A group of researchers led by Bob Mankoff tried to teach robots to distinguish between things that are funny and things that are not.
This wasn’t a casual question. Mankoff is a cartoon editor for a publication that has a caption contest. Having to sift through mountains of submissions made him wonder whether the first screening could be automated.
Many of the recent improvements in industrial robotics center on sensors. Can a sensor be developed that will let robots “see” funny?
In fact, Mankoff and his Microsoft research colleagues were able to make some significant progress, and they didn’t use sensors at all. They were able to train robots to judge between pairs of captions and choose the one humans considered funnier 64% of the time — better than chance.
The characteristics robots were able to grasp included things like the use of unusual words. “Poughkeepsie” is funnier than “Boston,” right?
Of course, this is not telling jokes. However, other researchers have used similar methods to teach robots to make up jokes. They’re not very funny and they often don’t make sense, like this example:
What’s the difference between leaves and a car?
One you brush and rake, and the other you rush and brake.
Like most robot jokes, this sort of looks like a joke. True, we don’t brush leaves or rush cars, but it follows a joke-like pattern. It’s like jokes made up by six year olds.
Servo motors will be required if we ever try to teach robots to tell jokes instead of just recognizing or producing them. It’ll be all about the timing. So far, at least, robot jokes can’t stand on their own.
As Wired put it, “If robots ever conquer the world, in other words, we’re in for a dystopian future of horrible puns.”