Data labellers build AI in secret


Behind every powerful AI system are huge numbers of people labelling and clarifying data to train it, contracted by companies like Remotasks, a subsidiary of Silicon Valley-based data vendor Scale AI, whose customers include the US military and OpenAI. Often the workers, who are assigned tasks they don't understand for a purpose they don't know, are sworn to secrecy. Yet labelling is crucial; it can make the difference between a car stopping to spare the person walking a bike across the road or misidentifying them as road markings. Tech companies expect that all these jobs, including those of the many thousands of people around the world who fill in the gaps in AI systems, are temporary, soon to be rendered unnecessary by technical advances.

Publication: The Verge

Publication date: 2023-06-20

Writer: Josh Dzieza