. The social media industry as a whole basically said, we're not doing this anymore. We're not doing the fact-checking thing. We're dialing this back. This was a mistake. Mark Zuckerberg apologized for what he called, you know, like errors in judgment, the way that he addressed this misinformation.

28:57Things might change if we had a problem that was as big as COVID again. Maybe the political winds would blow the social media industry back into addressing it. but they never really figured out an answer. Yeah, I mean, I would argue that social media actually has long figured out the answer. You know, there's been a lot of investigative journalism that has found internal documents within all the social media companies that show that they've done internal tests, that have determined that the way that their algorithms incentivize information to flow is the core problem. And yet, they would never change that because that is also the crux of how they make money is by spreading information in these kinds of ways.