almost parallel or contend with expert human comedians.” So in 2018, Jon the Robot made his stand-up comedy debut at an outdoor open mic in Los Angeles. Two years later, Jon and Fitter performed 22 shows in the greater L.A. area to gather data on how the comedic timing of a robot can impact its success among audiences. It turns out robots are funnier when they let the audience react to a joke. Although Jon’s jokes are written by Fitter and preprogrammed in a particular order to maximize comedic effect — punctuated by the small gesticulations he makes with his arms and hands — the small robot is considered autonomous because he makes use of adaptive timing. In other words, he can “read the room” by listening to the audience’s response, whether it’s applause or laughter, before he moves on to the next joke. He can also deploy machine-learning tactics in decision making, which involves adding a programmed “tag” following a joke, based on whether he gauges a negative or positive response from the audience. For instance, if Jon senses that a joke was successful, he might offer the quip: “Don’t you hate it when you’re trying to solve inverse kinematics equations to pick up a cup and then you get ‘Error 453, no solution found’? Don’t you hate that?” It means he’s got the audience on his side. Conversely, he might add: “Not a lot of robots in the crowd, I guess. That joke kills at Best Buy” if the crowd falls flat. As a stand-up comedian herself, Fitter says she’s been fascinated by how objective she can be when assessing the robot’s comedy wins and fails. “When I try to do this with my human comedy, like look at if a particular joke or different styles of jokes succeed or fail, it’s kind of excruciating,” says Fitter. “But with the robot, you can cue up [the performance] and it’s less risky than in the moment. Then you can get some full objective data on what tactics or what material is making what kind of impact, for better or worse.” Kory Mathewson agrees. He’s a senior research scientist with Google DeepMind, whose work focuses on human and machine creative collaborations. “Artificial intelligence can help us decode, analyze and understand humor,” says Mathewson. “This can lead to more sophisticated and personalized comedic experiences.” So what, exactly, goes into writing from a robot’s comedic perspective? As someone with comedy training and stand-up experience, Fitter says character and world-building exercises were really the launching point. “What is romance for a robot? What’s grief for a robot? What’s drugs for a robot?” Fitter asked herself. “So when I was starting to write, I tried to think about what are the most hack jokes that humans tell, and then I kind of mapped them to the robot’s perspective.” As gender disparity in professional comedy is still a weighted issue, Fitter explains that her choice to make Jon male-presenting was intentional but not without careful consideration. The robot was modeled after a typical Silicon Valley engineer, in regard to his monotone humanoid voice and commonplace name. While she did run tests with a female-presenting voice, Fitter found that it sounded almost “too human-like” and wanted to retain the robotic quality of the small performer. Moreover, in a study with university students, Fitter was relieved to conclude that “when you do modulate the gender presentation of the voice, there wasn’t a significant decline [in enthusiasm] that we could identify.” Roboticist and researcher Naomi Fitter with Jon the Robot Naomi Fitter and Jon perform standup in Corvallis. 19
RkJQdWJsaXNoZXIy MTcxMjMwNg==