Improbotics is a tech-infused improvised theatre and comedy show and a live Turing test-based scientific experiment. An actual artificial intelligence-based chatbot is performing in the show and tries to pass as human as it sends lines to one of the improvisers via an earpiece. Our impossible and hilarious challenge is to attempt to justify, physically and emotionally, AI-generated lines that may make no sense at all.
Improbotics was co-created by robotics researchers Piotr Mirowski (UK/France/Poland) and Kory Mathewson (Canada), lated joined by drama director Jenny Elfving (Sweden), science communicator Ben Verhoeven (Belgium) and communications and digital media expert Boyd Branch (US). Our show combines ideas from HumanMachine‘s Artificial Intelligence Improvisation, and the classical improv game called Actor’s Nightmare or Lines From a Play.
Improbotics has been featured in the New York Times, Wall Street Journal, New Scientist, RTÉ One, Globals News Canada and Bloomberg.
**** (Binge Fringe) and Most Innovative Show Award (Paris Online Fringe 2020).
***** “If I were introduced to a show like this as a kid, I would definitely have paid more attention to science class” (Phoenix Remix, Brighton Fringe 2019).
**** Edmonton Fringe 2018.
“This could be revolutionary” (Broadway Baby, Brighton Fringe 2018).
“I will stick with artificial stupidity” (Colin Mochrie, Edinburgh Fringe 2017).
Why are we doing this?
Improvisation is about risk, challenge, and cooperation. Improvisers take on seemingly impossible tasks, and through their skill and open mind, surprise themselves. In this show, we want to explore how human improvisers could seamlessly perform when a machine or another human feeds some of them lines via an earpiece. We want the team to play together in a single, grounded narrative improvisation, despite the challenge.
Similarly to the Actor’s Nightmare improv game, the AI-controlled improviser (Cyborg) needs to justify their lines as well as they can. However, in this case, their lines are interactive and depend on the context of the improvised scene. The other improvisers, who are not fed lines, need to make the scene look natural. All the improvisers, especially the Cyborg, need to play grounded characters. The AI chatbot, who listens to the conversations on the stage and interactively produces the next line for the Cyborg will probably be, at times, nonsensical. The free-will improvisers, who do not get lines fed to them, need to create strong dialogue. Or they can, to the contrary, pretend to be the AI and through deception, introduce a wildcard into the audience’s mind.
We want to push the deception as far as we can, and to encourage the improvisers to act in the most natural and intelligent way, and to play to the full range of their emotions and physicality. This setup gives many potentials for strong comedy.
The technology that enables remote control of a human player consists of a laptop, an FM radio transmitter/receiver with headphones, and our custom-made software that allows an operator to type sentences that will be sent to an artificial intelligence-based chatbot (based on HumanMachine‘s A.L.Ex, currently built around the GPT-2 model that we retrained on movie dialogue from OpenSubtitles and, more recently, on the GPT-3 model), that will in turn generate sentences to be said by the Cyborg improviser. All the performers wear headphones, and the audience typically does not know who is controlled by an AI, and who relies on their free will to speak as they wish. A small humanoid robot, produced by EZ-Robot and running our custom software, is the host of our shows. For our latest show, Rosetta Code, we also rely on Google Translate and speech recognition in the Chrome browser.