Expert (Knowledge-Based) Systems

Introduction

Knowledge representation and knowledge engineering are central to AI (artificial intelligence) research. Many of the problems machines are expected to solve will require extensive knowledge about the world. This fact was realized in the mid-1970s and implemented in the so-called expert systems, “computer programs that emulate the decision-making ability of a human expert” (Jackson, 1990, p. 2). This article covers the following topics:

• •

  Basic notions from the area of expert systems

• •

  Types of expert systems (diagnostic, generative) and examples of their applications

• •

  Architecture of an expert system (knowledge base, inference mechanism, uncertainty processing)

• •

  Building expert systems (life cycle, KADS methodology)

• •

  Knowledge engineering

Background

Expert systems represented the prominent research area within AI in the 1970s. In these times the search for a general problem-solving algorithm (using the formalism of the state space) has encountered its limitations in the domains that required specialized domain knowledge.

An expert system is:

The power of an expert system (ES) is derived from presence of a knowledge base filled with expert knowledge, mostly in symbolic form. In addition, there is a generic problem-solving mechanism used as an inference engine. Some other more-or-less typical features of expert systems are uncertainty processing, dialogue mode of the consultation, and explanation abilities.

Expert Systems

Expert systems should substitute human experts in the decision-making process. An expert system can play the following roles in this process:

• •

  Expert

• •

  Colleague

• •

  Assistant

The expectation that expert systems will play the role of an expert was too optimistic and unrealistic. The main reason for the fact that expert systems cannot always guarantee decision-making abilities of an expert is that a lot of knowledge of a real expert has a tacit form and thus cannot be transferred into the system. Another reason is that the users considering themselves to be experts were reluctant to follow the recommendations of the system if they disagree. So expert systems have to play either the role of a colleague or the role of an assistant. In the first case, the system has the same knowledge as its user, but as it never works under stress, never forgets, and can more quickly enumerate all possible solutions, the users check their decisions with the system, if necessary. In the latter case, the user is in fact an expert when compared with the knowledge of the system, but the system is useful to solve relatively simple, routine problems that usually prevail in various application areas.