#1 in a series of images exploring the parallels between human jokes and Generative Adversarial Networks (GANs). Each image was generated by a GAN using only the text prompt given in the image title.

Cognitive studies of humor suggest that jokes function as a kind of cognitive debugging. They exploit areas of overlap between two “scripts” or “frames” that can be applied to the setup line, tricking the joke recipient into applying the wrong one. (“There are two fish in a tank. One says to the other ‘How do you drive this thing?’”)

But this is similar to a GAN, in which a generator algorithm creates an image in imitation of its input dataset, and a discriminator classifies that image as either imitation or real. The GAN’s output consists of those imitations that trick the discriminator into assessing them as real.

In both, one intelligence tricks another with the goal of improving overall results.

Joking has often been thought of as uniquely human. But might it instead be thought of as the “meat version” of machine learning techniques that today stand at the frontier of artificial intelligence?