AI vs Primate Brain Visual Processing Mismatch

Researchers at York University have identified a significant discrepancy between artificial neural networks (ANNs) and primate brains regarding visual information processing. While AI models are frequently cited as brain-like due to their ability to predict neural activity during object recognition, this study reveals that the internal mechanisms of these models do not mirror biological processes. The findings suggest that AI often achieves correct visual outputs through strategies fundamentally different from those employed by the human brain. The research team, led by Assistant Professor Kohitij Kar, utilized a novel reverse predictivity test to evaluate the alignment between AI internal features and recorded brain activity. By testing 1,320 natural and synthetic images, they demonstrated that while AI can predict brain activity, the reverse is not true: brain activity cannot reliably predict the internal features of AI models. This indicates a lack of structural and functional equivalence between the two systems. This study provides a critical diagnostic metric for the field of neuroscience and AI development. The authors emphasize that relying on poorly aligned models to interpret human behavior, particularly in clinical research for conditions like autism or PTSD, carries significant risks. They have released a testing toolkit to help developers improve the biological accuracy of their models. For practitioners and researchers, this work serves as a cautionary note against assuming functional similarity between AI and biological systems. Future research should prioritize building models that not only predict outcomes but also replicate the underlying neural mechanisms of human vision to ensure reliability in clinical and behavioral applications.