Technology-Independent Design of Neurocomputers: The Universal Field Computer 1

We argue that AI is moving into a new phase characterized by biological rather than psychological metaphors. Full exploitation of this new paradigm will require a new class of computers characterized by massive parallelism: parallelism in which the number of computational units is so large it can be treated as a continuous quantity. We suggest that this leads to a new model of computation based on the transformation of continuous scalar and vector fields. We describe a class of computers, called field computers that conform to this model, and claim that they can be implemented in a variety of technologies (e.g., optical, artificial neural network, molecular). We also describe a universal field computer and show that it can be programmed for the parallel computation of a wide variety of field transformations. 1. The ‘‘New’’ AI Traditional Artificial Intelligence technology is based on psychological metaphors, that is, idealized models of human cognitive behavior. In particular, models of conscious, goal-directed problem solving have provided the basis for many of AI’s accomplishments to date. As valuable as these metaphors have been, we believe that they are not appropriate for many of the tasks for which we wish to use computers. In particular, symbolic information processing does not seem to be a good model of the way people (or animals) behave skillfully in subcognitive tasks, such as pattern recognition and sensorimotor coordination. Thus, the needs of these applications are driving Artificial Intelligence into a new phase characterized by biological metaphors. We call this phase, characterized by a combination of symbolic and nonsymbolic processing, the ‘‘new’’ AI (MacLennan, in press). The technology of the new AI already includes neural information processing, genetic algorithms, and simulated annealing. The new AI will allow us to make use of massively parallel computers (including neurocomputers), optical computers, molecular computation, and, we expect, a new generation of analog computers. Current AI technology has been quite successful in a number of tasks, for example, chess, diagnosis of blood diseases and theorem proving. Many other tasks remain beyond its capabilities, including face recognition, autonomous movement and continuous speech recognition. The interesting thing is that the tasks that AI has been most successful with are those that we commonly consider higher cognitive activities, specifically, those activities that can be performed by humans, but by few other animals. On the other hand, the tasks that currently stretch the capabilities of AI technology are those that are lower on the scale of cognitive accomplishment. Specifically, they are activities that almost any animal can perform with skill. A rodent may not be able to prove theorems, but it can effectively navigate its way through a complicated terrain, avoid predators, and find food — and accomplish this with a comparatively small brain constructed of comparatively slow devices. It has been truly said that computers will replace mathematicians long before they will replace carpenters. Unfortunately, many important applications of artificial intelligence require just the sort of activities that stretch the current technology. Therefore it is important to seek the reason for the limitations of the current technology, and to see if there is a way around them. Current AI technology is based on psychological metaphors; its algorithms mimic conscious, rational thought. Thus, this technology deals best with verbalizable knowledge (knowledge that), deductive reasoning and discrete categories. However, as we’ve seen, there are other kind of intelligent behavior.