A Comprehensive Evaluation of Expert System Tools

Abstract

1.0 ABSTRACT Current trends in knowledge-based computing have produced a large number of expert system building tools. This onslaught of high-tech software stems from the discovery that expert systems can be effectively applied to a variety of industrial and military problem domains. A variety of vendors provide expert system prototyping and development tools which greatly accelerate the construction of intelligent software. Todayfs expert system tool generally provides the user with a friendly interface, an efficient inference engine, and formalisms that simplify the creation of a domain knowledge base. This paper presents a formalism for expert system tool evaluation and critiques an exhaustive variety of commercially available tools. The multitude of Expert System Tools (ESTs) available in today's marketplace combine to form an elite repertory of artificial intelligence software that can be used to construct prototype expert systems in a wide range of application areas. These prototypes are then progressively refined, eventually evolving into operative expert systems tailored to specific problem domains. Given any application area, the chances that one of today's ESTs will address the problem are quite high. It is worth noting, however, that the converse is not always true (i.e., given an EST, one should not expect to be able to tackle problems in all application areas) . State-of-the-art ESTs, by virtue of their design and artificial intelligence capabilities, possess certain inherent constraints which better suit them to particular problem areas. The more constrained an EST is, the greater the chance that it will only function in a single problem domain. An analogy exists to Heisenberg's Law of Uncertainty, which states that as the certainty in location of a particle is increased, the certainty of its momentum decreases, and vice versa. In terms of ESTs, offering more constrained ways of representing and manipulating knowledge decreases the applicability of the expert system tool. If a wide applicability is desired, then a tool offering flexibility in representation and control should be sought. The problem with this idea is that no one EST will be able to maximize both. The spectrum of today's ESTs ranges along a dimension of constraint in knowledge representation and manipulation. At one end of the spectrum is the almost totally unconstrained environment afforded by the LISP programming language. At the other end is the highly constrained environment that a full-fledged expert system tool offers. This paper surveys all software packages claiming to be an expert system building tool. Tools at the unconstrained end of the EST spectrum have not been included since they are essentially symbolic programming languages, and not actually expert system building tools by our definition. This survey thus covers the spectrum from production rule languages like ROSIE and OPS5, to hybrid tools such as KEE and ART. Perhaps the most limiting factor of an expert system tools' computing power is its host machine. Embedded in this dimension is a price trade-off between the machine and the tool. The most powerful ESTs run on expensive machines and tend to be large and expensive themselves. Alternately, an abundance of computing power may not be desired, or the user may wish to integrate the EST into existing small scale hardware such as a PC. These ESTs are not extremely powerful but run on inexpensive machines and are for the most part small and inexpensive. Although some ESTs run on both small and large computers, the very fact that one version runs on a PC, and another version on a LISP machine indicates that these tools will have differing capabilities (i.e., just because the name is the same does not mean you get the same product).