Computer Metaphors

   Misgivings about Computer Metaphors of the Human Brain

   Within the AI community there is a growing dissatisfaction concerning the adequacy of sequential models to simulate the cognitive processes. . . .

   For an example of the dissimilarity between computers and nervous systems, consider that in conventional computers . . . each piece of data [is] located in its own special space in the memory bank [and] can be retrieved only by a central processor that knows the address in the memory bank for each datum. Human memory appears to be organized along entirely different lines. For one thing, from a partial or a degraded stimulus human memory can "reconstruct" the rest, and there are associative relationships among stored pieces of information based on considerations of context rather than on considerations of location. . . . t now appears doubtful that individual neurons are so specific that they are tuned to respond to a single item and nothing else. Thus, connectionist models tend to devise and use [i]distributed principles, which means that elements may be selective to a range of stimuli and there are no "grandmother cells." . . .

   Information storage, it appears, is in some ill-defined sense a function of connectivity among sets of neurons. This implies that there is something fundamentally wrong in understanding the brain's memory on the model of individual symbols stored at unique addresses in a data bank. . . .

   A further source of misgivings about the computer metaphor concerns real-time constraints. Although the signal velocities in nervous systems are quite slow in comparison to those in computers, brains are nonetheless far, far faster than electronic devices in the execution of their complex tasks. For example, human brains are incomparably faster than any computer in word-nonword recognition tasks. (P. S. Churchland, 1986, pp. 458-459)