Parallelism

Can parallel processing of information be useful? This question should be considered purely technical in a greater degree than it may seem at first glance. Indeed, at the level of identifying the simplest visual features, textural elements, characteristic features in stereoscopic vision or motor parallax, the use of parallel processing of information should be considered appropriate. At the next, “objective” level, it is more difficult to imagine how parallelism can be used, although in this connection it’s worth noting A. Guzman’s (1969) works on identifying “cores” and uniting them into related domains or D. Waltz (1972) on using semantic roots for sorting shadow lines performed on a special parallel network.

However, at higher levels of mental activity, the expediency of applying parallelism encounters a number of fundamental objections. In the work on pattern recognition, quite a lot of schemes were proposed for performing parallel operations — perceptrons, integral transformations, etc. These schemes, interesting both from a mathematical and a computational point of view, could apparently serve as components of the theory of processing sensory information, but no more. Integral methods are good mainly for working with isolated two-dimensional images, but on their basis it is impossible to solve the problem of identifying and recognizing objects in complex three-dimensional scenes. Why?

When analyzing complex scenes, areas belonging to different objects should be correctly identified, because only in this case the perceived picture makes sense; However, to solve this problem, which is equivalent to the “object-background” problem traditional in Gestalt psychology, it takes so much effort that, as noted in the work of M.Minsky and S.Peypert (1969), the very possibility and even the expediency of developing an isolated technique recognition is questioned. For three-dimensional images, this problem is even more complicated by the distortion of perspective, as well as by the fact that certain parts of objects are invisible due to other objects.

In new sign theories, hypothesis methods are used with their subsequent confirmation; these methods seem to us more productive. It is difficult to solve any truly complex problem without paying close attention to its individual components. Fortunately, however, it is possible to imagine a more efficient (compared to just the idea of ​​parallelism) sequential process, in which large, complex sign structures are considered as the simplest operands. This opens up theoretically a new opportunity for a quick search for large substructures and, apparently, will allow us to find the secret of the speed of the mechanisms of human thinking and perception of visual information.