Selective Neural Networks and Their Implications for Recognition Automata

Higher mental functions require the prior abitity to categorize objects and events according to sensory signals reaching the brain. The neuronal group selection theory postulates that this ability arises from a kind of Darwinian selection operating in somatic time on groups of interconnected neurons. These groups develop with varied and overlapping abilities to respond to patterns of input at their synapses. Groups that contribute to responses having adap tive value for the organism undergo modi fications in the efficacies of their synaptic connections that enhance their future re sponses to similar stimuli. Computer models of automata based on these prin ciples can carry out simple tasks involving recognition, categorization, generalization, and visual tracking. A general program for implementing such models is presented.