The proactive brain: Using memory-based predictions in visual recognition

Introduction You are speeding along an empty stretch of highway in the southwestern Nevada desert on a clear, starlit night. It is very late, and you have been driving for many hours, fighting boredom and encroaching sleepiness. Suddenly you notice a strange, dark shape near the road, a few hundred yards away. What is this thing? Your mind is racing through the possibilities: Is it a police cruiser parked at an odd angle? An unusual rock formation? Some top-secret weapons system that the U.S. military is testing in the desert? As whatever-that-thing-is fades in the rearview mirror, you still have no idea what it was. Perhaps it was just a harmless piece of highway maintenance equipment that became unrecognizable at night? The hypothetical example just described demonstrates a failure of the visual system to map input to any stored memory of an object. It is remarkable because of its rarity – most of our conscious visual recognition experience consists of recognizing things in our environment seemingly without effort. Recognition, by definition, relies on matching input to representations already stored in the brain. However, given that any real visual stimulus can be seen literally in an infinite number of views, determined by its orientation, lighting, distance, visual “noise,” and occlusion, it is highly improbable that the brain performs recognition by matching detailed input to a representation of every possible viewof an object.