Next: Genetic algorithms Up: Adaptive information retrieval Previous: IR issues

Electronic artifacts

It is clear to many scientists studying information systems that the potential impact on society of computers and the communication networks connecting them may rival that of Gutenberg's printing press. As we attempt to prognosticate about the future of these technologies, I have found it useful to consider two very old social institutions, science and the law, and the ``electronic artifacts'' they build and leave in machine readable forms. The goal is to then extrapolate from these classical activities to new forms of electronic publishing, such as Email and hypertext.

AIR was originally developed to learn the semantics of the technical vocabulary used by a single group of scientists to describe documents they share. More abstractly, the system comes to learn a ``cultural artifact,'' viz., the semantics of the indices used by that particular user group. There is a natural role for such a technology as an extension of the modern library, as it extends beyond its traditional concern with the physical warehousing of books [27]. (This research was extended by a grant to me and co-PI E. Hutchins, by the National Science Foundation [28].)

The potential relevance of AIR's approach to the legal domain was described at the first meeting of the AI &Law conference [17]. This potential was realized by Dan Rose's thesis, also described by Dan Rose and myself at the next meeting of this same group [45] and more completely described in an article solicited for a special issue of Intl. Journal of Man Machine Studies on ``AI and the Law'' [47].

On several occasions I have looked beyond AIR to consider information systems more generally. With colleagues at the University of Michigan I analyzed special features of the communications mediated by conferencing systems, for example imposed by timing constraints and the serial stream of messages [32]. Such analyses have helped to suggest new tools for managing electronic mail that make this potentially overpowering volume of information manageable and even a new form of literature [14]. In fact, I've argued that it is appropriate to analyze hypertext itself as a novel form of knowledge representation [19]. Such a view is consistent with other AI researchers investigating a role for ``text-based intelligent systems'' [23].

Next: Genetic algorithms Up: Adaptive information retrieval Previous: IR issues