PROBLEM OF TEACHING MACHINES TO IDENTIFY EXTERNAL SITUATIONS

A method for machine recognition of external stimulae, based on so-called potential functions, is proposed in the paper dealing with artificial intelligence. Individuals can recognize events and patterns, and teach others to do so, frequently without being able to explain how the process of recognition comes about. For instance, an illiterate person can be shown letters 'a' and 'b' and taught to recognize these letters irrespective of their shape. This process of information transfer is therefore based not on explanation, but on demonstration. This technique can be applied to learning, pattern-recognition machines, designed to respond to audio or visual commands. The problem of teaching the automaton to classify correctly a given input can be defined either in the deterministic or in the probabilistic domain. The report describes the application of potential functions to the probabilistic domain, and in conjunction postulates a third theorem. It is concluded that it is in principle possible to apply the demonstration technique to training of automata and that a rigorously scientific, rather than an empirical, approach to the solution of this problem is possible. (Author) Edited machine trans. of mono. Samoobuchayushchiesya Avtomaticheskie Sistemy (Self-Instructing Automatic Systems) Moscow, 1966 p3-8.