AI can now 'see' optical illusions. What does it tell us about our own brains? (www.bbc.com)

Recent research has revealed that artificial intelligence can be susceptible to optical illusions, much like humans. This finding offers new insights into how both AI and human brains process visual information. For instance, scientists used a deep neural network called PredNet, designed to model human visual processing through predictive coding, to examine its responses to visual illusions, such as the rotating snakes illusion. Impressively, PredNet was tricked into perceiving motion in still images, mirroring human perception. However, the AI lacked an attention mechanism, causing it to perceive all elements uniformly, unlike humans who can focus on one segment of an image. This research is significant for the AI/ML community as it helps bridge the understanding of visual perception between artificial and human systems. The study not only advances our knowledge of deep learning models—demonstrating both their similarities and their limitations compared to human cognition—but also opens doors for ethical experimentation in neuroscience. By studying how AI responds to illusions, scientists can explore the intricacies of human vision without the ethical constraints that accompany direct brain experimentation, providing a fresh perspective on our cognitive functions.