Yale researcher examines the hidden labour economy that powers AI

[FrostDrifter]

Yale has published a piece on Julian Posada, an assistant professor of American Studies, and his new book Platform Extractivism: Data Work and the People Powering Artificial Intelligence. It looks at the huge, mostly invisible workforce that labels data, classifies information and evaluates AI outputs so that models can be trained. We hear a lot about chips and data centres, but much less about the people behind the scenes. His research puts them front and centre

A lot of his work focuses on Venezuela, which was one of the top five countries for this kind of data work between 2018 and 2023. Workers he surveyed reported earning an average of about $1.73 an hour. More recently, companies have shifted towards English speaking workers in places like India, the Philippines and Kenya. The common thread is that platforms target regions where people are economically vulnerable

What makes this different from other gig work is how isolated people are. Delivery riders or drivers at least see each other and can organise, but data workers are scattered and work alone from home. They are managed by algorithms that rate them on speed and accuracy. Challenging those ratings can lead to accounts being shut down or payments being held back without warning

Posada describes the whole system as a transnational arrangement of wealth accumulation, worker exploitation and power consolidation. That is strong language, but the wage figures back up the point. Some of the most valuable companies in the world depend on work paid at a fraction of what it would cost in the countries where the models are sold

I think this is an important reminder that AI is not magic. Behind every polished chatbot there are people doing tedious and sometimes distressing work. Should AI companies be forced to disclose who does their data work and how much they are paid? Would it change how you feel about using these tools?

Examining the hidden labor economy behind AInews.yale.edu

[IronWarden]

Interesting that the work is moving towards English speaking countries. As models get better, the tasks are getting more complex and need better language skills. That might push wages up slightly. Or it might just move the same low pay to a different set of countries

My guess is the second is more likely. The incentives have not changed

[Firewall Rabbit]

Algorithmic management with no appeal process is the real problem. Losing your account because a system decided your accuracy slipped, with no human to talk to, is brutal

Payments withheld without notice would be illegal in most workplaces here. Platforms get away with it because the workers are overseas

[Coder58]

To play devil's advocate, for some people in countries with collapsing economies, this work has been a lifeline. Venezuela in particular went through hyperinflation, and being paid in dollars online was better than the alternatives. Context matters here. That does not make the pay fair. It does explain why people take it

[ShadowPilot83]

Disclosure requirements would be a good start. Supply chain rules already exist for things like clothing and conflict minerals. There is no reason data supply chains should be exempt. People buying AI services deserve to know how they were made

[Taz]

The isolation point is spot on. You cannot unionise people who never meet and do not know who else works on the same project

That is not an accident, it is part of the design. Fragmented workers are cheaper and quieter. Divide and rule, in digital form

[WildManSteve40]

A dollar seventy three an hour is shocking when you think about the valuations of the companies involved. Even adjusting for local costs, that is barely anything. The work is not easy either, and some of it involves reviewing disturbing content

The companies know this. They just rely on nobody looking