In today’s Tech@Work, a company in Finland is having incarcerated people train its AI model; and democratic lawmakers are probing tech companies that use data labeling ghost work.
As reported in WIRED this week, a Finish startup called Metroc is using prison labor to train its AI model. The startup has created a search engine aimed at helping construction companies find new building projects online. Their idea is to use a large language model to distinguish between open versus closed bids and work opportunities in online municipal documents and news articles. To do so, they need people to manually label tons of construction bid data, so that the model can “learn” to do this itself.
As we wrote about this past summer, companies like OpenAI (who makes ChatGPT) outsource this data labeling work to a company in Kenya, where English is commonly spoken and wages are comparatively low. This work has turned out to be traumatizing, exploitative, and highly underpaid, and workers have been organizing to change working conditions, and have brought litigation against both the local and American companies. As the WIRED article points out, it is likely that the Finnish startup has looked to incarcerated people within Finland because very few people outside Finland speak Finnish. While the content of the data labeling is not traumatizing in the same way that the ChatGPT labeling work is, the fact of Metroc using people in prison to do this work speaks to its low-paying and potentially exploitative nature. Some say this work is a good alternative to other jobs available in prison. Others quoted in the article worry it offers no skill building opportunity. In any case, the use of prison labor in this instance highlights a more general phenomenon wherein technology that is automated on the surface requires large amounts of human labor behind the scenes. This “out of sight” element may only increase the risk of exploitative conditions or dynamics.
Responding to this general phenomenon of exploitative labor behind AI, Bloomberg reported last week that Representative Pramila Jayapal (WA-07) and Senator Ed Markey (D-Mass.) have written a letter to nine tech companies asking for transparency around the conditions of workers similar to those in Finland in Kenya. They characterize the situation as one in which workers “are often paid low wages and provided no benefits but keep AI products online by completing tasks such as labeling training data and rating chatbot responses for accuracy and safety.” The letter, which was signed by Senators Bernie Sanders (I-Vt.), Ron Wyden (D-Ore.), and Elizabeth Warren (D-Mass.) and Representatives Jamaal Bowman (NY-16), Katie Porter (CA-47), and Mark Pocan (WI-02), states that “tech companies must not build AI on the backs of exploited workers.”