WHAT IS THE NET WORTH? SOME THOUGHTS ON NEURAL NETWORKS ANO BEHAVIOR

Main Article Content

M. JACKSON MARR

Abstract

Neural network models have played major practical roles in engineering as well as theoretical roles in cognitive science, and now are being explored in behavior analysis. What contributions can these kinds of models make to a science of behavior?

Neural networks and behavioral processes can show similarities in dynamical properties, dependency on some variation of a contiguity mechanism, instantiation of some sort of memory, and operations in accordance with some kind of delta rule leading to a quasistable state. However, because these are basically inherent properties of all network models, they are grossly indeterminate. Conversely, they may be best described as "implementory", as opposed to "explanatory" models in the sense that they, with few exceptions, only simulate what they were specifically designed to simulate. This "curvefitting" quality sets them apart from predictive and otherwise suggestive quantitative models. One rationale for their exploration is their putative value in simulating neural mechanisms in learning. Not only is it the case that we understand little about these mechanisms to begin with, but network models capture virtually none of the complexities of the nervous system at any level. Despite all these difficulties, these models are worth further development and exploration as potentially powerful quantitative approaches to behavior, independent of any possible relations to real neural systems.

Article Details

How to Cite
MARR, M. J. (2011). WHAT IS THE NET WORTH? SOME THOUGHTS ON NEURAL NETWORKS ANO BEHAVIOR. Mexican Journal of Behavior Analysis, 26(2), 273–287. https://doi.org/10.5514/rmac.v26.i2.25159