Expert Systems and AI-Based Decision Support in Auditing: Progress and Perspectives †
William E. McCarthy
Michigan State University
Brigham Young University
University of Massachusetts
When all the AI rhetoric is boiled away, expert systems are simply com puter programs much like general ledger packages or even like video games. Writing a new payroll program in COBOL is not research, and neither is building another auditing expert system.
Since the development of AUDITOR at Illinois, there have been a num-ber of auditing expert systems designed and built by both academics and ac-counting professionals. For surveys of this work, see Messier and Hansen , Gal and Steinbart , Bailey, Hackenbrack, De, and Dillard , and Bailey, Graham, and Hansen . However, as encapsulated by the statement above, a continuing criticism of this work (indeed, a criticism of any knowledge-based work in accounting) is that it constitutes more devel-opment than research. In this paper, we contend that such blanket criticisms are unfounded and are in fact more attributable to a critic's lack of schooling in computer science than to any conceptual shortcomings in the actual sys-tems research methods. More specifically, we will look at several auditing expert systems and evaluate them in terms of some informally developed dif-ferentiation heuristics, heuristics whose rationale depends heavily on the work of March  and Cohen and Howe . We will also try to chart new directions for research in knowledge-based auditing systems. Our central pur-pose throughout this paper is to try to develop a framework of analysis so that when someone proposes a new audit expert system or enhancements to an existing audit expert system, we can type its contribution as either pri-marily research or primarily development or both.
†Support in the development and preparation of this paper was provided by the Department of Accounting at Michigan State University and Arthur Andersen & Co. Steve Rockwell provided numerous comments and criticisms.