THEaiTRE: How AI Dialogue Became a Play in 2021

Photo on the right depicts Rudolf Rosa who is part of the research team

What happens when a language model supplies dialogue and theatre professionals bring it to the stage? THEaiTRE brought researchers from Charles University together with Švanda Theatre and the Academy of Performing Arts in Prague to explore that question. Their project culminated in AI: When a Robot Writes a Play, which premiered on 26 February 2021. It began with a reversal of a familiar story: a century after a writer imagined robots in a play, a computer system would help write a play about human beings.

The team used GPT-2 to continue dialogue from a supplied scene setting and opening lines. The anniversary of R.U.R. supplied the project’s inspiration, rather than a requirement to produce a biography of its author. Karel Čapek wrote R.U.R.; his brother Josef is credited with suggesting the word robot. That distinction matters because the play’s authorship and the origin of its now-familiar term are related but separate parts of the story.

Rudolf Rosa, who is part of the team, said:

"The main idea behind our study came from Tomáš Studeník, an innovator who noticed that the 100-year anniversary of the theatre play R.U.R. is approaching. This was a key moment for robotics, as the idea of a robot, including the word 'robot' itself, was invented by Karel Čapek and his brother Josef, who wrote this play. Tomáš thought that this should be properly celebrated and came up with the idea of turning the story around: 100 years ago, a man wrote a theatre play about robots; what if today, robots wrote a theatre play about men?"

The collaboration made the division of creative work part of the experiment. Rather than treating human involvement as an invisible finishing stage, Rosa described it through the “human-in-the-loop” concept:

"Basically, all computer-generated 'art' (if we want to call it so) is touched and retouched by humans in some way, but often this is not very transparent. In our work, we are trying to be very explicit on what the machine does and what a human does, making their cooperation an integral part of the system design, not a post-hoc fix."

The project was one of several contemporary attempts to explore machine-generated art. Its distinctive question was theatrical: could generated dialogue sustain a performance when spoken and interpreted by actors? Language generation supplied material to work with, while the rehearsal room tested what could be made of it. The result depended on those encounters between text and performance, not simply on whether a computer could produce grammatically plausible lines.

Rosa went on to explain the creative process:

"When we fed GPT-2 a scene setting and a few lines of the drama script, it generated further lines in the same style and focusing on the topic of the input script chunk. This way, we did not have to train anything (yet), as we restricted the generator a bit to keep to the task and not to diverge elsewhere. We can thus make use of the great large GPT-2 model trained for a very long time on very large texts, which we could not afford ourselves to train on our hardware, as only the largest tech companies can train such models nowadays."

Rosa’s account describes the system at that stage of the project. His comments about training resources and available models belong to that historical setting. The quotations capture a practical concern that shaped the experiment: how to make a general text generator continue a dramatic scene without losing its direction.