Plato's Cave: Evidence from Physiology

Neurophysiology and Perceptual Deficits

Lesion studies performed on experimental animals as well as studies of deficits in humans due to strokes and head injury indicate that specific aspects of perception are associated with specific areas of the brain. For example certain areas have been identified with the perception of motion, form, and color. A significant aspect of such perceptual deficits is that the missing features due to the lesion disappear from the world, rather than from "inside the patient's head". In subjective terms, if you were to suffer a stroke in your "color center" (if it can be called such), the color would disappear from the world around you, rather than from inside your head. A stroke in the "motion area" would remove the percept of motion from the world, and a stroke in the "form area" would deprive objects in the world of their recognizable forms. If you imagine therefore the perceptual effects of successive lesions it would seem that as each center is ablated, specific aspects of the outside world would disappear with each lesion until there is nothing left. The outside world therefore would seem to be composed of multiple perceptual features generated by specific brain areas, which appear to lock together into a single coherent percept of a solid world.

Even more direct evidence for spatial perception can be found in certain bizarre perceptual deficits such as left hemianopia. This is a condition caused by lesions to the right parietal lobe, and results in a syndrome whereby the patient becomes unaware of the left side of space. This is different from being blind to the left side, as for example in visual scotomas, or blind spots on the retina, because in the latter case the patient will be aware of their blindness and will advance with caution, expecting unseen objects to emerge from the blind sector. In the case of hemianopia on the other hand, it is not that the patient is blind to objects to the left, but is blind to the very space as a potential holder of objects. The existence of this kind of deficit in human perception is strong evidence for a spatial mapping ability.

Another perceptual deficit that is instructive in this discussion is the phenomenon of visual agnosia. Farah [Farah 1990] makes a distinction between "apperceptive agnosia" and "associative agnosia". Apperceptive agnoias are those in which recognition fails because of an impairment in visual perception, which is nonetheless above the level of an elementary sensory deficit such as a visual field defect. Patients do not see objects normally, and hence cannot recognize them. In contrast, associative agnosias are those in which perception seems adequate to allow recognition, and yet recognition cannot take place. It is said to involve a "normal percept stripped of its meaning". This condition was popularized by Sacks [Sacks 1970] in "The Man Who Mistook His Wife for a Hat". What is interesting in this example is that the patient, while completely unable to identify objects in his world, could nevertheless perceive them spatially, and could even describe them in geometrical terms. He would surmise that moving pillar-like objects were probably people, but when they stood still he could mistake them for objects such as a coat rack. What is remarkable in this condition is that the patient could go about his daily life, walking the streets and avoiding obstacles. The existence of this condition clearly separates the function of spatial perception from that of recognition, showing that it is possible to have spatial perception in the absence of recognition. Those models of perception that simply skip this stage and proceed directly to recognition or abstraction from the raw sensory input are based on the assumption that this stage of processing is somehow superfluous or unnecessary, and yet the condition of associative agnosia indicates that this spatial integration stage, in the absence of the subsequent recognition stage, provides the great bulk of functionality of exactly the sort that is most difficult to duplicate in artificial robotic systems, i.e. the ability to move around in the world and avoid obstacles without regard to their identity. This condition of spatial perception in the absence of recognition, I propose, is the kind of perception enjoyed by animals, whose recognition abilities are considerably inferior to our own, and yet whose spatial perception appears to be similar to ours.

The other kind of agnosia, apperceptive agnosia, represents a failure of this spatial integration stage, creating in the patient a subjective experience of visual disorder and confusion for which the higher level recognition system is powerless to compensate. The fact that the higher level recognition system is intact is manifested by the fact that the patient can describe the disconnected pieces of their confused percepts with considerable precision. Such a patient can list the individual features that they observe in an object, but cannot integrate those disconnected features into a single spatially coherent whole. Apperceptive agnosics will attempt to identify an object by a process of adding up visual impressions, performing the integration of features cognitively rather than spatially. For example such patients often use color to identify objects, mistaking vanilla ice cream for scrambled eggs, and a piece of white soap for a piece of paper. A patient might recognize dice from their size and the presence of black dots on a white surface. This kind of reasoning is reminiscent of the kinds of algorithms proposed for recognition in Artificial Intelligence systems, which use high level reasoning on the low level input. For example such a system might reason that


  (IF (SMALL) AND (WHITE) AND (HAS-BLACK-DOTS) THEN IS-DICE).

These algorithms exhibit the same propensity for being sidetracked or confused by irrelevant shadows or stray edges as do the agnosic patients themselves. Furthermore, even if such a system correctly identifies an object, that information is encoded purely symbolically, without any spatial context. Consider the difference between the statement "there is a pair of dice in front of you", and a view of those dice. In the former case it is impossible for example to reach out and pick them up with anything more controlled than a blind grope, while in the latter it is possible to plan a trajectory for your hand, and to orient your fingers appropriately, even if you had never seen dice before and were unaware of their identity or purpose. I suggest that computational models of vision will continue to suffer from apperceptive agnosia as long as they fail to incorporate this vitally important stage of spatial integration.

Return to argument

Return to Steve Lehar