AI Perceives Optical Illusions: Predictive Coding in Brains and Machines
#artificial_intelligence #optical_illusions #predictive_coding #neural_networks
AI Perceives Optical Illusions Like Us
Artificial intelligence can now fall for optical illusions, just like human brains, revealing striking parallels in visual processing. Scientists have shown AI models, such as PredNet, "hallucinating" motion in static images like the Rotating Snakes illusion, where high-contrast patterns trick the system into predicting rotation.[1][4] This isn't a flaw but evidence of predictive coding, a efficient mechanism both brains and neural networks use to anticipate the world from past data.[1]
Why This Mirrors Human Vision
Trained on natural videos, these AI systems interpret shading as motion cues, looping false predictions much like our neurons do with Necker cubes or Rubin's vase, flipping between interpretations.[2] Unlike humans, AI can't "freeze" illusions by focusing attention, lacking that mechanism.[4] Quantum neural networks even replicate these effects via quantum tunnelling, outperforming some conventional models.[2]
Insights for Brains and Machines
This convergence validates AI's path toward human-like cognition, turning "hallucinations" into strengths for understanding reality.[1][3] It challenges views of illusions as mere bugs, suggesting they're mathematical inevitabilities, and could unlock new brain research amid deepfakes and perception studies.[2][3]
