The brain knows how to draw reality

During the experiment, the volunteers were shown blurry black-and-white pictures (images of Mooney), which are difficult to make out anything. Then they were shown a clear version of the same object. After that, people instantly began to recognize the hidden image in the blurred photo.

This is “learning in one go”. The brain retains a clear pattern (prior) and uses it as a clue next time. The researchers found that this template is resistant to resizing the image, but is sensitive to its rotation or movement. This means that it is the visual pattern that we remember, and not an abstract concept (for example, “dog”).

To find a specific place in the head where this mechanism is activated, the team used MRI and EEG. They were looking for a zone whose activity would coincide with the behavior of people during the tests.

The tracks led to a high-level visual cortex. It is there that the visual “blanks” are stored, which the brain superimposes on the signal coming from the eyes. Electrodes implanted in patients for medical reasons have confirmed that a surge of activity in this area occurs exactly at the moment when a person experiences an epiphany and recognizes the picture.

Based on the data obtained, the scientists created a computer model. They taught the neural network to store the visual experience in a separate module and use it to recognize new images.

The result exceeded expectations: artificial intelligence began to learn as quickly and efficiently as humans, overtaking standard algorithms. This opens the way to creating machines that can learn on the fly, without requiring millions of examples to train.

Doctors hope that understanding this mechanism will help in the treatment of schizophrenia and Parkinson’s disease. In such patients, the “prediction” system is too active, forcing them to see things that are not really there. Now, doctors have a specific purpose in the brain to develop a new therapy.