Symmetry is ubiquitous in Nature: the bodies of animals are symmetrical, plants are symmetrical and man-made objects are at least partially symmetrical, too. It follows that symmetry can, actually should, be used as an informative prior by any intelligent seeing system, such as our visual system. This clearly is the case: our 3D vision is based on an inference in which our retinal images are combined with symmetry priors. Technically, this inference is accomplished by maximizing a Bayesian posterior or by minimizing a cost function. The resulting percept is called veridical, when some permanent properties of an object, such as its shape, remain unchanged (stay invariant) when they undergo the transformation from the physical world to a mental representation in the visual system of the observer. The three concepts that characterize perceptual events, namely, symmetry, inference, and veridicality resemble the three concepts that characterize physical events, namely, the conservation laws that are derived from symmetries of a physical system by applying a least-action principle(Noether, 1918). We also know that not only the concepts are similar: The underlying mathematical formalisms are similar, too. The observed similarity between concepts and methods in Perception and Physics served as my starting point for redefining our contemporary science of Perception. The talk will conclude by suggesting that most, perhaps even all cognitive functions beyond Perception, such as Thinking, Concept Formation, Linguistic Communication, or Social Cognition can be included in this new Science of Mind because they are all based on inferences derived from symmetries operating in our physical and social environments. The questions that need to be asked next are (i) how all of this might actually be implemented in the human brain, and (ii) how can it be emulated in an Artificial Intelligence system.
*NOTE: Time change to 4:30 pm for this presentation for this date only. Thank you.