Second generation expert systems, explanations, arguments and archaeology

It is claimed that second generation expert systems provide a means of solving some of the problems put forward recently by writers on AI and archaeology (Huggett 1985, Huggett and Baker 1985, Baker 1986, Reilly 1985). Various kinds of second generation expert system are presented in particular a variant I call the arguing expert system (AES). A program the Argument Support Program for Archaeology (ASPA) is proposed to illustrate these ideas. It is shown to provide possible solutions to (a) the problem of the inadequacy of the abstraction captured in the knowledge base of an expert system and the attendant danger of its fossilization and (b) the problem raised by the inability of expert systems to model nondeductive reasoning. The former is dealt with by giving the user the chance to get the system to 'change its mind' by engaging the system in a sustained argument exchange. This has the effect of allowing the user to supplement or modify the knowledge base in a principled way thus reflecting changes in factual knowledge about the domain. The user can also argue about the relations between facts contained in, for example, rules. This allows the user, at least in part, to modify the theoretical viewpoint reflected in the knowledge base by modifying how the knowledge is structured and manipulated. The user is also given the facility for modelling multiple viewpoints. Each viewpoint can reflect either a particular theoretical standpoint or the views of an individual archaeologist. Problem (b) above is dealt with by incorporating within the system the capacity to recognize and assess arguments from analogy. The argument support paradigm although designed to integrate many aspects of expert system research is thus shown to be particularly suitable for applications in archaeology. This is because of the patterns of reasoning used in archaeology and the nature of its theoretical support. It is claimed, in brief, that interaction with ASPA will be more natural and more flexible than with traditional expert systems.