Evolution Versus True Perception
February 4, 2013, Electronic Imaging Conference, Burlingame, CA—Donald Hoffman from UC, Irvine asked the question “does evolution favor true perceptions?” in his keynote talk. Evolution has always moved to work around poor designs with fixes that work with the least energy.
We now know that about half of the brain is involved with visual processing, about 50B neurons and 100T synapses are involved in this effort. Vision is compared with a camera, where the eye is like the lens and sensors. The difficulty is that the eye has a lot of internal processing and is built wrong. The inter-neurons are in front of the sensors, rather than behind then as in a high performance camera.
In essence, the visual system is a reality engine that takes color, shapes, objects, motion, textures, and depth to give the viewer a sense of events. All of the processing takes less than 100mS. At issue, is that the eye is easily fooled. Optical illusions and perceptions of motion and structure emerge from a set of dots that change colors, but not location, in a sequence.
Natural selection tends towards veridical perception. Perception is a reconstruction of the true properties of some function, and evolution reinforces this abstraction. Perception helps us survive in out niche and guides adaptive behavior. Evolutions considers the cost in time, information, and computation in making any adaptations. Evolution can be considered as applied game theory. As a result, perception strategies changed to accept a relation between real and structure.
In some regards, perception is a UI to the world, and this interface has the purpose to guide adaptive behavior while hiding the truth. The interface shows all of the relevant properties of an icon, and any parameters that are not key to survival are irrelevant to the real object. A hierarchy of response modes from naïve, to critical, to interface shows that utility is not a linear function of reality.
Truth has not shown to be necessary to survival. Instead, the fitness and costs in time, energy per bit, and frequency of choice matter. Simple abstractions are better than the truth, since an organism is penalized for the increasing number of bits and time to process higher resolution and truth. At the same time, there is a distinction between taking something seriously and the logical or literal understanding of an event. For example, if a car is an abstraction of the truth, then stepping in front of a moving car is logically irrelevant, compared with taking the images seriously and avoiding the car.
Space and time provide a standard view of the objective world. The mind and vision construct optical illusions that we take as the closest approximation of reality. All simulations so far, have shown that truth doesn’t realize in long-term evolution. The scientific methods and good models can prove the distinction between our perceptions and the truth, but the distinctions only matter on an evolutionary scale.
