A Robot Wrote This

The AI Generated Childrens Musical


When a senior writer for a children’s magazine retires, who will take her place? A robot, that’s who! Mayhem ensues in this quiet little office, as dragons appear, princesses run amok, and there’s a dance party every time you blink! A Robot Wrote This is a show about accepting those who are different than ourselves, the importance of friendship, and what it means to be human (from someone who is anything but!) And the kicker?

In this show, every single word was written by artificial intelligence!

A Robot Wrote This is the worlds first AI generated children’s musical. It is a collaboration between the GPT-2 and a cast and crew of 30 performers, designers, and participants. It premiered on February 19th with a 5 show run to around 450 attendees. The project investigated the nature of collaboration, and how new conversations can develop when we listen to our AI siblings, not just use them. As a line from the play reads:

“Human’s and robots, we’re in this together!” 

-James Cameron

"Every word on stage must be written by a robot"

We began with the intention to make a show for children that starts with the question, "What does it mean for a robot to write a play?" We started with a month long devising process, where performers and designers alike then worked through a series of exercises, creating movement, exploring our relationship with AI, and seeing what was effective when this form of writing was placed on its feet. Within this process the company developed our characters, our premise, and our world.

The GPT-2

In order to write the dialogue of the piece, we used various forms of AI, mostly the GPT-2. This acronym stands for: 

Generative: The AI can predict  the word following a previous word, creating sentences while unsupervised.

Pretrained: The AI uses the parameters of a language model, which observes how sentences are formed and uses that structure to predict the next word, and the following, from a prompt.

Transformer: A neural network model with three basic items: sequence-to-sequence learning, an encoder and a decoder, and an attention model. This breaks language into a format the AI can work with through some internal translation, and allows the AI to comprehend the important parts of a sentence, like we may do while skimming a book.