AI has independently learned to perceive visual information in a way similar to humans.

Researchers from Osaka University have found that Vision Transformers (ViT), a type of generative AI model trained using the DINO method, can spontaneously develop visual processing mechanisms similar to those of humans. By comparing human eye-tracking data with the models' behavior, they observed that ViTs without fixed filters focused on faces, figures, and backgrounds in ways that resemble human perception.

These models processed visual information similarly to how adults watch video clips, unlike models trained with fixed filters. The spontaneous emergence of such mechanisms shows that self-supervised learning enables AI to develop fundamental aspects of visual perception.