Artificial intelligence is already doing some incredible things, for example in the field of healthcare. But how much can we rely on AI to improve things on the lighter side of life? Can a computer make us laugh?
Anyone who has chatted with an AI bot will know that they come out with some pretty weird things, interspersed with reasonable chunks of conversation that mostly much make sense. When it comes to asking AI to make jokes, the results are pretty similar – mixed.
Why jokes don’t come easy to AI
Vinith Misra is a data scientist at Netflix (and consultant to HBO’s Silicon Valley), and is interested in the humorous potential of AI.
Getting AI to make jokes is not easy, however. AI needs to follow rules to get to a solution, but there is no set formula for joke-making. If there were, then humans could all be taught to be hilarious just like we can be taught to use a simple computer program, for example.
“Humor is one of the most non-computational things,” Misra says.
A joke told by British comedian Lee Dawson demonstrates the complexities of deconstructing jokes.
The joke is: “My mother-in-law fell down a wishing well the other day. I was surprised — I had no idea that they worked!”
Explaining why this joke is funny (assuming you do think it is and/or are not an easily offended mother-in-law) is not simple.
The listener has to use societal context that’s understood outside of the joke, and it’s a big challenge to imbue an AI program with a lifetime’s experience of broad social understanding.
A different approach
Abhinav Moudgil, a graduate student at the International Institute for Information Technology in Hyderabad, India, takes time out from his day job studying computer vision to investigate computational humor.
His approach is to work with a recurrent neural network, a popular type of statistical model.
The difference between this and rule-based models is like the difference between showing and telling. Rule-based algorithms depend on work put in by coders from the get-go. The result is a structured system and structured jokes.
Here are two examples:
“What is the difference between a mute glove and a silent cat? One is a cute mitten and the other is a mute kitten.”
“What do you call a strange market? A bizarre bazaar.”
Basically nice word play that isn’t hilarious but does work perfectly well.
These jokes were produced by a system that analyzes hundred of thousands of jokes, harvested from around the Internet, character by character. It noticed the probabilities of certain letters appearing after other letters, and its jokes then ended up following this pattern.
As a result, many started with “What do you call…” or “Why did the…”, because the letter “w” had a high probability of being followed by “h”, the letter pair “wh” had high probabilities of being followed by “y” or “a,” and the letter sequence “wha” was very likely to be followed by “t.”
There were some jokes produced by the same system that were just plain odd (although some may find them hilarious if they like absurdist humor):
“I think hard work is the reason they hate me.”
“Why can’t Dracula be true? Because there are too many cheetahs.”
“Why did the cowboy buy the frog? Because he didn’t have any brains.”
“Why did the chicken cross the road? To see the punchline.”
A third approach was taken by Stanford University’s He Ren and Quan Yang, who used a particular comedian to inspire their system. Their neural network ended up trying to emulate talk show host Conan O’Brien.
Here are some of the results:
“Apple is teaming up with Playboy in the self-driving office.”
“New research finds that Osama Bin Laden was arrested for President on a Southwest Airlines flight.”
True, it sounds like Conan if he’s lost his mind.
Ren and Yang asked for human reactions to the jokes, and found that around 12 percent of the jokes were considered to be funny, which is not a great hit rate. We assume Conan would score much better – but AI may catch up with him in time.