AI Attempts Improv Comedy And It Goes Hilariously Wrong

Besjunior/Shutterstock

Rosie McCall 17 Sep 2018, 13:07

A human, a puppet, and a cyborg walk into a theatre... and attempt improv comedy. 

Your friend may find Adam Sandler movies hilarious while Chris Rock's standup has you in a fit of giggles. Needless to say, everybody's sense of humor is (at least slightly) unique. However, when artificial intelligence (AI) stepped up to the mic for a project led by the University of Alberta, its sense of humor fell flat. Let's just say it wasn't very good by anyone's standards.

Kory Mathewson, a PhD candidate at the University of Alberta with the Alberta Machine Intelligence Institute, and Piotr Mirowski, a drama school graduate, oversaw the project, which put a new spin on the Turing test – first developed by Alan Turing in 1950 as a way to assess a machine's capacity to display intelligent behavior. Their plan was to convince an audience of punters that their artificially intelligent improv comedian (called A.L.Ex) was, in fact, human. 

The study involved a set-up inspired by a short comic play by Christopher Durang called The Actor's Nightmare, which requires actors to fit inappropriate lines into the scene.

"One actor might be reading War and Peace, and the second has an objective to ask for a divorce, and keep the Jaguar," Mirowski told New Scientist.

"Obviously War and Peace and an acrimonious divorce don’t really make sense together."

Except in this particular performance, only one of the three actors was a human with the free will to "write" their own lines. The second was a puppet whose lines were decided by a human off-stage, while the third was a cyborg whose lines were machine-generated (via A.L.Ex). A.L.Ex had been trained on movie subtitles so that it would produce credible and context-based dialogue during the performance, though a curator was also in the wings to input screen prompts and then choose the best reply.

Both the audience and the actors were asked to determine whose lines had been human-generated and whose had been machine-generated. From the results, it looks as though it may be a while before AI bots win this particular Turing test – the audience was able to identify the cyborg every single time. (There were six performances in total.) Interestingly, on two occasions they thought there was a second cyborg.

Full Article
Comments

If you liked this story, you'll love these

This website uses cookies

This website uses cookies to improve user experience. By continuing to use our website you consent to all cookies in accordance with our cookie policy.