Hidden humans make artificial intelligence

Human raters have played a significant role in the rapid improvement in the machine learning models that fuel modern AI. The raters evaluate the algorithmic output of search engines and AI chatbots and provide "Reinforcement Learning with Human Feedback" (RLHF) – the technical name for the deployment of such ratings to improve AI models. The efforts of these workers, who are mostly located in the global South but include thousands in the US, is downplayed by the technology companies to whom they're subcontracted, and are poorly paid and high-pressure, while subcontractor companies such as Appen reach billion-dollar valuations supplying their work to companies like Google, OpenAI, and others. Raters working for Appen have begun organising with the Alphabet Workers Union-Communications Workers of America in order to win greater recognition and better pay and working conditions. Appen fired several raters active in organising or speaking out about conditions, reversing the decision when AWU-CWA filed a complaint with the National Labor Relations Board.

https://www.theatlantic.com/technology/archive/2023/07/ai-chatbot-human-evaluator-feedback/674805/

Publication: Atlantic

Publication date: 2023-07-26

Writer: Matteo Wong