UW–Madison artificial intelligence tool helps choose New Yorker cartoon captions

Tool also identifies genetic markers for disease
UW–Madison artificial intelligence tool helps choose New Yorker cartoon captions
Photo by newyorker.com

Finding the funniest captions for hundreds of cartoons in The New Yorker magazine is not the ultimate application of an artificial intelligence tool developed at the University of Wisconsin-Madison.

Electrical and computer engineering professor Rob Nowak and his graduate students didn’t set out to mathematically facilitate a humor contest — or assist the U.S. Air Force or identify genetic markers for disease — but their creation of NEXT, an open-source cloud-computing system, has done all that.

“It’s basically a machine learning software system, but it’s designed specifically to handle cases where the machine is learning a model to predict things,” Nowak says. And one of the first, best and ongoing test cases for NEXT has been predicting how funny people find suggested captions for New Yorker cartoons.

It works like this: The weekly magazine posts captionless cartoons on its website and asks readers to submit their own humorous verbiage. Every week the magazine receives more than 5,000 submitted captions and then 10,000-plus people rate many of them “funny,” “somewhat funny” or “not funny.”

The algorithm-aided system — in which the captions increasingly rated as “funny” are shown more frequently to subsequent voters — was implemented after Nowak approached then-cartoon editor Bob Mankoff in late 2014. Before the algorithm was applied to the contest, New Yorker interns had to evaluate the thousands of submitted captions individually.

The algorithm, which The New Yorker uses free of charge, “is a fun application and it’s helpful to them,” Nowak says. “More importantly it’s helped us refine and improve the algorithm.”

Nowak says similar algorithms have aided the study of virus replication in fruit flies at the Morgridge Institute for Research, housed in WID. The NEXT system has commercial applications, too, as Lands’ End, American Family Insurance and Amazon have found. “Amazon has hired at least three of my students,” Nowak says.

Nowak was hired as director of the Machines, Algorithms and Data Lab, or MADLab, established in May on the UW-Madison campus with $5 million from the U.S. Air Force.

Nowak says the military branch has a great deal of data in need of analysis but much of it is classified for reasons of national security or privacy. Nowak says MADLab won’t be doing classified research. “The basic research we’re going to do at that center could benefit the Air Force but it could [address] general problems for humanity,” he says.

Joel Patenaude is associate editor of Madison Magazine.