Move over Watson. Microsoft has trained an artificial intelligence program that can pick through the thousands of submissions to the New Yorker’s cartoon caption contest to help find a winner. Bloomberg writes that Microsoft researchers worked with Bob Mankoff, the cartoon editor at the New Yorker to train a computer to understand what makes a New Yorker cartoon funny. Mankoff, who goes through an assistant every two-years because they suffer burn out reading so many caption entries, is excited about the AI possibilities.
Dafna Shahaf, a researcher at Microsoft, used the database of cartoons to train the program to understand commonalities and differences in the millions of cartoons, which lets the AI run through the entries the New Yorker receives each week for its back-of-magazine cartoon caption contest. About 55.8% of the time the humans agree with the captions the AI selects, which is a pretty good percentage.
From a technical perspective, computers have not suddenly learned a sense of humor (not even the dry, New Yorker sense of humor). They have learned how to emulate a specific style of humor, although that in itself is pretty cool, because learning how to emulate something, as opposed to being told exactly how to do something is an important step. If you ran the same AI against Dilbert or Ziggy caption suggestions you’d get poorer results because the Microsoft AI is trained against the New Yorker’s specific archives. However, getting computers to “understand” humor is an important step in almost all aspects of interacting with humans and translation.
Thus, efforts to make the lives of Mankoff’s assistants easier has the added benefit of possibly making Microsoft’s Cortana or Skype’s translation features better as well. And maybe one day, instead of just helping weed out caption suggestions, researchers can build an AI that will write one on its own.