Representational Eclecticism a Foundation Stone for the New Ai?

In this paper I support the view that neural networks and connec-tionism constitute a `new AI' but argue that the new eld incorporates both the reactive systems movement and work on genetic algorithms. The common theme in these approaches is, I propose, the idea that cognitive agents may use representations in an eclectic and opportunistic fashion rather than in the neatly structured manner proposed by classical AI. The way in which this idea is being developed (in connectionism) is considered and the outstanding problems are illustrated using a simulation. 1 New versus Old: What is the diierence? Does the eld of neural networks (connectionism) constitute a `new AI' or just a new set of methods for dealing with diicult problems? There is no clear answer. There are obviously methodological diierences between connection-ism and classical AI. Connectionism concerns itself with network architectures and training algorithms whereas classical AI concerns itself with algorithmic methods and knowledge representation. But are these diierences deep-set or merely superrcial? Some researchers (eg. 1]) have suggested that there is a strong continuity from classical AI to connectionism noting that even though the latter paradigm is concerned with a new type of computational architecture it remains concerned with the problems that were central to classical AI (e.g. representation).