New study identifies how AI fails to reproduce human vision

Deep neural networks in artificial intelligence resemble the human brain and can identify features like faces and cars with great speed and accuracy.

However, a recent study found they cannot fully reproduce human visual recognition or account for human neural responses to objects. The study suggests that deep learning models could benefit from a more human-like learning experience focused on behavioral pressure and training to accurately emulate human vision.

This discovery provides insight into why neural networks may fail to understand images with ecologically relevant object categories like faces and animals.