Images of Real Human Faces Recreated From Monkeys’ Brain Signals

By monitoring macaque monkeys’ brain cells as they look at human faces, scientists may have just unlocked a key part of how facial recognition works.

Doris Tsao of the California Institute of Technology and her team found that individual cells can generate infinite faces by joining forces—their combined activity creates a composite image.

Composite image generated by the algorithm, versus the person’s actual face

Here’s Andy Coghlan, reporting for New Scientist:

In their study, Tsao and Le Chang tracked the activity of cells in monkey brains. Working together, the combined signals from these cells could encode 50 different aspects of a face—for example, face shape, eye distance, skin texture, and so on.

“The key is that even though there’s an infinite number of faces, you can describe all of them with just these 50 dimensions,” says Tsao. “It’s like computer-generated imagery, except it’s in our brains.”

Tsao and Le Chang inserted electrodes into 205 of each macaques’ specialized facial recognition cells and proceeded to show the macaques 2000 images of human faces. They found that each facial cell interprets a particular face in a slightly different way—and that an algorithm designed out of this feedback constructs a face that bears remarkable resemblance the real-life version.

This research could be the key to advancing the field of facial recognition technology, as there are still many limits to what engineers can do. Tsao and other colleagues surmise that this system could work in parallel with another system that scientists already knew about—one that encodes more familiar faces in the brain. That makes this new finding a big deal, since it fills some crucial gaps in our knowledge.