Future of Theatrical Production: Plays Written By AI

Plays written by AI, the technology used, and feedback by viewers.



Technology is evolving in such a manner that it has almost replaced humans in many fields of work. Not only IT companies, fintech industries, healthcare or other industries are replacing humans with AI but also in theatrical production houses, literature and plays are now being written by AI. This article emphasizes the plays written by Artificial Intelligence, the technology used, and the reaction of viewers.

Karel Čapek, a Czech writer, came up with a play written by AI. Together with his colleagues from the Institute, he developed a system for the automatic production of play scripts based on machine learning. He used the language model GPT-2 from OpenAI which processed 40 GB of text from eight million web pages. The play was written 90% by the AI and only 10% was fixed by humans. The team provided the AI with only an introductory scene and the rest was produced automatically.

The play includes eight scenes describing a robot’s journey in the human world and it’s a mixture of black comedy and existential drama. David Košťák, the dramaturgist said that it was surprising to know what the play says about humans. It shows that human’s most common topics are violence and sex. The play also reflects the time when it was written. 

The team realized that it’s relatively easy to produce a play script but the main challenge was to teach the AI that characters in a play are living with their own goals and motivations. Also, the system is trained with general texts with only one speaker and not trained for dialogues with more people. 


How Does it Take Place? What is the Technology Behind it?

At the young Vic theatre in London, an experimental production called AI played a three-night run. It is a 30-minute script written by AI. viewers attended the show to experience a play written by AI. They expressed that AI was an exploration of what happens when technology is implemented in the creative process. 

There is a system that is used to produce content live on stage called Generative Pre-Trained Transformer 3 (GPT-3), a language model capable of generating high-quality written content in pretty much any form is incorporated in the process. 

Developed by a company called Open AI, GPT-3 bases its output on text prompts, which decide the topic, format, and style of the content, as well as any characters that might emerge. GPT-3 was trained on a massive library of information derived from the open web. Until recently, it was the largest AI model ever designed, built on 175 billion parameters (variables whose value is drawn from the training data).

During the performance, text prompts and GPT-3’s output was exhibited on the pyramid above the stage. GPT-3 is capable of remarkable precision. 



According to the viewers, the individual scenes were colorful. Also, they highlighted certain limitations of the AI models. They mentioned when individual plays were threaded together it became an unclear collage. In traditional ways, scenes are dodged together in a very conscious way but here the scenes are loosely connected. The play reflected imperfect humans. The commotion is more like looking into a circus mirror.