The Google Brain Team Is Trying to Teach an AI to Tell Jokes
Artificial intelligence is paving the way to technologies previously seen only aboard the USS Enterprise: teleporters and cyborgs have already been realized, and lab mice were recently observed enjoying a real-life holodeck. AI even seems to be artistically minded, writing songs and creating beautiful, photographic landscapes. So why then, can't AI tell a joke?
Douglas Eck, a researcher on the Google Brain team, has spent the last 15 years exploring the relationship between AI and music. In fact, he's currently leading Magenta, an open-source research project by Google that's at the forefront of using AI in creative endeavors. Eck recently spoke with the MIT Technology Review on why AI can't tell a joke, explaining that his team initially asked the question, "can we understand that component of joke-telling which is about surprise?" Here's what he came up with:
Eck's team was using what are called "truth vectors" to understand semantics in conversation and then use that information to manipulate it. Specifically, they used the joke "The magician was so angry she pulled her hare out." The use of puns was a good challenge for their budding system. The challenge is proving to be particularly difficult, however, because "Not only did this model not tell any jokes, funny or not, but we didn't actually get the code to converge."
There is an ongoing issue with AI being close, but not exactly close enough, to understanding and mimicking human behavior. If robots are doomed to be humorless, however, that'll be a difficult thing to accomplish.