Feeding the machine: the hidden human labour powering AI
01:28:25 Share

Welcome to the LSE Events Podcast by the London School of Economics and Political Science. Get ready to hear from some of the most influential international figures in the social sciences. All right. Welcome, everyone.

So we're so excited for this hybrid event this evening. This is an opportunity to discuss the rich and wonderful book, Feeding the Machine, The Hidden Human Labor-Powering AI. So I'm Kate Redenberg. I'm an assistant professor in the Department of Philosophy here and a UKRI Future Leaders Fellow. I'm joined by two of the three authors of the book.

So here we have our guests Callum Kant, senior lecturer in management in Essex Business School and James Muldoon, a reader in management in Essex Business School. And we're also very fortunate to be joined by Kirsten Senbrook, professor in the International Inequalities Institute here at the LSE.

And, you know, one aspect of her work, a reason that we're very, very lucky to have her to comment on the book today is her work on conceptualizing and measuring quality of employment, especially in the global south. So first, some logistics. Everyone loves logistics.

So following the discussion, there will be a chance for you to put questions to our speaker. So if you've brought pen and paper, please write down some questions so you don't forget them. For those of you joining online, there'll be a chance for you to ask questions as well. So please include your name and affiliation. If you're asking a question online, they'll come up to me with this high-tech iPad and I'll be able to ask all of our speakers.

So for those of you in the theater, of course, I'll indicate when I'm opening the floor for questions. We aim to have at least half an hour for questions, and we really want to provide you enough context in order to discuss this really rich and insightful book. And as well, right when we get to questions, there'll be a microphone. The microphone will come to you. I will call on you. Very straightforward.

I'm required to remind you to please put your phones on silent. Please avoid shouting in the middle of the lecture and otherwise making disruptive noise. Also, you should know that the event is being recorded and will be available on your YouTube channel. If you ask a question, you and your question will show up. We will include the questions as well, so just be aware of this.

There's no fire alarm scheduled, so if there's a fire alarm, there's a fire, and we all need to leave in an orderly fashion. All right, with those logistics in place, we can turn to our panel discussion of Feeding the Machine: The Hidden Human Labor Powering AI. So, it's really excited to be able to chair this event because this is such an important and particularly a timely book

that really uncovers all of the hidden human labor and capital that goes into building as well as deploying and maintaining AI systems, right? So when you interact with ChatGPT on your phone, you've got this lovely interface. You wouldn't really guess that there's this whole kind of data supply chain, huge series of human labor that goes into this.

And in particular, right, I think we all have reason to care about what this book discusses, the various inequalities that AI creates and perpetuates in the workforce. We have reasons of solidarity, right? We have reasons to care about the conditions of workers in the global south and the global north.

subject to unjust working conditions. We have reasons to care for all of us as well, self-interested reasons. So as we'll get to, one of the themes of the book is also algorithmic management, the ways in which AI systems are used to manage all of our work. This probably has already come for you. If not, it's coming for you. We have reasons to be concerned, and as we'll talk about in the end, right, we need to think about ways to resist these trends. All right.

So, what I'd like to do, because this is such a rich book that touches on so many themes, is turn it over first to our authors, James and Callum, just to give them a chance to give you a little bit of an overview into the book,

and then we'll dive into some specific questions for our panel. So James and Colin, please. - Great, do you want me to start off? First of all, thank you so much for attending this event, Efrain. It's always such a pleasure to share your work with people and we're really delighted to have Kirsten here to help comment on the book and also to learn about her research and thank you to Kate for moderating this session.

We usually start talks like this by talking about one of our characters in our book, one of the research participants. And her name is Mercy. And Mercy works as a content moderator at an outsourced center in Nairobi. And she is looking at social media posts and looking for kind of toxic content.

One day when she's working, she sees a fatal car crash that has been filmed. And she kind of looks closer at the front seat and realizes that one of the victims is her grandfather. And she actually learns about the death of her own grandfather by moderating toxic social media posts.

what's more, when she kind of gets up from her seat and screams and tells her supervisor what's happened, the workplace where she's at, the conditions are so bad that her supervisor suggests that she should finish the rest of her 10-hour shift. So for the next few hours, again and again, she sees...

video footage of this images text description of this incident that she has to continue to moderate. And when we heard about this story, in a must have been like a five hour interview session on a Saturday afternoon at a cafe in Nairobi, we thought, this is a story that everyone needs to hear, right?

And again and again, what I've found is that even people working in AI, even people, like I give this talk sometimes at like AI conferences, and like more than 50% of the people in the room have no idea this is taking place. So what we're really trying to do in this book, Feeding the Machine, is take you behind

the kind of smooth surface and the slickness of the technology to kind of show you the hidden global production networks that make this possible, right? So the supply chain view of AI, how does it work? So in the book, we actually explore quite a few different themes and we've structured the book so that every chapter takes the perspective of

of a different worker in the supply chain from someone who works at a data center to a data annotator and even the kind of manual laborers and intellectual laborers that also feed the machine so to speak so where you look at void vocal actors artists people who are writing books and films and paintings all of this that goes in to the data sets that make ai possible so it's a

It's a riveting read, in my opinion. We do try to kind of have a kind of academic trade book crossover type thing. And what we'll be focusing on today is the more data annotation side of this. So what is that? It's basically like people who are drawing bounding boxes or little squares around blocks.

pictures of like people, maybe a street scene, so that autonomous vehicle software can learn how do you see a road, how do you kind of picture, how do you differentiate a person from a tree, that kind of stuff. It could also be textual analysis, right, like grading answers from chat GPT, that's a kind of data annotation as well. Anything that involves curating, cleaning up, making...

our lives legible to computers, right? Making the data sets able to conceptualize all of this kind of work. And what we really wanted to show first of all was the incredibly exploitative nature of these systems, right? The kind of grueling conditions

that data annotators face in some of these centres. So the fieldwork that this was all based on was at a company called Sama, previously called SamaSource, actually started out as an NGO and then kind of transitioned to a for-profit business. And they thought of themselves as a social enterprise, right? They actually were so...

about conditions in their East African centres that they invited us in. You know, they had a negative Time magazine story published about them. You might have seen it. Workers getting less than $2 an hour under appalling conditions. And they thought...

Look at these good chaps here from the University of Oxford, where technically we were associated with at the time with our PI, Mark Graham. Look at these good chaps. They're going to tell everyone that that nasty man from Time magazine was wrong. And actually, our workers are basically working in the Silicon Valley of Nairobi.

And so when we went there, we found an incredible disconnect between what head office was saying in San Francisco and the conditions on the ground in East Africa. We found a senior leadership team that had no idea, essentially, or at least as they presented to us, what was going on. Otherwise, why would you invite us in?

And so this was all done as part of an action research project called Fair Work that Calum was at the time was a postdoc on. I was a kind of tag along research assistant and Mark Graham was the director of who's our third co-author of this book. And the methodology behind Fair Work and what we do is essentially

To put it most simply, we grade companies out of 10. We talk to workers of digital platforms, platforms like Uber, Deliveroo, and we assess what working conditions are like. Do they have a fair contract? Do they have fair pay? Are they able to unionize? All those questions.

And we give the companies a rating out of 10. Now, this is kind of part of this civil society pressure group kind of campaign of raising working conditions, right? The companies purportedly are supposed to care about their reputation, they want to get a good score, they get graded against their competitors. And this has led to over 200 changes that companies have made, pro-worker changes to improve the conditions of workers at these firms.

you know, this methodology basically which has been applied to digital platforms. We wanted to kind of transition to see what AI companies and all of the companies in the AI supply chain would do. And so this project was a part of that. I think I might leave it there as like the intro to the book. If you wanted to add some intro material, we can talk about working conditions maybe after. Yeah, great.

I think one thing that would be quite useful to lay out a little bit is this book's called Feeding the Machine. What's the machine? What's the metaphor of the machine? How do we use this metaphor? How does it help us investigate questions around AI? For us, the machine is what we call the extraction machine. Because we conceptualize AI as not a complete novelty, not a completely new development that's broken human history in half and rendered inoperable previous theoretical frames, but rather a

a latest development in a long process of industrial development whereby resources, be they labour, energy, water, get put into a machine, converted through technology into profit. Fundamentally, the dynamics we're exploring here are not necessarily completely novel ones. The exploitation of labour in Africa is, as everyone here will know, not a new thing. It's a fundamental dynamic.

whether that be through slavery or a more contemporary version. So what we're trying to do is situate these stories of contemporary exploitation, these stories of contemporary colonialism and imperialism, within that very long durée history, and the social dynamics and structural causes underlying that. So the story of the seven workers that kind of make up the book are situated as the latest chapter in a very long story.

and the latest dynamic in what is a much longer history of technology production and work.

Thank you, James and Colm. And yeah, I should add as kind of context and another plug for the book, we're really going to focus on work today, but the book really takes a very expansive view on this machine and what it is that discusses question about investment, who controls investment and technology, infrastructure, environments, all of these kind of big, hard political questions. We're going to focus on work today, but check out the book if you're interested in these other issues as well, say the environmental impact of AI.

So thank you for that setup. So I think with that I might turn to you, Kirsten. So we have in the book, and we've sort of, James has started us on a description of the working conditions of many data annotators or data workers. So we had, for example, also Anita, a data annotator who works in Gulu, Uganda, works punishing hours, it's grueling, gets paid very little, no opportunity for breaks, no opportunities for organization in the workplace, etc.

I think, you know, given the kind of presentation in the book and our reaction to the story that James just told us, we might think that's a really bad job, right? We get gripped, like there's something bad about this job. But of course, as scientists, social scientists, also normative thinkers, we want to fill that in and think why exactly is that job

bad and then how can we measure how bad these various jobs are. So I'm going to turn it over to Kirsten and ask her, given your kind of research and disciplinary perspective when you look at these data annotators and content moderators, would you say these are good jobs? How can we conceptualize that? And then how should we think about measuring that?

Well, first of all, thank you so much for coming tonight and for being here. Thank you guys for being on the panel. I just wanted to add actually that apart from being a very informative book, this is a very well-written book. And one of the reasons why it's so interesting and well-written is because it opens each chapter with a fairly extensive description of different workers and their cases and

which provides an introduction to the issues that are then discussed in more depth within the chapter. I have to say I've been working on labour markets and labour markets in developing countries in particular for a number of decades now and I thought I'd heard it all, mostly.

I wasn't expecting to be surprised by this book, but I was. Not just by Mercy's grueling story of watching her own grandfather's death on the internet and having to content moderate that,

but also the various descriptions of these jobs that followed. Actually, the one thing I wanted to ask you before I go on into discussing whether this is a good job or a bad job, could you just provide a little bit more detail perhaps for the audience about why these jobs are mentally so stressful, so exhausting, and what the consequences are in terms of the mental and physical health of the workers? Because I think that's a very key issue here that we need to focus on and describe perhaps in more detail. So,

we can discuss whether they're good jobs or bad jobs and other associated issues. Yeah, I can do that. So there's a really fascinating line from the research on call centres back in the day where two people writing an article, I think 1999, the start of bank operations being outsourced to call centres. They talk about the creation of an assembly line in the head and

and the way that a lot of the dynamics they're observing in call centres are fundamentally the same as those that used to be observed in the assembly line in the factory, but instead they've just been converted into a white-collar context. I think that line, the assembly line in the head, very accurately describes a lot of what's going on in these data centres. I remember myself and Mark sat down

to do some observation of a group of people working. They were on a bank of desks, maybe 15 people, one team leader, 14 people working with them. And their job, I think, due to an NDA, I can't describe the exact details of it, but it was incredibly boring, right? There's images, you need to draw boxes around certain parts, and you need to do this for the entirety of your shift.

Now, you're pressured all the time to match a time per task, right? So there's a number of seconds allocated for every job. If you don't meet that number of seconds, your productivity starts falling, your likelihood of getting given a contract again after the end of your short-term contract falls, you're asked to stay in late and work unpaid overtime. So you constantly have to meet this time per task. You have to produce a certain amount. And that time per task is being dynamically adjusted all the time based on the top performance in the team. So if the people who are getting the bonuses in your team are doing really, really well, you have to speed up to keep up with them.

up them. And we're watching this work and I sat there for a few seconds and I immediately started to feel like I can only liken it to how I used to feel in like maths class as a kid where I was like oh my god this is the most boring and awful thing I've ever experienced as you may have guessed I wasn't very good at maths and Mark was sat next to me and he was like Jesus like this is horrendous it was like unpleasant watching people were incredibly quick incredibly fast at completing these tasks

But you just wondered, like, what is a day where you spend the entire time outlining a very specific feature of a very specific image over and over again with no capacity to slow down? And actually, Mark likened it to work he used to do at a factory in Germany, where he used to put a little electric component in the same place every day for a number of hours. So he directly made the assembly line comparison. I think it's the most accurate one, because all the phenomena one experienced...

back in the day working on an assembly line, or still working on assembly lines in a kind of a Fordist sense, you also experience in this, whether it be repetitive strain injuries, but also profoundly mental health consequences, both from the actual organisation of work, how boring, how distressing, how few resources people are given to cope, the lack of autonomy, the lack of meaning and direction. All of these have very profound effects on people, but also the system of work.

Because it's a very coercive system, one whereby if you don't meet your targets, you're put under direct pressure by managers, where because local labour market conditions are so bad, especially in places like Nairobi, you have huge urban surplus populations constantly looking for work. So the pressure to keep your job is colossal because otherwise you're going to be selling eggs on the roadside.

So that pressure to keep your job then means that managers have a huge amount of power over you and this leads to all sorts of abuse, be it abuse of women who are on pregnancy leave and not inviting them back to their work or straight up sexual abuse.

And so this environment is one in which workers are constantly pushed to produce faster and faster and faster and faster. They are dominated by managerial power that's accentuated by conditions in the local labor market. And it's work whereby they produce a huge amount of value for someone entirely located elsewhere. Elon Musk makes a vast amount of money off these workers. They make absolutely nothing, more or less, in relative terms.

based on their intensified production. These dynamics aren't just located purely in the South. They aren't just located in Kenya and Uganda. When we go to the story of Alex and the West Midlands, Alex has a fascinating story because he used to work in a Jaguar factory. The Jaguar factory in Coventry closed. The closing of the Jaguar factory was part of this massive wave of deindustrialisation. It was one of the last bits to go, but there had been this colossal wave from the 70s onwards.

And basically he ends up working back at the same site of the Jaguar factory, but it's been converted into an Amazon warehouse. And he struggles with all the same problems, you know, meaningless work, very low relative wages, a lot of physical strain, a lot of mental strain, repetitive movements over and over again. So there is this combined dynamic where we see very, very high work intensity, limited resources, huge stress and strain, leading to massive negative consequences for the people involved with limited avenues for address apart from self-organization.

Just one very quick Mark story that I heard him tell at Yale. Recently, our other co-author kind of visited a similar place in the Philippines and he said they had a task they needed to do. They needed to do it every 15 seconds but the task itself didn't kind of time them. There was no timer on the screen. They just had to keep that up. So everyone had mobile phones out and this is like a big challenge

kind of factory setting of like several hundred people with a timer every 15 seconds and for whatever reason they all had like chickens as their buzzers and so it was just this like cacophony of this bukka bukka bukka

Like the whole and the entire floor was just filled with these chicken noises. And he said he was there for like 30 seconds and he felt like he was going to go insane. But that was like a 10 hour shift for them, just with thousands of chickens going off all the time.

- Yeah, yeah, yeah. So we have this horrific cocktail of algorithmic management, right? You have to do these tasks super, super fast. You've got precarity, so you could be fired at any moment. And it's super dulling. And now, maybe, Kristen, with that, we'll turn back to you and say,

Would we want to call this a bad job with that kind of background in place? And if so, why? Probably yes is the short answer. So a couple of things. I mean, of course, human history is full of employment situations that are absolutely horrific. And we've been through this over centuries and we've found ways routinely of changing that, improving situations, and the ways through which we do these situations

or effectuate these changes haven't really changed that much over time. They basically relate to, on the one hand, regulation, on the other hand, workers organising themselves, balancing out the power structure and so on. But these situations are very particular and the fact that many of these jobs are in developing countries and as James and Callum were saying, are taking place in situations where there are literally millions of people looking for work,

Working under... The alternatives are also pretty horrific, so standing by the roadside selling sweets or something, often in intense heat, exposed to the elements, etc., isn't a very pleasant job either as an alternative. And so what happens in those scenarios is that you have policy makers often sort of making the "any job is a good job" or "any job is better than no job" argument.

Or alternatively saying, well, you know, employment is the best form of social security. Arguments like that are very common and very much focusing on the quantity of jobs rather than on their quality without really any regard to what the impact of that lack of quality may be. So I'm going to actually draw a parallel to an area of public policy that we...

pay much more attention to than labour markets, which is education. So it's pretty obvious now, regardless of how low the level of development in a country is, that we have to focus not just on extending the coverage of education and the quantity of education, but we also need to focus on the quality because otherwise the whole investment is pointless and

kids come out of schools having learned nothing. We sort of learned this the hard way because we instituted conditional cat transfer programs, often encouraged by development institutions, that expanded coverage but didn't really focus on the quality. So at the end of the day, you had kids sitting in school all day but still not knowing how to read and write.

And a similar sort of thing is going on in the labor markets here where we do still tend to focus on quantity with this argument that any job is better than no job without really thinking about the consequences. So for example, one of the descriptions that I found very powerful in the book is that the mental health consequences to these workers are so severe in terms of depression, anxiety, suicide attempts, et cetera. And we see this all over the world in many types of different jobs.

that at the end of the day you end up with a worker who's so burnt out that they cannot work anymore. And in advanced economies we learned this a long, long time ago and we've instituted protective mechanisms that because these jobs are happening in the global south and developing economies we're not paying enough attention to this.

However, in other forms of the production chain, we have done so. So we've all heard about sweatshops and appalling conditions there, and gradually there have been improvements, and both this sort of top-down and bottom-up approach. So companies in the global north who are buying from these suppliers have pressured for better working conditions, and workers themselves, as conditions improve, as countries develop, have more power to organise jobs.

But the crucial role, I think, in the space between these two forces is also government and regulation. And this is why it's so pernicious if a government says, well, any job is better than no job and we shouldn't worry about equality. Because at the end of the day, that government will then also reap the negative externalities of jobs like this in the form of workers who become inactive, workers who become sick and ill,

and who can't sustain their families. So I think that's a very important consideration that we need to focus on, and developing countries, even if they are relatively poor, can do a lot to regulate these situations.

At the same time, I have to say that regulation tends to generate resistance, so there's always this worry that these jobs will be lost and companies will move elsewhere, etc. That concern is always present, but again, it's not a long-term solution, so what we should try and avoid here is a situation where there's a race to the bottom.

that is so detrimental to workers and productivity overall without bearing in mind the negative externalities, the long-term consequences of having jobs like this, even in less developed economies. I'll leave it at that. I'm sure we'll come back to some...

- Yeah, thank you. I'm gonna pick up on, and I think we'll return to a lot of this, but just two threads in what you've said. So the first, give you all both a chance to comment on the any got jobs, better jobs than no jobs line, which is something you also heard from government officials, and the kind of trade-offs in keeping work in a country, expanding work to more people versus increase the quality of work. Do you think there's a trade-off

How should we manage that? But also second, this thread of jobs in the global north and how well insulated they are given by existing regulations. So you brought up in the book, you have this nice account of Alex, the operator in an Amazon warehouse in the Midlands. You already brought this up, Callum.

I know someone also who's worked in an Amazon warehouse in the U.S., and here, existing safety protections seem really inadequate precisely because of the kind of punishing pace of algorithmic management. So if you're a worker in a warehouse, you're constantly being monitored, you know you'll be fired at any minute, especially in the U.S., right? There are no employment protections. You can be fired at will. If you fail to bring enough boxes to a certain place in a certain amount of time, you will work through injuries, you will get sick, you will get stressed,

and all of this will have a huge, as Kirsten said, right, it's an externality for all of us. All of us are paying into, you know, joint healthcare, all of these things. Companies are not bearing the cost of this, but we as the kind of citizens are while companies are profiting. So I'm curious, given the story of Alex, right, also how well you all think jobs in the global north are insulated by safety protections and other regulations. Yeah.

Do you want me to do the ethical AI and you can talk about Alex? So I think the problem that we encountered when we were at these data centres was that they were part of a global labour market in which there was a race to the bottom on pay and conditions. So one of the largest tech companies, again, we signed an NDA, but one of the big boys was

essentially parcels this work out to seven or eight different global data center annotators and annotation centers rather across the globe, right? So there's like one in the Philippines, one in Nairobi, one in India, etc. And they're actually all in competition. And how they know they're in competition is the tech company sends them a list every week of how well they're all doing.

So it's not just a metaphoric labour market in which they're potentially in competition, which a hypothetical worker or work centre could take their work. They're literally in competition. And the highest one at the end of the week will get a little bump in their contract. They might get a little bit more work or a slightly better cut, so they're encouraged to do as well as possible. But what this actually means is that the middle managers, the people running these centres...

see themselves as having to enforce these really strict standards of labour discipline, of keeping pay down, of making sure people work as many hours as possible, maybe even getting a few unpaid shifts in. And we did document many instances of unpaid overtime at the centre we were at.

in order to protect their workers, right? In order for jobs to stay in that centre, right? If you want jobs to stay in Nairobi, in Gulu, you need to work harder, right? So the whole system and the whole game of...

having workers, their names are up in green, if they fall below their targets, everyone can see their name goes into red. The speed targets, the efficiency targets, all of this is about saving jobs. It's for their own good. And this kind of brings us to the point of the specific company we went to markets itself as an ethical AI company. They were an NGO.

Their CEO, Leila Janna, wrote a book called Give Work, Not Aid. It's like a critique of the aid industry. It's a critique of these do-gooders who think that they can help the poor, starving Africans by sending money overseas, by donating to Oxfam or some other charity. Here, Leila is starting a new enterprise, giving people work.

allowing people to enter the global digital economy. And so the NGO, which becomes a company that she founded, along the way basically gets caught up in this labor market in which they're in competition with all these different companies. They're forced to become a for-profit company to be a more efficient enterprise, to be able to protect their workers, to expand their mission.

And in the process of that, they really, from our perspective, reveal the kind of structural limitations of social enterprises in this particular industry, right? That it's really hard to pay your workers well. It's really hard to make working conditions better when you are in a league table, essentially, with eight other companies competing for a contract from Tesla, Meta, whoever it might be.

And so there is this kind of structural pressure on these companies to act in a certain way. So we asked ourselves then, well, you know, where are the empowered actors in this network? Who can actually make a difference? Because there were just limitations on the extent to which middle management in East Africa or wherever it was in the world could actually do something, right?

The workers were doing everything they could. They were struggling, they were organizing, they were putting demands to the workplace. But they're on temporary contracts, right? There are limitations to the extent to which they can be successful in those kinds of union and collective action tactics. Because anyone who is found to be talking about that kind of stuff is just not having their contract renewed, right? It's very easy for the company to kind of manage that stuff.

It's the lead firms in these supply chains that have the power. Facebook, one of the reasons why one of the centers had wellness counselors

by the way, not a trained psychologist, just some random pulled in off the street who was called a wellness counselor. You know, workers doing 10 hour shifts of toxic content moderation in which they're moderating one ticket every 55 seconds, right? Thousands of instances of horrific abuse and pornography and suicide. You know, they would have a mandatory 15 minute session with a wellness counselor every week because Facebook said so, right? So,

They do have the power to set minimum wages. There could be a living wage for their country. They do have the power to set working conditions because whatever they say, you know, the data annotation centres will, you know...

Facebook says jump, they say how high, right? That's the kind of relationship they have within these global production networks. So in addition to what Kirsten was talking about and this idea of government regulation, which could either be through supply chain laws in the global north...

It could also be local laws at the location of these centres, although there are kind of issues with both. We also found that putting pressure on the lead tech companies was one possible way of trying to name and shame and get them to take responsibility for conditions in their supply chains because it's at the top rather than at this kind of middle section where we thought that you could have some serious possibility for change.

Just to touch slightly on the dynamics of the North-South relationship here. Mike Davis wrote this fantastic book called Planet of Slums, I think it came out in about 2004, which is a magisterial global overview of the development of huge slums on the peripheries of cities in the Global South where people who were ejected from agriculture as agriculture became increasingly productive and centralised in agribusiness ended up working in informal economies or not working at all.

and were held there by the particular dynamics of global labour mobility. Because when we had huge populations expelled from agricultural production in England, where did they go? They went to Australia, right? They went to America, they went to New Zealand. They colonised half the world. This settler colonial project was part of dealing with that surplus population, of having an outlet.

But now if you're in Kenya and Uganda, you can't go anywhere. There are militarized borders. If you attempt to come to Europe, there will be camps and wire and boats and drones, and they are trying to stop you get there however they possibly can. Fortress Europe does not want you. So historically, we've seen this outlook. The surplus population expelled from agriculture finds some other place, goes and often kills a load of people in another country. Instead, now what we've got is this dynamic where this global wage differential is maintained by the trapping of surplus labor in the global south.

And that means that imperialist capital in the global north can then use that as a resource, right? Because where is 80% of manufacturing done? It's not here, right? It's wherever labor is cheapest. If you go and look at your Foxcons, if you go and look at where the vast majority of production is happening now, manufacturing production, but equally this AI data chain, it's the same thing. It goes wherever cheap labor and cheap resources are.

So increasingly, the reality for the vast majority of the world's population is you are treated as a expendable resource, part of a giant population, not all of whom can possibly hope to get work, who are basically gonna be worked to the bone for the profits of people like Elon Musk. It's a dramatic summary, I think, of a very dramatic situation, but so much bigger than just necessarily the technology of AI. On how this reflects in our context today,

There's often, like, a very comforting story that, like, that's the situation there and everything's obviously much better here, right? You know, everything's all right, everything's fine, no-one's hyper-exploited, no-one's suffering work like this. We've talked about Alex's example, which is maybe one start of a counterfactual, but I also want to talk about another example where algorithmic management is important, and this is the platform economy, right? We're all in London. We've all seen the number of food couriers whizzing back and forth all the time. How many people...

really think about the reality of that job. Because at the moment, people working in that job do not make the living wage, especially once you take off costs. They are working in absolute poverty. They're often paying 20% of their wage to someone who rents them their account, paying some of their wage to someone who rents them their moped, paying some of their wage to someone who rents their

accommodation. They are constantly dodging immigration enforcement. The vast majority of jobs in London now are done by undocumented migrant workers. They are paid absolutely nothing. If they come off their bike after colliding with a bus, it's them that has to deal with it or not deal with it, as the case may be. They're dying every day on our streets and they are hyper-exploited. And the way now you can see companies like Uber and Delivery, they've become profitable

They finally made the turn to profit. And you know what they used to tell us? It was like this technological dream. We're going to become profitable because we're going to invent automated delivery or we're going to invent the driverless taxi and this is going to cut costs and allow us to massively expand and we're going to become a monopoly, but in a good way. Instead, what's actually happened is it's become profitable through hyper-exploitation. It's become profitable through the fact that the wages these people are paid to do this work are absolute pennies. You know, three pounds for a three-mile delivery that's going to take you half an hour and you have to drive your own moped and pay for your own petrol.

This is the situation we're at now increasingly. And so the stories of how the UK's model of employment regulation is meant to work just doesn't apply when increasingly employment is not the dominant relationship. The norm of work post-2008 is not permanent, standard, full employment models. Instead it's precarious. It's zero hours, it's a fixed term, all the rest of it. But increasingly not even employment at all.

And when you circumnavigate the employment relationship, you circumnavigate the entire body of statutory regulation that's been built up around the employment relationship. So you get rid of minimum wage, you get rid of holiday pay, you get rid of a load of these rights. So I would really like the story, in a way, to be...

There's a bastion of security and the goal is to up-level everyone to this bastion of security. The truth is work in the global north is deteriorating at an incredibly rapid rate. Algorithmic management is facilitating that by facilitating, for instance, the fiction of self-employment or massive intensification of work in a warehouse like Amazon.

But it's not the case that we are safe and secure and this dynamic is affecting other people elsewhere and we must act on it because we are like nice charitable people. It's not about charity, it's about solidarity, right? The fact that exactly the same dynamics that

that allow ultra-rich tech billionaires with obvious and increasingly concrete connections to fascism to determine how technological development operates in our society, to profit off the work of millions and millions of people. That exact same dynamic that enriches them in places like warehouses and platforms in the north, exactly the same thing that immiserates people in the slums of Nairobi. It's not...

It's not even two parts of the same story. It is the same story. We're exactly the same people in the same position. We just have to be in different places within it.

Yeah, to pick out a thread and then maybe turn back to you, Kirsten. So we have these structural conditions of the labor market, which were nicely outlined, that with the kind of capabilities of AI really give us this perfect storm of deteriorating jobs. So we have this race to the bottom and globalization that James talked about. We have this issue of precarity, which we see in the platform economy. A lot of people working outside of traditional firm employment relations, they're not protected by labor laws, if those exist.

We have the power of managers through these constant surveillance tools. And then, of course, we have this issue of surplus labor and the kind of precarity that surplus labor creates. So I wanted to ask you, Kirsten, are we missing anything, first of all, from these kind of structural conditions of the labor market? And then are there any variations that you've seen in your research?

And then secondly, maybe also for the whole panel, what are the kind of difficulties in actually measuring and figuring out these structural conditions, and why has that been hidden from us for so long? And I think on the one hand, there's a story about academic research itself, as you said, right? Economics is really focused on the quantity of jobs, not the quality of jobs. So we just have a lot of work to do to conceptualize what is a good job and where are we seeing bad jobs and why,

But then also, we'll come back to you both, these issues of, well, if companies control information about their supply chains, if these supply chains are so complex, how can we as consumers even know what's happening and raise this call for solidarity or put pressure on companies if we have no idea what the working conditions of people are?

actually are. So yeah, if you want to add anything to these structural conditions of the labor market, and then just the difficulty in figuring out what's happening in labor markets globally.

I've got two anecdotes which I think illustrate and help us understand this. So without going into too much detail, but I just wanted to mention a long time ago when I was working in South America, Chile had signed free trade agreements with Europe, the US and South Korea. And it just so happened to be that we could actually measure the relative impact of these free trade agreements on the sectors in the Chilean economy that were affected by them.

In the case of the bottom line basically was that the employment conditions in those sectors affected by the European Free Trade Agreement improved. The US was neutral and the South Korean one deteriorated employment conditions in Chile. For very simple reasons, the European Free Trade Agreement came with clauses about employment conditions and the European buyers, large companies would come down and inspect locally what the employment conditions were.

In a way, it was much more effective than local government regulation, which was always resisted and people always found ways of trying to circumvent that. Because if the European buyer comes and inspects that there are toilets, protection from the sun and so on, then companies... It's a bit like what you were saying, if Facebook says jump, then the supplier says how high. There's no discussion or no debate about the fact that this has to be done, it just gets done.

The problem with that mechanism is it isn't universal and it isn't rolled out in the same way. It's just those suppliers who happen to have inspectors that they do send down. They do illustrate the kind of power that we have and the fact that the institutions and the agreements, the way we organize supply chains, that really, really matters. And a lot of that power is in the global north and the pressure that we put on governments, on retailers and so on to organize these things.

But the other thing that I wanted to pick up on is what you were mentioning, both of you actually, about the fact that we are in a very different situation now compared to decades ago in that we have a global labor market where we insure workers locally.

and this issue of what you're saying about employment conditions in the UK not being so different as in other places, there is in fact a global convergence going on. It's been driven in part by development institutions encouraging companies and governments to flexibilise their employment legislation, generating a race to the bottom, but it's also been generated by companies like these generating new jobs that are by definition precarious.

And we have an increasing proportion of them in the UK. Now, the big question is, well, how many of these jobs are there? And I'm pretty certain that if you try and answer that question based on your average labor force survey data in the UK, for example, and in the US even less, they've got 500 pages of survey. They ask three questions about hours, wages and wages.

Not much else. But these surveys aren't capturing any of this. So how many gig economy workers do we have in the UK? Well, I've seen estimates that range from half a million to five million. I'm sure it's changed since then. It's probably doubled. Some of these workers are working partially in the gig economy. Other workers are working full time. Some of them are legal. The conditions vary significantly.

And so the employment conditions that we have in our labour markets have become so fragmented. If you think about the traditional employment relationship where you had an identical employer, you just had one employer, you worked full-time, you had a fixed salary, you probably had a permanent contract, you knew how much you were going to be earning this week, next week and the month after. Those conditions are...

I won't say they're gone, but they're increasingly becoming fragmented in part through increased proportions of self-employment and through precarious conditions, even within formal employment relationships. And the fact is, as you were saying before, this generates a whole host of negative externalities, which we don't measure. So, for example, the accidents and the gig economy that the National Health Service then has to deal with, well, how much does that cost us?

and we are not charging employers who employ these people more in terms of their national insurance rates, in terms of their tax rates. We don't require them to publish information on how they hire workers. So, for example, the information that we publish in our accounting and reports on labor relationships is just so minimal, it's ridiculous.

And of course the government so far has not considered implementing differential tax rates or national insurance tax rates that could cover some of the costs of this on the one hand, but also disincentivize employment relations like this. The other sort of very quick anecdote I wanted to refer to is

If we don't measure this, so the first thing is gathering data. We're not very good at that. And second, we don't really conceptualize what constitutes a good job or a bad job, what constitutes a negative externality, how much does that cost? All of this is sort of happening under the radar. And what happens then is that you actually then subsequently try and implement policy. So in my case, for example, I was working on unemployment insurance in Chile.

And when they came out with the design of the legislation, I said, well, this is never going to work. Well, why not? Because your employment conditions are so precarious that people who will require unemployment insurance will never have complied with the conditions you've just written into this legislation. Oh, no, no, no, that's not going to happen.

Well, one year later, the administrative data comes in and they realize, well, actually, of the people who were unemployed, nobody was covered and nobody was claiming unemployment insurance because a very, very high proportion of their contracts going into the insurance, and this is formal good jobs, right?

So many of those were on very short-term contracts and people didn't last long enough in their jobs in order to accumulate potential benefits. So this is what's happening when we're not measuring these things. Labour markets are changing under our noses and we are putting very little effort...

into following up on this. And the irony of it all is that the one thing that could help us to both figure out what's going on and revise policies, target policies, and calculate the cost of this is AI.

So one example is I left the UK when I was a PhD student to do research on Chile because there was no information there. So it was a perfect place to do a labor market survey. It was the first survey on employment conditions, et cetera. This was more than 20 years ago. By the time I left Chile, 20 years later, Chile has a linked up administrative database with all the data from the

the health insurance system, the education system, labour markets, pensions, unemployment and so on. You can link all of this information up and therefore you can use it on the one hand to target social policy. You can also track people over time. You can do much better research. And I come back to the UK and we still have a labour force survey which seems to be clumsy both in terms of its response rates and the information it provides us with. And meanwhile the UK has put

some effort into linking up administrative data so that we can use it for this purpose, but really not enough at all. And so I think we really need to rethink the way we approach these things, conceptualize, measure. I mean, there are methods out there for measuring how many jobs are of poor quality. And unfortunately, they're quite basic because, for example, your 10-point score from Fair Work

But there's nothing in that, in the standard labour force surveys or understanding society or household surveys. And so the other thing is, of course, we tend to think about these things as siloed policy areas. So you've got a household survey that's really designed to look at the social circumstances of family, but not at your employment situation. And we can't link that up with the employment data. So I think we need a fundamental reboot, really, here, both in the global north and the global south.

Yeah, and I think one of the really, I think for me, especially impressive things about the book is it does precisely this, right? You wouldn't have thought, oh gosh, the mental and physical health of workers, the infrastructure that powers AI, the environmental impacts have anything in common. And what the book is aiming to do is really

bring these big threads together. But I want to pick up on something that Kirsten brought up, this kind of hopeful use of AI. So AI, you might think, right, it has a lot of potential. We're not seeing it.

in the labor market, one reason, and I want to put this as a question to you both, that might be are just these huge inequalities in power that you chart in the book. So we've already seen some come up already, right? The inequality between your big tech company and the company employing data annotators, right?

inequalities in power between workers and the managers that they work for, even between consumers, all of us, and our governments and tech companies. In many cases, sometimes too right between governments and tech companies. Tech companies can say, well great, we're gonna pull work out of your country or we're not going to provide you with needed infrastructure, right, if you enforce, say, certain labor laws.

So can you talk us through a little bit what is so troubling for you all in those inequalities in power and how can we start to redress them? I was going to do, well, I think one of the dynamics that comes out in the book, which me and James kind of thought through together, is really the tendency towards monopolistic concentration that emerges in AI.

and how the very, very intense resource and compute and other requirements associated with artificial intelligence leads towards this process whereby all of the dominant AI companies have immediately been bought up by previously existing tech monopolies and concentrated within them.

So the actual path of technological development that is shaping how AI changes over time is not being determined by anyone thinking about what's the possible social use value of this stuff. It's being determined purely by people like Peter Thiel. People don't know who Peter Thiel is.

he's a Silicon Valley investor, very friendly with the extreme right in the US who is highly focused on surveillance technology. So we kind of chart a bit in our discussion of investment. We went and talked to some venture capitalists and it was really fascinating because we were trying to understand like how does the path of AI development get charted?

Who's making decisions at any one time about where this annotation work should go or what it should be optimized for or what kind of outcomes would be of most value? The disturbing thing you discover when you do some of this research is that basically it's all about increasing valuations and there's nothing, absolutely nothing that compels anyone involved in this process to think at all about the social impacts of what's going on or how the technology can be used for better.

So it strikes me, we also tell the story in the book of a guy called Donald Mickey. Donald Mickey was a... He used to play chess at Bletchley Park but was absolutely the worst one playing chess at Bletchley Park. He got beaten by everyone. But he did kind of invent machine learning. So he came good in the end. So he was developing this technology, machine learning, in the 60s using matchboxes, getting them to play noughts and crosses, basically.

There's nothing about that technology, nothing about that development, nothing about that science, nothing about that way of thinking about problems that's inherently bad. Nothing about what we point out and say, "Oh, that's evil." But the problem is the whole infrastructure of technological development and the social relations that this technology is emerging within are conditioning it in such a profound way that all of those emancipatory opportunities, the opportunity to automate bad work, to discover new things, to achieve new progressive developments,

are not being realized in the meantime. What's happening instead is this technology is being put to the ends in which it's been developed, right? And those interests are not our interests. Fundamentally, those interests don't really overlap with our interests whatsoever. So this concentration of power and control in kind of monopoly capital is one of the huge overwhelming barriers that we have to get through if we want to think about how AI could be used for better.

And in this sense, I think AI is quite an interesting technological cipher that forces us to think about these structural problems in a very big way. Because actually the solution to that isn't just five new laws. The EU AI Act is going to do nothing about that concentration. It's going to do nothing about how that process of technological development is happening. If you want solutions on that level, you have to think at a much, much higher level. And you have to think about much more transformative processes that are going to shift that concentration of power.

And shifting concentrations of power on that level means something much more than new regulation. We have a five-point plan. We realise that we academics that we're trying to write a book for a more commercial audience. And so what do books do at the end? And you have a number of points that you need to take. The fifth point is overthrow global capitalism. We don't really say it.

We don't say, we kind of gesture to it. Like, if the first four points don't work, well, there is global capitalism seems to be at the root of some of these issues. We never really, yeah, there's not quite a red flag at the end. But the kind of steps leading up to that

things like supporting workers. And one of the main points we say here is that we need to be more transnational in our organizing, right? That tech workers, and they are tech workers in places like Kenya and Uganda,

have reason to have solidarity and vice versa with workers in Silicon Valley, right? That they are part of the same system. And we do see some instances of that kind of transnational solidarity in action. But I think we need to see a lot more. The second step, one is kind of building and supporting collective workers power. The second thing is putting pressure on these through civil society campaigns. So things like the Fair Work Project that we did this research through.

and other NGOs and civil society groups that are trying to both make these supply chains more transparent, to shine a light on what these companies are doing and to put pressure on them to kind of change their behaviour. The third step

What's the fourth step then if the third step is regulation? Oh, don't test me. Maybe it's... No, that's right. The third step is the possibility of actually having workers' cooperatives in this space, which I think we... Because that's a fairly dicey kind of...

But are there ways in which we could have businesses owned and controlled by workers, or maybe even partially owned and controlled by workers, that might leverage a slightly different workplace? The fourth step is government regulation, and we look at some of the difficulties of that. I think the biggest hopes that we have at the moment are on something like the EU's corporate due diligence directive.

which essentially looks at businesses trading in the EU and turning over more than 150 million and saying, "Look, you need to be responsible for what's happening in your supply chain." So that is actually going to include a lot of the big US firms if they want to continue trading in Europe. And that came into force this year, and I think people have got two years or something to kind of get their house in order. But it includes things like paying workers a living wage,

and allowing them rights to unionize. Two things which are a huge problem in some of these countries and centers. So there are possibilities here. I do want to end on a slightly more positive note. I'm looking forward to doing different research. This is a very depressing topic. I feel that every new project is going to be even more depressing than the last, though. Yeah, all my other research projects are getting more depressing.

More depressing, okay. I had a previous book called Platform Socialism. It was all hopeful and optimistic and looking at different ways in which you could organise the digital economy, but honestly it feels like a lifetime ago, like doing that kind of stuff. And a lot of the cooperatives that I actually analysed are kind of no longer in existence. So it is a little bit depressing. But we do have a five-point plan. So the most important thing, obviously, is that you buy the book.

And then you have access to the five-point plan, which you too can institute in your workplace. And we'll get rid of global capitalism after you, by the way. That's the last... Order's in now, and then... Yeah, then it's all together in the solidarity economy.

once you have the book absolutely and thank you for bringing up the five point plan because of course we said in the event outline the event description that we would also tell you what we need to do individually and collectively to fight for a more just future right a lot of the event today we focused on all these issues around bad work but i totally agree one really nice thing about the book is we have this kind of five point plan at the end

showing, okay, what can workers do? What can companies do as we've talked about, right? Care about your supply chains, contract in a smart way, use your power in a nice way. But obviously as we've seen, that's not gonna cut it. We don't want companies to have kind of too much power to decide at their whim. So there's an important role for policy makers, governments as well.

So I think with that, what we're going to do is turn over to the audience and hear some questions from you and give a chance for all of our panelists. Although I might give Kirsten a final word on what we should do because given her experience, I feel like you've got a lot to add here before we jump to questions. Yeah, there's just one final point I'm just sort of itching to make, which is

We have a way of thinking about labour markets as sort of, you know, the economists for their trade-offs, public policy makers are highly influenced by economists, and we tend to reduce labour markets to a couple of variables, basically the quantity of jobs and unemployment and also wages. And the thing I've been trying to highlight over the last sort of decades with all my research, but I think that this book brings this out really well, is that

it's just not enough to focus on not just the quantity of jobs but also wages in particular as a substitute variable for quality of jobs is a completely inadequate variable. One of the things that I really like about the book is that it highlights this because

The jobs that are being carried out, you could pay them double or triple or whatever. And it's not that wages don't matter. This would definitely improve the lives of the workers concerned. However, you can't pay for destroying somebody's mental health or physical health. And we've seen this over the history of time.

in labor markets all the way through, and it's still true today. But the problem is that public policy tends to reduce these very complex situations to a couple of variables, and the most important one there is wages. And these jobs really highlight that wages aren't everything. You can increase the wages,

but it's not going to resolve other fundamental issues that will also basically in the end incapacitate you. And that's global. That's a global point. It doesn't matter whether you're in the West Midlands warehouse in Amazon or whether you're in Nairobi, etc.,

Yeah, and it's really, I think, it's a nice point to end on because work is really a holistic problem, right? It impacts our health. Education plays into where we can work, how we can work. And as Callum said earlier, this is really like a big political problem where we need big structural change. It's not something that can be solved by limiting working hours a little bit if you don't address, say, the safety issues around jobs.

So with that, we turn it over. So we've got a chance both for the audience in the room and the audience online to ask questions. So audience online, feel free to ask questions. And what I will do is make a note of all the questions, and I will indicate you by my perception of the color of the top that you are wearing.

So the gentleman in the blue jacket-- sorry. And I will, as I just assumed gender but I did not intend to, I will just call you person with the blue top. So please, person in the blue top in the back. Thank you. My question is for Dr. Kalam. So I am from India. And we provide electric vehicles for people. And they in turn run them on the platforms.

So one point of discussion which we didn't have was that because of the human labor involved in the AI related activities,

A lot of participation, especially in countries like India, the women participation has increased a lot. They've got a lot of flexibility, they can work from home, especially after the COVID. They've got a lot of flexibility, even if it's a more, you know, job, menial job on social media, you know, reading comments or whatever, you know.

replying to commands but uh... a lot of women participation and uh... the flexibility of work hours and work has increase in india uh... and uh... secondly my point is that uh...

how these workers right now are being, you know, getting trained on the AI tools. So I am not an AI expert, but I would rather like to know that how they are kind of ascending their ladder in their workplace. Like, I'm sure they will be equipped with good, they will learn how to use, you know, future AI tools much better compared to a newer workforce. Yeah, that's it.

- Thank you. Yeah, we'll just kind of answer questions one by one. - Yeah, great. Yeah, I mean, I think women's participation in labor markets is definitely changing through platformization. It's a huge dynamic. I wouldn't say necessarily

we should view that as like being an unblemished positive. I mean, one of the major examples I have for increasing women's participation in labour markets is the automation of textile manufacturing in the UK during the Industrial Revolution. The use of machinery, one of its major effects in the Industrial Revolution was to allow women and children to work in the factory, doing jobs that previously had been male only. So we can see these kind of very regressive developments packaged with positive elements. This is kind of characteristic of capitalist development, I think.

and shouldn't perhaps bamboozle us and say it has to be one or the other, right? Like both of these are possible at the same time and unpicking them is part of our challenge. As far as skills, I really wish you were right. Like it would be really nice if you were. Harry Braverman, who wrote a fantastic book back in the day about work,

had this idea of de-skilling, right? And part of the dynamics of capitalist work increasingly is that the level of skill involved is reduced in order to increase the possible labor market, in order to reduce the power of workers over their own work, and to increase the power of management to coordinate labor.

And this is really one of the key features that we observed was that the level of skill involved in the work was very, very low. It was routinized. There was a specific homemade piece of software that workers interacted with that didn't really allow them to develop any skill. And you could work there for an incredibly long time and not actually move up the ladder. So movement up the ladder, skills-wise, was almost non-existent. There was a very hierarchical organization of work.

where one in every 15 people or so could move up to team leader, then one of every 15, 20 team leaders could move up from there. But by and large, most managers were hired in from previous managerial experience and other business process outsourcing firms. So there was very little internal mobility. So whilst there was this idea of upskilling, the reality was really one of de-skilling. Actually, if anything, the level of skill lowered over time.

and managers would sometimes say to us you know we sometimes hire people with degrees but it's completely unnecessary you know anyone could do this job and that's part of the point of how it's organized so I wish there was kind of like a progressive narrative here about people upskilling and this leading to process of development but much as you can observe in the rest of the development literature you know dependent economies are not moving out of their dependency they are still trapped in that situation and AI is just reproducing what we've seen with every other manufacturing industry for the past few decades

Yeah, and just to add, of course, that to get back to Kirsten's point about externalities, often societies, when education is required and upskilling, societies are the ones bearing the cost of that, right? Companies are employing this kind of precarious workforce where they're contracting a worker to do a single job. They're not educating them. They're not providing these relevant skills, right? Societies, we're all paying in a fund to do that, but companies are benefiting from that.

person on the far left in the mustard jacket. Hi, I'm Angelo from the Philippines. Thank you very much for the

presentation it's uh it's good to see cross-country case study research and my question is across these countries you mentioned about nigeria philippines and india in the context of assessing the labor standard and the risk reward from a worker's perspective like in the philippines this sector has been an outperformer in terms of growth it has been a major driver of job generation

And at the same time, from the business perspective, they still have to get some adequate supply of workers. So there's a shortage of workers. So that would mean that businesses are willing to pay more or provide, say, a premium on top of the wages, maybe better multidimensional type of working environment. So in that context, has there been some research from your end

the risk-reward across countries and on an intertemporal basis over time, will this hidden human labor feeding the AI be replaced given a certain time horizon? And in that context, what's the policy implication given that countries are also wanting to become more competitive in this sector? It's an important generator of

foreign exchange, trade in services, in the case of the Philippines. So they also want to become more competitive, but they also need that type of labor supply that is, of course, optimal in terms of the welfare.

that is being provided to this sector, not just wages, but also other factors or perks as well. Thank you. Hi, I'm interrupting this event to tell you about another awesome LSE podcast that we think you'd enjoy. LSE IQ asks social scientists and other experts to answer one intelligent question, like why do people believe in conspiracy theories? Or can we afford the super rich?

Come check us out. Just search for LSE IQ wherever you get your podcasts. Now back to the event.

Maybe just to start with your final question first, is this work likely to be replaced, automated, maybe AI will take over? Not in the near future. The companies are making no plans on this whole industry getting wiped out. They're all here for the next 10 to 20 years. People talk about something called synthetic data as perhaps being able to replace data annotation. It's

there are technical kind of hurdles that would be involved there. Like a recent research paper showed that language models, so like ChatGPT, that use synthetic data, so stories that ChatGPT wrote, essentially using that as then the data to train it on, that that kind of eventually leads to model collapse.

There are also problems with any kind of data annotation. The main issue is all the work they're doing is all the contextual, tricky, fiddly, annoying stuff that it's really hard for machines to do because

they need very clear tasks. And the whole point of data annotation is it's not clear, that you actually need to basically clean it all up so that a computer can read it. One of their other key tasks is not just annotating the data sets, but verifying the machine learning output. So basically, you train an algorithm, it produces some outputs for you, and then you get a human to go through and go, oh, that doesn't look good, this isn't working. So it's all of this...

tidying busy work it's really hard to get a machine to do that so not anytime soon will this kind of be automated away with the kind of comparative analysis

To be honest, a lot of us, we're kind of in the sociology of work field. Most of us do kind of case studies. We do have people that kind of go out and do a little bit of comparative work. I know that our PI, Mark, has kind of visited the Philippines and done some work there. I remember him telling me that he had a kind of discussion with one of the government officials, and what they told him was, look,

we're not interested in this to be high-wage work. We have no intention of kind of helping workers demand high wages because we see ourselves competing in this global labour market, right? If we try and push wages up, the jobs will go. So it's not our intention as a government, and I think he was talking...

to a senior person within the Labor Ministry at the time. It didn't seem, at least from that report, that they had any intention of pushing for that because of this global competition. Did you want to add anything? - Yeah, I think you raised a really important point. At the moment, what we do have in terms of information on these situations is mostly qualitative work and studies like this.

And it's very, very difficult to come by internationally comparable statistics. And to give you an idea, we just used UN data on Latin America to study how many jobs are of poor quality. And we had four variables that we could compare. Wages, the hours people work, the type of contract that they had or whether they had a contract and whether they contributed to Social Security. That's all we could compare. And that took time.

forever to get that together. One of the problems that we have is it's not just national governments who will design their own surveys, but it's also international institutions who try and help them. They don't really have much power in this field and can't say, okay, do it this way, do it that way, or standardize your surveys, and so on and so forth. And the problem then, of course, is that without this internationally, it's the same as locally with an individual economy. If you haven't got

internationally comparable data, then it's very hard to figure out what's going on. Who's competing with whom? What's the lowest common denominator? And where is this race to the bottom taking us? So the data requirement in this is-- I can't overemphasize it. I keep on coming back to this point. Because without the data, we can't really figure out what's going on. So we'll take two questions online and then return to these ones in the room.

So the first comes from Ahmad. How do we address the issue of cheap labor in regard to provisions of raw materials in sub-Saharan Africa as a means of revenue generation to the individual communities and countries? So expanding this to the kind of physical inputs into AI systems, and again, these labor questions and exploitation questions as well. Did you get that?

It was about raw materials, right? Yeah, exactly. Can what you've said about the kind of structural conditions of the labor market speak to the questions about labor in regard to the raw materials that also go into these AI systems? Yeah, I mean...

It reminds me, people might know Marini's Dialectics of Dependency. It's a Brazilian work from the 70s thinking about the role of Latin America in the development of European economies. And what Marini argues basically is that Latin America, workers there underwent a process of super exploitation because they were having to produce raw materials and inputs that were going into a Western economy, but none of the capital was being accumulated in their local context.

So instead of having this kind of rise in the productivity of labor that could allow for higher real wages over time in the local context in Latin America, instead all of that capital was kind of exported to the imperial core and there was no increase in the productivity of labor locally and this reproduces conditions. I think we can see exactly the same dynamic going on whether it be with cobalt or whatever. Even with lithium, right? You look at some of the contemporary conditions with lithium production, you see very much the same thing. And rare earth minerals are very much reproducing this exact same dynamic.

So it is kind of the constant return of the same, right? The same dynamics over and over and over again. And...

And I think it's profound in a way to recognize that constant return of the same dynamics and to see this as like one set of historical cycles playing out repeatedly. Because it does suggest to us that any of the solutions we've attempted to develop previously have not worked, right? So at least it delimits the field of like possible non-solutions being like, okay, so do we think that trying to do fair trade cobalt is going to work? Well, did it work with coffee? Not really, right? So maybe we need to think in another register.

Thank you. And that brings us really nicely to our second question from the online audience. So George asks, what can/should really be done? So I take an emphasis on really because we heard your five-point plan. Maybe George is a little skeptical of the five points. Or maybe-- actually, maybe what George is asking is like, look,

the three of our panelists, if Keir Starmer called you tomorrow and was like, "Hey, please come to my office, please give me one thing I should do about what is kind of like one big thing that you would want Keir Starmer to do regarding everything that we've talked about today?" You can maybe give him two things if you want. I can think of a thing I want Keir Starmer to do, but it wouldn't help. Yeah, fair enough, fair enough. So is there anything you'd like to add to the kind of previous discussion of what really could be done, and in particular maybe for all of us, right?

I mean, maybe we have to read the tone of the... What is to be done is a very good question. I don't know about the really, but what is to be done certainly is a good question. One of the points we come to towards the end of the book is actually that it's impossible to write a universal prescription for how everyone should act because everyone's positionality is somewhat different and people have different connections and links that allow them to do different things.

I happen to think that most of us in this room probably work for wages to some degree, and that work is overwhelmingly a source of our leverage. So when I am organising at work, I'm a member of UCU, currently the University of Essex is going through a

sticky situation as the rest of UK higher education. When I'm organizing the workplace around that, I also think about how I can use that power as a worker to address other questions. So whether that be like the university's complicity with BAE systems in Palestine and the genocide that's occurring there, whether that be with the way that tech companies might be interacting with our computer science department or whatever. So my emphasis, I mean, and this is partly my personal experience with organizing, is very much on organizing as a worker in the workplace.

I think most readers do not have the ear of Keir Starmer or the ability to design fantasy policy programs and even if we could design them and give them to Keir Starmer, the likelihood of him following them would be next to none given that we can't get the most basic employment rights legislation through the door. So I think what is to be done depends on who you are but asking that question and starting to think about it certainly and thinking about it with other people around you is the key.

- Thank you. All right, we'll turn back to the room. So we had a question, yes, over here. - Hi there, thank you all very much. I can also just sense the passion for this subject, which has really made this engaging. I'm here, I'm doing a master's at King's, I'm with a couple of classmates, I hope that doesn't get me kicked out of the room here. I'm interested to hear about how any of you feel that

this is getting positioned to people internal to these large tech companies, but then also stakeholders like investors or regulators. Because as someone just sort of sitting here kind of learning about it, it's easy to picture just a chain of people standing there beating each other on the head with a stick. But in reality, and that goes all the way from the CEO to the manager of that woman who

made her go and watch that awful content for the rest of her shift. But I have to think there's a more tactical way that these tech companies are going about positioning this from an operational perspective, from a PR perspective, from a legal perspective. Because I find it hard to believe that any one individual would want to perpetuate that cycle of exploitation, but ultimately it still happens.

And then I guess the second part to the question is how does that sort of compare to historical examples or is there really something kind of uniquely complicated and sinister happening here with this kind of new form of industry and exploitation? Thanks. Yeah, let me just start. So there are obviously people within the companies who...

Firstly, there are people within the company we visited that thought they were running a great kind of show. But there are also people who, even when they find out that things are going wrong, they want things to be better. And so I think Google took the lead on this in 2019 through the partnership on AI, which is kind of like an independent group that was formed by some of the large tech companies. They put out a white paper about...

conditions for, and they call them data enrichment workers. And so they had this big checklist and all these kind of things that you should look at. Maybe you think about pay. Where is it going? What's the relationship? There's a huge amount of work on human-computer interaction dealing with the kind of

biases and stuff the subjective biases of data workers that kind of get inputted into the models and might lead to problems with the model's output i mean just to give you one example though so in the company that we work for there was you know a kind of champion of this project and of kind of making the conditions better i'll call him as a pseudonym alex

And Alex probably had a master's degree in human rights and was really involved in the tech for good. And you imagine him as the kind of person who would really like to make a difference. The problem is, Alex was not the one calling the shots. So he's in the senior leadership team,

But the idea that any individual here, just through sheer force of their good nature, would be able to change conditions, I think, is just a misunderstanding of how this whole system, how we all get into this in the first place. Because the company is under pressure as well. I know you had the stick-beating analogy.

the people that are beating the company in the senior leadership team are the investors. Because they have taken venture capital money and the venture capitalists want to see like three times, five times, ten times their investment within five years. And so it's just not possible for someone like Alex who has a human rights degree who wants to do good

to really change this because of the structural incentives that exist both within the business and within the broader market. I've forgotten the second part of your question. Do you remember the second part? Is it uniquely more complicated? No. No, I don't think it is uniquely more complicated. I mean, I don't have the training in psychoanalysis you'd need to have a more in-depth analysis of what some of the people in these contradictory locations experience, but

I mean Marx talks about people as character masks right well like often we are kind of the bearers of relations that act through us regardless of our own personal intentions and there is a degree to which I think competition disciplines people like those with good intentions to the degree that they actually can't like even if they were supremely good moral people who made decisions it would be very very difficult to have any kind of like meaningful impact without immediately just being kicked out of the company because you'd be endangering the company's survival in its current form

So this is, I think, why we're so focused on structural change, because it's not that the people involved are uniquely evil, it's that structures involved are reproducing specific dynamics. And to change them, you need to act at the level of structure, not personnel. There's a question in the front, in the striped shirt, and maybe if you could just wait for the mic so everyone can hear you. Yes, very quickly, it's a comment in a way. I mean, why do you give so emphasis on AI...

First question. Second, I don't think it's so much AI the problem, it's the data. So this means we are talking about data and the linkage between data and AI and people and business is some kind of data that is circulated. It's a circular economy, we can say, where then you can balance data

all these type of issue and problem from social, economic, etc. So the question is simple for me. I think there is a lack of education. I mean, if you see even here at London School of Economics, the question of data science is recently, they start even at the schools,

So it's the domain of discipline, traditional discipline that does not want to change or to transform. That is my concern. So this means data is at the basis more than AI. It should be more a circular data where everybody can benefit. Thank you.

Thank you. Yeah, I think that's a really helpful comment that is very much in line with the book and the book's real emphasis on thinking about what we think of as the kind of AI system, the end product, but through the entire chain of building it all the way from the data through to the energy use through to the final product that consumers see. So that's super helpful. Thank you. We had a second question up here in the front. That's very interesting. Thank you.

When you mentioned Planet of Slums, that sort of triggered a lot of, put sort of things in place that these sort of captured populations. I just, I guess two questions is, I'm still unclear how we go from Planet of Slums, that narrative, to the narrative of these people in the

in the data centers. So, you know, there must be some kind of regulatory form or educational or state-al kind of apparatus that allows for these persons to then

even if it's sort of as routine, routinized as you say, so maybe you could just fill that in a little bit. The other thing which I thought was interesting was you did have this comment about EU and UN regulatory form that was governing over, to some extent, particular kinds of labor practices. And that struck me as well, is that just there as a generic thing?

governing over garment industry and AI, or are there other people hopefully around who know what you guys know? I just want to know, what are those regulatory intellectuals located in the ivory towers of the World Bank Group or UN? How are they being informed about the narrative that you've given us?

I'll say the first part and you can take the second. I'll leave you to talk about policy. Okay, so how do we get from Planet of Slums to work to this story? Let's take the example of Gulu. So Gulu is the town in northern Uganda where we went and did field work.

People remember Joseph Kony and the Lord's Resistance Army. That was one of the areas where during the Ugandan Civil War he was active and his regular armed force was active. So this is a society that has existed kind of on the periphery of Ugandan development for a long time, so not even necessarily particularly integrated

into the Ugandan economy during the colonial period. But during the Ugandan Civil War, the Ugandan government makes the decision to concentrate the population to bring people out of the countryside into the towns. And that's what creates Gulu, right? It is a refugee camp town. People who've been living in traditional Acheolian society, dispersed around a really large area of land, are suddenly concentrated into a town, right? And with it, you get the first mass contact with things like the cash economy, right?

with informal work, with work and wage work in general at all. So people go from living on family compounds with relatively little connection to the dynamics of global capital to suddenly being in a refugee camp, needing to work and not having access to land, and then going and finding work. Now, this is where SAMR comes in as one of the early projects to give work. The idea was we're going to offer work to people in this area.

and we're going to give them a way into, you know, earning their own living and developing over time. So you can imagine, I mean, there's some really excellent writing on Gulu during this period and how Ashiolian society goes through a profound transformation. People talk about a boda-boda lifestyle. A boda-boda is a motorcycle taxi, right? And basically a lot of people in Ashiolian society talk about the way that life has been completely transformed because the young men all want to work as like moped taxi drivers now, right?

And it's the number one way to get around Gulu and I never felt quite so badass as when we were doing research on the back of a motorcycle. It was a brief moment of glory. But so for a lot of these people, their transition from...

living in a form of society relatively disconnected from the cash nexus, relatively disconnected from contemporary capitalism, to working producing annotated datasets for big tech is an incredibly rapid one. It's like integration in the space of a number of years and it drives people straight from the economic periphery to producing value for Silicon Valley.

So there you have a story very much of a direct transformation. If you buy the book, you can read more about what that looks like in reality. But ultimately, it's not a story whereby that integration leads to a massive increase in quality of life.

- Yeah, thank you so much. And I think since we're running short of time, I would say it's a great second question as well about regulators and what they know. And I would just refer you back to all of the things that Kirsten mentioned, this difficulty of actually getting cross national data, governments aren't producing data that's directly comparable to allow us to do the kind of science we need.

as well as the limitations in the ways that many have thought about work and ignoring all of these both larger questions about how it impacts health, education, as well as these other kind of richer ways of conceptualizing work. So with that, I'm afraid we're at time. So I would just ask you to join me in thanking Colm, James, and Kirsten.

Thank you for listening. You can subscribe to the LSE Events podcast on your favourite podcast app and help other listeners discover us by leaving a review. Visit lse.ac.uk forward slash events to find out what's on next. We hope you join us at another LSE Events soon.