Meta Recruited Striking Actors to Train AI

Tech giants like Meta recruit on-strike actors for AI emotion studies, raising questions about compensation, rights and potential propelling the replacement of human actors by AI avatars.

In the midst of Hollywood's historical dual strikes by the Writers Guild of America (recently ended) and the Screen Actors Guild (SAG-AFTRA), Meta and London-based emotion AI company Realeyes found a unique opportunity. With the entertainment industry at a halt, a plethora of out-of-work actors became the perfect candidates to teach AI systems to mimic human emotions.

The "emotion study," which ran from July to September, sought actors for a two-hour shoot. These actors, including a young aspirant identified as T, were asked to enact various emotional scenes. Their performances were recorded not for human viewership but to train AI databases. This data, T later discovered, would be used to create "virtual avatars" for Meta and algorithms for Realeyes.

While the actors were compensated, the broader implications of their participation remained unknown. The study promised that no individual likenesses would be used for commercial purposes. However, the actors had to sign certain rights away "in perpetuity," allowing their data to be used for technologies that may not have even been invented yet.

SAG-AFTRA's chief negotiator, Duncan Crabtree-Ireland, emphasized the gravity of the situation, stating at the San Diego Comic-Con panel, "This isn't a contract battle between a union and a company… it's existential." The central concern is that with AI becoming increasingly sophisticated, actors — especially background actors — might soon find themselves replaced by avatars trained using their own performances.

Realeyes's vice president for growth and marketing, Max Kalehoff, clarified that their study was unrelated to the strikes and the timing was an "unfortunate coincidence." However, many actors, including a New York-based actor named Jessica, expressed concerns about the potentially exploitative nature of such projects.

The growing field of artificial intelligence is powered by data — often human-generated. Companies like Meta and Realeyes are shifting from scraping data from the internet to professionally building datasets, which approach is deemed more ethical, but it clashes with the publicity rights of actors. Before AI, actors controlled the use of their name and likeness. With generative AI, realistic replicas can be generated using just a handful of data, and thus actor’s control over the use of their names and likeness may be threatened. 

While the Realeyes study offered a substantial pay, especially for rising actors like T during a difficult time, the ambiguity surrounding the usage of their digital performances remains a cause for concern. Actors were essentially signing rights to their emotional expressions, potentially paving the way for AI entities that could one day replace them.