Tickl-AIing the Funny Bone: Exploring Humor, Language, and Artificial Intelligence
For the longest time, humor has been thought of as a deeply human trait that blends different qualities of social intelligence, creativity, and wit, given how cracking jokes is a masterful art requiring cultural skills, and linguistic knowledge, along with an astute sense of social awareness. This gives rise to some intriguing questions: Can AI crack jokes? Can it understand jokes and analyze why something is funny? Can it create a truly original joke that is rich in context, or is it restricted to making jokes about general things? More importantly, can AI be funny beyond generating jokes as output?
Even as artificial intelligence can diagnose rare illnesses, perform convoluted calculations, play games, and construct maps, with an extremely high degree of efficiency and precision, humor is still a human thing. This is because humor is predicated on the interpersonal relationships and social phenomena that exist between different groups of people: humor is how people bond with each other, and make sense of the world around them.
Therefore, can machine learning learn to deploy humor by ingesting vast amounts of data, working through algorithms, identifying common patterns, and therefore generating humor-rich outputs? The answer to this question is that yes it can. But to just spew out jokes is not necessarily to be funny. There are far too many people who can deliver the funniest joke in the world, and still not solicit a laugh. This is because comedy is often contextual and lies in the subtext. One needs to intuitively understand it. While one can be taught to explain a technical process, explaining a joke requires analyzing and deconstructing it, which is a deeply painful process because it takes the humorous punch of the joke away, and makes it extremely un-funny.
Further, because AI works through predictive analysis, it has to identify patterns and work in accordance with those patterns. This type of functioning goes against the grain of humor: because most often, the funnier a joke is, the more its punchline disrupts a listener’s unconscious expectation of what is to follow. Because AI excels at being formulaic, it lacks the ability to truly surprise or shock a listener.
In an experiment spearheaded by Naomi Fitter, an assistant professor in the School of Mechanical, Industrial, and Manufacturing Engineering at Oregon State University, scientists have come up with Jon, a tiny android who performs comedy when a button is pressed. And even though it tells the same jokes always in the same order, over time, Jon has learned to respond to its audience by varying the time between jokes, the length of the audience's laughter, and even delivering small one-liner quips about the joke based on audience feedback.
Therefore, even as AI can generate jokes that might even be good enough as filler material on a comedy show, AI neither understands what makes jokes funny nor can being funny be described as one of its qualities. This also helps us think about how language is a social phenomenon, embedded in a shared context, and how machine learning can approximate that context, but can never subsume it.
Sources