Firms are assembling teams of performers to generate specialized training datasets for AI laboratories.
Top artificial intelligence enterprises have begun actively onboarding improvisational and theatrical actors to produce the necessary training materials for refining models that generate and recognize human emotional states. A notable contractor in this space is Handshake AI, which supplies data, among others, to OpenAI.
Job postings reveal that actors are invited to take part in compensated virtual sessions. During these sessions, they collaborate with fellow performers to enact prescribed scenarios and emotional vignettes. The objective is to gather instances of genuine expression, emotional transitions, and the conveyance of intonation nuances that are difficult to replicate using standard text-based datasets.
These organizations point out that current language models, despite achievements in various complex domains, often exhibit an “inconsistent” performance level: while capable of handling technical queries, they falter in straightforward everyday or emotionally laden contexts. Filling these gaps necessitates specific annotation and novel forms of data input.
Handshake AI and its rivals, including Mercor and Scale AI, are expanding their contractor networks to incorporate experts from diverse fields—ranging from legal professionals and physicians to screenwriters and actors. According to reports from The Verge, the demand for these services has surged multiple times over the past year, with Handshake AI achieving a turnover of $150 million by November 2025.
Some performers harbor concerns that their involvement in these projects could inadvertently hasten automation and subsequently result in job displacement within creative and knowledge-based professions. Nevertheless, companies persist in scaling up data volumes to enhance AI’s capacity to identify and reproduce the intricate elements of human interaction.
About the Author
