An interesting application of artificial intelligence (AI) techniques, notably ontologies, to a difficult information retrieval problem--the search of legal databases by nonexperts--is presented in this paper. Contrary to what the title might suggest, the paper is neither a general study of Internet-based multiagent systems, nor a discussion of methodological issues concerning multiagent systems. The author claims, however, that the particular implementation he describes is an example of a useful application of agent technology to a complex problem. The main weakness of the paper is that no compelling evidence is presented that might justify this claim.

Although the system architecture includes interface agents, LawBots, and proactive agents, it is hard to see why these could not simply be described as modules, rather than agents. The usage of the term agent in AI is somewhat moot. In its most useful form, it is connected to the practice of ascribing propositional attitudes, such as beliefs and intentions, to machines, in order to better understand the machine’s structure, its behavior, or how to repair or improve it [1]. The paper makes no such ascription. Instead, it uses the term agent to suggest that its components have other properties often associated with intelligent agents, namely: flexibility, responsiveness to changes, and autonomy. The usefulness of ascribing these much more vague properties to a system is debatable. There are objectively measurable attributes, such as precision and recall, that could have been used to assess, for example, the extent to which ontologies improve information retrieval. Unfortunately, the paper does not discuss these attributes.