Scientists read monkey minds to draw eerily accurate portraits

As superpowers go, being able to read a monkey’s thoughts is right down there with invisibility when nobody is around. Still, with the brain almost as much of a mystery black box as it’s always been, a study from the California Institute of Technology that does just that potentially represents an amazing breakthrough.

In short the researchers were able to show photographs of a human face to a monkey, and then algorithmically draw astonishingly accurate portraits by interpreting the brainwaves. To be clear, these aren’t like shonky police efits, or the kind of tragic mugshot you might get in a knock-off copy of Guess Who – these are virtually indistinguishable from the originals, as the picture below shows. In both columns, the photo they were shown is on the left, and the algorithmic output is on the right.

This is more than a neat party trick, though. Aside from potentially revolutionising the flimsy world of eyewitness accounts, it puts doubt into our previous understandings of facial recognition, which date back to the 1960s. The best guess previously was the idea that specific neurons would be attached to specific people – the grandmother cell theory. As Professor Rodrigo Quian Quiroga, a neuroscientist from the University of Leicester, told The Guardian: “This paper completely kills that.”

The way the discovery was made was almost entirely accidental. Doris Tsao and her colleague Stephen Le Chang were originally working on a computer vision project, and had come up with a series of 25 measurements that would assemble a dot-to-dot face shape. This was quite abstract, so they came up with 25 more measures to explain missing characteristics such as eye colour, skin tone and musculature.

To the pair’s surprise, the scheme they had come up with independently seemed to mirror neurons activity in two male rhesus macaques when shown photographs of human. “The predictions were so good, I was kind of amazed,” Tsao said, tracking back to ensure no mistakes had been made in the process. Later, however, Tsao concluded that the system they had picked was simply mathematically the most efficient way to convert faces into numbers. “If you look at methods for modelling faces in computer vision, almost all of them … separate out the shape and appearance.”

We’re a long way away from being able to read minds, but we’re one step closer to unraveling the mysteries of the mind. And another step along the road to proving that “monkey see, monkey do” is an unfairly simplistic assessment of our simian friends’ faculties.