The Invisible (and Underpaid) Data Workers Behind the "Magic" of AI

Who are the invisible human data-workers behind the “magic” of AI, and what does their work really look like?

In this episode of THE PEOPLE'S AI, presented by Vana, We pull back the curtain on AI data labeling, ghost work, and content moderation with former data worker and organizer Krystal Kauffman and AI researcher Graham Morehead. We hear how low-paid workers around the world train large language models, power RLHF safety systems, and scrub the worst content off the internet so the rest of us never see it.

We trace the journey from early data labeling projects and Amazon Mechanical Turk to today’s global workforce of AI data workers in the US, Latin America, Kenya, India, and beyond. We talk about trauma, below-minimum-wage pay, and the ethical gray zones of labeling surveillance imagery and moderating violence. We also explore how workers are organizing through projects like the Data Workers Inquiry at the Distributed AI Research Institute (DAIR), and why data sovereignty and user-owned data are part of the long-term solution.

Along the way, we ask a simple question with complicated answers: if AI depends on human labor, what do those humans deserve?

Timestamps:

• 0:02 – Krystal’s life as an AI data worker and the “10 cents a minute” rule
• 2:40 – What is data labeling, and why AI can’t exist without it
• 6:20 – RLHF, safety, and the hidden workforce grading AI outputs
• 9:53 – Amazon Mechanical Turk and building Alexa, image datasets, and more
• 14:42 – Labeling border crossings and the ethics of unknowable end uses
• 25:00 – Kenyan content moderators, trauma, and extreme exploitation
• 32:09 – Turker organizing, Turker-run ratings, and early resistance
• 33:12 – DAIR, the Data Workers Inquiry, and workers investigating their own workplaces
• 36:43 – Unionization, political pressure, and reasons for hope
• 41:05 – Why humans will keep “labeling” AI in everyday life for years to come

The People’s AI is presented by Vana, which is supporting the creation of a new internet rooted in data sovereignty and user ownership. Vana’s mission is to build a decentralized data ecosystem where individuals—not corporations—govern their own data and share in the value it creates. 

Learn more at vana.org.