Cartoon Humor Test for AGI

I wrote a post nearly four years ago titled Brief Q&A on Artificial General Intelligence (AGI), which has held up pretty well. Despite a lot of progress in specific domains since then there is no obvious sign that AGI is around the corner, although as the post points out, the jump to AGI could be very nonlinear. The Turing test is of course a well known proposal establishing that a machine has reached a level of general intelligence (and fans of Blade Runner may think of the fictional Voight-Kampff machine).

It strikes me though that a better test might be to see if a machine can explain why proposed cartoon captions in the weekly New Yorker cartoon caption contest are funny and then come up with its own proposal for a funny caption. Here is a particularly apropos example from some time ago:

A machine that can consistently perform well on explaining the humor in these captions, strikes me as a strong indication of artificial general intelligence and much harder to fake than sounding reasonably like a human in a conversation.