Showing papers in "Journal of the Mount Sinai Hospital, New York in 1965"

Computer learning and the scientific method: a proposed solution to the information theoretical problem of meaning.

Abstract: This discussion outlines and implements the theory of an inductive inference technique that automatically discovers classes among large numbers of input patterns, generates operational definitions of class membership with explicit levels of confidence, creates a continuously updated \"self-organized\" coded hierarchical taxonomic classification of patterns, and recognizes to which already discovered class or classes, if any, a new input belongs in an information-theoretically efficient way. Relationships to the \"scientific method\" and learning are discussed. Learning processes which provide the individual and his species with an increased potential to cope with his material, biological and social environment have what evolutionists call \"adaptive value\" (e.g., see Dobzhansky (1)). Measured on such a scale, the most impor.tant aspect of such learning processes is the development of implicit or explicit techniques to accurately estimate the probabilities of future events. And, since the \"calculus\" of future events from analysis of past experience is the forte of science, it seems reasonable to turn to the methods of science for some useful clues when searching for efficient learning techniques. Modern scientists, and physical scientists in particular, who work regularly with the scientific method and operationalism (2), are keenly aware of some very efficient and sophisticated techniques for learning more about the universe. But if we ask of these who, it seems, should best understand the learning process, \"How shall we build—or indeed, is it possible to build—a machine that can learn and think?\", the replies are varied (3-8), and, as is indicated in a recent knowledgeable review by Selfridge (9), so far have been relatively unproductive. In this discussion, I hope to provide a basis for establishing considerably greater confidence in the broad potentials for machine learning via automated inductive processes. This will be attempted by sketching out, in operational terms and against a background of common experience, many theoretical and a few of the practical problems. * This work represents a revision of \"Pattern recognition, morphology and the generation of hypotheses\", presented at the Symposium on Machine Methods in Biology, A.A.A.S. convention, New York, 1960, in combination with a revision of part of the final report on P.H.S. Contract #SA-43-ph-3096. From the Division of Cell Biology and the Cell Research Laboratory of the Department of Pathology of the Mount Sinai Hospital, New York, N. Y.