Continued speculations on the evolution of learning

Albus (1991) offers without proof the theorem that natural intelligence, like the brain, is a result of the process of natural selection. He offers four system elements of intelligence: sensory processing, behavior generation, world model database, and value judgement. The author has previously suggested (1991) that the latter two might have evolved from organisms with only the former two; and has raised (1994) the question how learning or memory might have evolved. All artificial neural nets (ANN) learning schemes assume the existence of an untrained pool of neurons whose weights are modified, but biologically, this untrained pool of neurons seems a contradiction, as these neurons serve no survival purpose prior to their training. It seems highly unlikely that a single mutation simultaneously produced this untrained pool and the training rule. How did both untrained pool and the training rule evolve? However, organisms possessing neurons serving control functions might, by mutation, beneficially combine the afferent signals of neurons and thereby improve their survival probability. The increased connectivity might allow the emergence of memory. This paper determines and models minimal control functions. Also, the effect of several control functions operating in parallel is studied.<
>