ABSTRACT
The idea that machine intelligence might be substituted for the memory and searching skill of the reference librarian is not particularly new. For several decades, programs have been developed to teach reference through machine simulation of the reference interview; to offer low-level assistance to users when skilled staff are not available; and to capture and model the decision-making steps involved in negotiating a question and selecting appropriate sources to answer it. Early programs developed on mainframes were interesting experiments in simulation and instruction, but lacked the flexibility, speed, ease of development and alteration, and portability required of a reference program intended for use in an actual library setting. The advent of expert systems that can be mounted on microcomputers has promised to overcome some of these limitations. Fully programmed systems can offer sophisticated natural language interfaces and embed detailed knowledge about sources and users. Through the use of expert systems shells, prototype programs can be developed relatively quickly and modified easily. Systems mounted on microcomputers can be placed where library patrons are mostly likely to use them. Through some systems, users may be linked to several databases or to full text almost transparently. The continuing rationale for developing these programs appears to be twofold: to make the expertise of the reference librarian available in a more consistent and timely fashion; and to reduce the burden of repetitive, predictable questions for the professional.
