Handshake hires improv actors to teach OpenAI human emotions

Handshake AI has posted a job opening for improvisational actors — not for theater or cinema, but for training language models. A contractor for OpenAI and other leading AI laboratories is seeking people capable of authentically conveying emotions and maintaining character throughout an entire scene. Handshake AI is one of the few companies specializing in producing niche training data for AI laboratories.

The company collaborates with OpenAI and a number of other major market players. The position involves not standard text annotation, but live acting work: spontaneous reactions, improvised dialogue, convincing transmission of complex emotional states. This is fundamentally a different level of requirements for performers.

Previously, internet texts and relatively simple annotated datasets were sufficient for training language models. But as model capabilities grew, requirements changed. Voice assistants, support agents, systems for role-playing games, and therapeutic chatbots — all of them need not just grammatically correct sentence formulation, but also to capture subtle emotional subtext, react to sarcasm, convey confusion, genuine joy, or anxiety.

This is the hardest thing to simulate with synthetic data. Improvisation as a genre is perfectly suited for this task. Improvisational actors are trained to react instantly without losing character, maintain the emotional tone of a scene, and do so convincingly — without prepared lines.

This is exactly the type of live, unpredictable human speech that is so lacking in standard training corpora, where dialogues often sound sterile and formulaic. The hunt for human uniqueness is becoming a notable industrial trend. In recent years, AI companies have already hired writers, lawyers, doctors, programmers, and translators — anyone capable of providing quality examples of professional thinking and speech.

Now it's the turn of creative performers. Apparently, Handshake's clients have formulated a specific request: simple "diverse dialogue" is no longer enough — they need live, emotionally rich improvisation. How much is paid for such work and in what exact format it takes place — the job posting does not specify.

However, the very fact of its appearance is telling. AI companies are increasingly understanding that the next frontier of model quality lies not in scaling the volume of data, but in its character. The number of tokens in training corpora has long surpassed trillions — now what matters is not the number, but the subtlety.

For actors, this is an ambiguous opportunity. On one hand — real income for a unique creative skill. On the other — participation in creating tools which, in the view of many creative professionals, will ultimately displace them from the market.

Discussion of the ethical aspects of hiring people to train AI, which potentially threatens their professions, has been ongoing for a long time, but has not yet led to meaningful changes in regulation or practice. Handshake AI's activity signals that the race for human data quality is accelerating. The more realistically language models imitate human speech — the more live and specific the material for their training needs to be.

And as long as this loop remains closed, improvisational actors have work — even if its meaning stirs mixed feelings in them.