Microsoft is developing a new AI system that is learning what's considered to be funny, at least by the standards of The New Yorker magazine. The system is designed to weed out below-average submissions to the magazine's weekly cartoon caption contest so that human editors can concentrate on the ones that might actually be picked to win.
According to Bloomberg, the project started when a Microsoft team member, Dafna Shahaf, attended an address by Bob Mankoff, the cartoon editor of The New Yorker. She decided to see if an AI system could figure out how funny a caption for a cartoon could be like:
"For the study, Shahaf fed cartoons and captions from the New Yorker's database into the system and trained it to find the funniest choices among captions that make similar jokes. She relied partly on crowdsourced input from contract workers, using Amazon.com's Mechanical Turk. Then she moved to the harder task of ranking jokes. Because typical computer vision software is designed for photos, not drawings, the researchers had to manually describe what was pictured in each cartoon. They organized this into two categories: the context and the anomalies."
The final results of the study were that all of the captions that were considered the favorites of the human editors of The New Yorker showed up in the top 55.8% of the ones picked by Microsoft's AI program. That means that Microsoft's system could eliminate nearly half of the over 5,000 captions that are submitted to the magazine's cartoon contest each week. Microsoft says they want to keep working on the system so that it could create its own jokes which could be used by, for example, its Cortana digital assistant.