Emergence in Neural Network Models of Cognitive Processing

Abstract

This contribution is devoted to assess whether a basic hypotheses underlying the connectionist approach is firmly grounded and useful for the research activities concerning the Psychology. The hypothesis asserts that the observed macroscopic consequences of cognitive processing are nothing but collective effects emergent from the interactions between suitable microscopic units. The implementation of the above assertion is based on mathematical models making use of artificial neural networks. In this contribution we investigate whether: (a) these models concretely exhibit emergent collective effects; (b) these collective effects are characterized by the same features which we observe in behaviors produced by human mental processes. Our conclusion is that only particular models of this kind (not including Perceptrons) can give rise to emergent collective effects. Moreover, only the use of specific strategies and techniques of data analysis allows to use the models themselves in a way useful to experimental psychologists. Our contribution discusses the application of our proposals to a specific case study in order to illustrate the nature of the difficulties encountered when dealing with a concrete implementation.