Researchers have found that certain artificial intelligence systems can be fooled by optical illusions in ways that closely resemble human perception. In one experiment, a neural network trained to predict future video frames was shown static images that humans perceive as moving. Despite the images being completely still, the AI interpreted them as containing motion, demonstrating that machine vision can be susceptible to the same visual tricks as the human brain.
The illusion used in the study relies on specific patterns and contrasts that cause the brain to anticipate movement where none exists. When humans view the image, different circular elements appear to rotate unless they focus directly on one area. The AI model reacted similarly, suggesting that its internal prediction mechanisms responded to visual cues in a way comparable to human predictive perception.
However, important differences were also observed. Humans can suppress the illusion by focusing their attention on a specific part of the image, effectively “stopping” the perceived motion. The AI system, lacking a true attention mechanism, continued to interpret motion across all elements of the image without distinction. This highlights a key gap between human cognition and current AI vision systems.
The findings offer insight into how both artificial and biological systems process visual information. While AI can mirror certain aspects of human perception, it still lacks the adaptive awareness and selective attention that define human vision. These experiments help researchers better understand not only the limits of AI perception but also the underlying processes of the human brain itself.
