The best way to approach AI challenges is often a topic of vigorous discussion. The two main camps are connectionist1.8 and symbolic AI. This is perhaps best illustrated by an example of speech recognition. For symbolic AI, form and meaning of words are represented in terms of labelled symbolic nodes [RMK06, p. 4]. Connectionists see the mapping of form to content as something that should develop in a neural network through training by examples. Connectionist pursue a bottom-up approach, while symbolic AI tends to analyze the higher level relationships and model those.
Erik de Bruijn