Long ago, back when twirling things was both an entertaining diversion and a serious mode of scientific inquiry, several researchers noticed an odd optical illusion. Spinning a white disk with a series of black lines produced colors where there weren’t any. A toymaker named Charles Benham turned the optical illusion into a toy, and Benham's Top was born. You can try it for yourself below.
Over the past 200 years, scientists have continued investigating this bizarre phenomenon (now termed “subjective color”), but it’s still somewhat mysterious. The best working theory for subjective color is that it comes from feedback in our visual cortex. Our eyes feed information forward into our brains, but our brains aren’t just sitting there waiting, they’re anticipating and predicting information to fill in the gaps. It’s this feedback that results in seeing colors in moving black-and-white images (fMRI studies seem to back up this model).
After two centuries fiddling around with the Benham’s Top illusion, scientists have uncovered several strange idiosyncrasies. For starters, arcs positioned at different distances can evoke different colors, and the colors can reverse when the direction reverses. But what’s even weirder is that artificial intelligence networks can see the Benham’s Top optical illusion too, according to a new study.
Read more: “The Illusion Machine That Teaches Us How We See”
Researchers from the National Institute for Basic Biology in Japan developed an artificial neural network (ANN) designed to predict “what image will come next” and spit out a visual representation. They trained their ANN on a series of first-person videos of natural landscapes (for example, video shot on a head-mounted GoPro from someone taking a relaxing hike). These videos, researchers say, give the network the same sort of visual training input a real person might have.
After the training, they showed their ANN Benham’s Top, and the images it produced also included faint colors. In fact, when they tweaked the spinning disk to move the arcs or reverse the direction, the ANN “saw” the same things a human would.
According to the team, the artificial neural network had learned to work in the same way our biological neural networks do. Because rapidly moving objects are especially challenging for the ANN to predict, they have an outsized impact during its training. The result is a “visual cortex” that works remarkably like our own, complete with the same feedback. Basically, it’s more confirmation for the feedback model of subjective color.
Benham’s Top gets a new spin, nearly 200 years later. 
Enjoying Nautilus? Subscribe to our free newsletter.
Lead image: MstMoushumi / Adobe Stock
