From AI Gig Work to Fake Papers

Reading Karen Hao’s excellent Empire of AI: Dreams and Nightmares in Sam Altman’s OpenAI, the most infuriating portions — and it’s a high bar — come in her descriptions of the exploitation of gig workers in Africa and South America to annotate, summarize, and describe content for LLMs and the AI tech stack, how this is outsourced through various companies that provide these services, and how the workers are discarded as expendable once their tasks are completed, leaving many with no options and only sporadic hopes for more work.

The first population to be exploited was in Kenya, where companies like Sama leveraged the country’s online infrastructure and impoverished workforce to get a lot of content tagged and created in what Hao calls the “data labor supply chain.”

This appears to have either initiated or worked in conjunction with another form of gig work — the writing and translation of scholarly and scientific works, with a 2024 documentary noting that as of 2024 up to 72% of all online labor work in Kenya involved writing for students in Global North universities.

The person behind the documentary — Patricia Kingori — estimates that 40,000 workers in Nairobi alone are writing papers for people in these universities. Whether they are also fake science papers is unknown, but it strikes me as likely. How many of these end up in high-volume OA journals is another question that immediately springs to mind. After all, the numbers don’t really add up unless there is a lot of manufactured science in our midst.

My head swims with this on a few levels:

• The historical exploitation of Africa by the Western world continues, but in the shiny guise of Big Tech
• The circle from annotation for AI systems to using AI systems to write more papers faster to be annotated for AI systems is an Ouroboros distorting reality
  • Hao tells of AI workers fired for using AI systems to speed their work filling AI systems, so this is an established and predictable practice
• Platforming of content for money via Gold OA has driven quantity-based publishing, creating incentives for sweatshops to create papers for academics who want rapid publication in journals more than happy to be paid to provide it
• Large publishers are creating their own AI systems and outsourced editorial services, and it’s unknown whether this is all the same laundromat — that is, the outsourced AI tagging shop also generates fake papers and then turns around and processes these when the publisher pushes them through their systems

Lately, John Wiley & Sons — a company that built their business on being academia- and society-friendly — has been making a lot of noise about “ethical” and “responsible” use of AI. These terms mean nothing in the larger culture and are cheap lipstick on the hallucinated pig of AI, especially if the systems were themselves assembled unethically, a topic Wiley fails to touch on, instead preferring to preach behaviors to the rest of us while washing their hands of the AI space’s rapidly degrading reputation and behaving as if adoption or acceptance of AI is inevitable:

As stated by Jay Flynn (Wiley EVP & General Manager, Research & Learning), “writers and researchers are already using AI tools, whether publishers like it or not. At Wiley, we’d rather embrace this shift than fight it.”

Um, I’ll stay on the “fight it” side, thank you, Jay. Allow me to explain why . . .