A sample of a New Yorker cartoon, analyzed by Microsoft
Computers have undeniably evolved at a very fast pace today, revolutionizing the way we live, with smartphones, tablets, laptops, and the like. However, no matter how smart these things have become, there's one aspect that these gizmos have always struggled with: understanding humor. Despite this, researchers at Microsoft have recently partnered with The New Yorker, an American magazine, to teach an artificial intelligence (AI) system to assess what is funny and what isn't.
To achieve this, one of the researchers, Dafna Shahaf, together with her colleagues fed an AI with an archive of cartoons and caption contest entries from the said magazine. They went on to train the system in finding and ranking the funniest choices among captions that make similar gags, using crowdscourced input from Amazon's Mechanical Turk. Typical computer vision is only programmed to see photos, and not hand-drawn images, so the researchers manually described to the system what each cartoon is displaying. The descriptions were then organized into two categories: context and anomalies.
Eventually, Microsoft's researchers were able to devise an AI system that helped the magazine's cartoonists' jobs easier. All of the editors' favorite cartoons reportedly appeared in the AI's top 55.8% of choices. With this in consideration, the system could be used to eliminate at least 2,200 bad puns, without missing the good entries. "I do think the future is human-machine companionship, computers can be a great aid," said Bob Mankoff, an editor for The New Yorker.
Such a project is quite comparable to Skype Translator, Microsoft's speech translation application. The program helps teach machines translate one language to another in real time, in order for two persons not speaking the same language to understand each other.
7 Comments - Add comment