Human-like intelligence relies on machinery of high sophistication in terms of processing capabilities. In this chapter I'm going to discuss artificial counterparts to two mechanisms that have played a key role in the development of intelligence in biological systems: evolution and neural networks (section 3.1).
In an attempt to understand and a hope to eventually harness the strengths of this machinery, artificial neural networks and evolutionary algorithms have been created. Different versions of these mechanisms have been created to answer different questions or to harness different qualities of these algorithms.
Neural networks have been modeled at the level of the synapse to answer neurobiological and neurophysiological questions and to test hypotheses about the workings of the brain. Other implementations were used to classify images or patterns. The most common implementation of the artificial neural network (the canonical neural network), however, is not the most detailed nor biologically plausible version. The canonical neural network will first be presented, and the discussion that follows will address whether essential features are left out, or whether other implementations are feasible.
Implementation of another instance of evolution, next to natural evolution, brings forward many questions [Ray99]. What assumptions can be made, etc. Again, multiple perspectives on evolution result in different implementations of artificial evolution (see also section 4.2). Like with neural networks, a common evolutionary algorithm will be used as a reference example.