Recently I came across an interesting field: computational humor, a subfield of computational linguistics focused on creating computational models of humor, and making a computer generate jokes.
First of all, what is the motivation for studying this field? Many researchers offer practical reasons, citing the benefits of humor in childhood development, in the workplace, and in coping with tragic events. There is also the obvious reason that it is fun to watch computers perform, or at least attempt to perform actions that are generally thought to be exclusive to humans. For example, it was at one point thought that computers would never be able to beat grandmasters in chess, and today the best grandmasters stand close to no chance against the best computer chess player, AlphaZero. What if we could have computers tell better jokes than humans?
One of the main differences between teaching a computer to be good at chess and teaching a computer to tell funny jokes is that there is no clear formula for humor, while the game of chess can be calculated to find the best and worst moves. Researchers are still studying what exactly makes certain statements funny and why humans laugh at them. There is no definite rule or equation for humor–unfortunate for computers, built solely to unrelentingly follow exact rules and commands. Many of the challenges computational humorists face are similar to the ones that computational linguistics face. Whereas humans can use inference to learn what is funny and what is not, researchers must explicitly tell computers what is funny and what is not, and they must go further to tell computers how to construct a joke from nothing. In essence, researchers need to feed computers the “rules” of humor when there are none, or at least they are too intangible to explicitly state.
Attempts have been made to combat these difficulties. The rapidly rising field of artificial intelligence is one of the most potential fields to break through the divide between computers and humans. Right now, the current methods of artificial intelligence are some of the closest attempts we have to modeling inference as humans make them. Computers can use patterns, where given a database of jokes, they can use pattern detection to construct a joke from scratch by starting with a word and looking up the most common word to follow that one, and so on.
The jokes that computers have made give a resemblance of a joke, but still most of them make no sense. For example, one is “Why did the cowboy buy the frog? Because he didn’t have any brains.” Kind of like a joke, but no real punchline.
Overall, I’m excited to see new breakthroughs in making computers more and more like humans, from making jokes to having conversations. It’s especially exciting that something as pivotal as this could be happening in my lifetime.
Michael Chen 