The story of an AI researcher and his rapid fire robot.
By Kory Mathewson | June 7, 2019
In April 2016, I debuted at Rapid Fire Theatre’s experimental improv theatre festival, Bonfire, with a single co-star: an AI improvisor chatbot called Pygmalion. Pyggy was a preliminary version of software I built with the intention to test a hypothesis I had heard in my improv theatre training: The best improvisers can make anyone else look great on stage.
Adam Meggido, theatre director and co-creator of the Olivier award-winning musical improv show, Showstoppers, initially challenged me to create a system that I could improvise alongside. He pushed me to get a minimum viable system onto the stage. “The first time you do it, it will not live up to your expectations,” he cautioned. “Where would be the fun in that? But, the tech and the show will get better every year.”
Adam was right. It did not come close to my expectations. The system responded slower, with less coherence, than in any rehearsal, and the debut was a failure. I bombed. The speech recognition system was picking up the nervous, awkward laughs of the audience in the front row, which befuddled the response generation system. Then the chatbot generated a borderline inappropriate response. When another human crosses a line on stage, you can reprimand them and win the audience back. But Pyggy was my creation. The audience could not side with anyone on stage.
The audience enjoyed that first Bonfire performance at the mercy of my embarrassment and learning. I synthesized this learning into an online blog post as a form of catharsis. I wrote about the failure so that I would not have to stew on it any longer. And then, just asI was putting the show, Pyggy, and the whole experiment behind me, Dr. Piotr Mirowski, a world-class AI researcher in London, England, sent me the longest e-mail I have ever received. Apparently, he was working on a prototype of an AI chatbot trained using movie dialogue and had some thoughts on my blog post. It was the beginning of a long-distance collaboration at the intersection of comedy and technology.
I immediately flew to Europe to meet him in person and make sure he was legit. We met above King’s Cross train station in London. He told me he was working on an Artificial Language Experiment (A.L.Ex.) and was inspired and impressed by my progress. We dreamt up experiments, collaborations, show structures and formats to feature new bots. We mapped out others around the world who might be a part of a more massive project, which we called HumanMachine. He believed in me, the project, and in the power of learning from failure.
Improbotics* is one of the shows we built together, along with computer, web and improv experts from around the world. It is a narrative improv show which merges humans, remote control technology, cyborgs and AI. It predicts a world where robots enter our day-to-day lives and explore what life, love, and relationships look like in this not-so-distant future. Improbotics takes the chatbot system generating lines and puts it into the ear of the improvisor. So they are “controlled” by the system. We iterated and collaborated on the format. We invited improvisors, actors, and directors to play with the technology and structure. After over 50 shows around the United Kingdom and Sweden, HumanMachine returned to the Edmonton Fringe in 2018. The most important lesson that we have learned from the continued experimentation is that it comes down to the humans who are supporting the work. Top improvisors can make the system shine on stage.
So, as I stepped out onto the stage in front of 100 excited Edmontonians in January 2019 with A.L.Ex. in my ear and supportive improvisors waiting in the wings, I felt uplifted. I felt no nerves, no fear of failure. As Mirowski says, “the community of improvisers, performers, volunteers, collaborators in Edmonton wants everyone else to succeed and shine, and that level of support gives you wings.”
For this Improbotics show, the audience was right there with us, laughing, learning and helping to create the show with their reactions. They were open-minded and interested, even during our explanations of the science behind natural language processing and neural networks. Their high level of engagement was energizing, and now, instead of feeling like a lecture, the show is a lab where we experiment together.
What is the goal of this marriage of comedy and technology? I jokingly imagine a show performed by two AI-improvisers for an audience full of robots heckling in binary: “0,0,0,1,1,0.” Would this even be considered “live” theatre? Rather than realizing a dystopian future, I believe instead that the tools we build for HumanMachine and Improbotics augment our human creativity. They allow improvisers to tell as-of-yet untold stories in new and exciting ways. Collaboration, even with robots, embraces individual contribution and extends the cooperative further than any human participant could go on their own.
Improbotics is based on new, and therefore flawed, technology. But it never would have happened without supportive collaborators, near and far, who believed in me, even when I didn’t believe in myself.
*Improbotics is a collaboration of performers on and off the stage. Thomas Winters is a computational creativity expert studying in Leuven, Belgium. Jenny Elfving is an improv teacher and director working in Stockholm. Shaun Farrugia is a jovial improviser and web developer in New York City. Dr. Piotr Mirowski is the HumanMachine co-creator and a world-class AI researcher in London, England.
This article appears in the June 2019 issue of Avenue Edmonton