Human Cognition and the Expert Systems Interface Mental Models and Explanation Facilities

Abstract

Expert systems have been used to automate complex decision processes, enabling a human problem-solver to use them as an aid in making decisions. Often, these decision would not have been possible in the time available. The purpose of this research was to investigate human cognition and the expert system interface, exploring two key variables affecting user/expert system interaction: (1) User Mental Models and (2) the nature of the expert system Inference Explanations. This research explored the reasoning, problem solving and decision making processes of both expert systems and human decision makers. The research empirically established the variables affecting user/expert system interaction and contributed to the development of the theoretical relationship between human cognition and the use of intelligent machines. The experimental testbed consisted of an expert system interface program that operated on a personal computer (PC). The testbed was an event-driven program that allowed for both textual and graphic explanation of expert system decisions. To create different mental models of the expert system's inference logic, the expert system was taught in different ways to different subject groups, using text-only, graphic-only or text-graphic models. All subjects completed explanation/training in only one mental model condition, but interacted with each type of explanation mode to solve the problems.The testbed also recorded all subject interactions with the expert system. For each problem, the keystrokes and the way subjects traversed the inference tree were captured in data files. These files were used to plot the Problem Behavior Graphs (PBG) which described the subject's path through the problem space for each problem. In addition, before-task and after-task questionnaires were completed by subjects.The evaluation of data and questionnaires showed: (1) A better mental model increased user/expert system performance. Mental models improved as the conceptual model shifted from text-based to graphic-based to multimedia-based. (2) Multimedia (text-graphic) explanations lead to higher performance than textual explanation.