Increasing the Biological Inspiration of Neural Networks

Abstract

We describe three extensions of current neural network models in the direction of increasing their biological inspiration. Unlike “classical” connectionism, Artificial Life does not study single disembodied neural networks living in a void but it studies evolving populations of neural networks with a physical body and a genotype and living in a physical environment. Another extension of current models is in the direction of richer, recurrent network structures which allow the networks to self-generate their own input, including linguistic input, in order to reproduce typically human “mental life” phenomena. A third extension is the attempt to reproduce the noncognitive aspects of behavior (emotion, motivation, global psychological states, behavioral style, psychogical disorders, etc.) by incorporating other aspects of the nervous system in neural networks (e.g., sub-cortical structures, neuro-modulators, etc.) and by reproducing the interactions of the nervous system with the rest of the body and not only with the external environment.