Human factors considerations in the design of a multiple source expert system for military applications (abstract only)

Abstract

In this paper we present a prototype expert system characterized by two major premises: there are multiple sources of knowledge within the knowledge base;human factors considerations must receive paramount attention.The domain of the prototype is AirLand combat planning. (Airland combat is a new Army warfare doctrine developed in the early 1980's. Its emphases include active defense, interdiction of the second echelon, early counterattack, and other tactical principles.) The potential users are Army division commanders and staff who are engaged in or are training for combat in Central Europe.Our prototype's knowledge sources are representative of those encountered in a wider class of applications, through its use of sources which may be incomplete, ambiguous and conflicting, such as: doctrinal knowledge found in policy statements, Army regulations, and professional manuals (FM 100-5). Doctrine can be viewed as a constraint on tactics, combat organization, fire and maneuver schemes, and command and control systems;knowledge from an expert from the domain environment (provided by the historian, Colonel Trevor N. Dupuy);distilled wisdom gleaned from the historical writings of great military thinkers, theorists, and commanders. The first sources to be included are the Maxims of Napoleon and material from von Clausewitz;previous experience with analysis of outcome of appropriate historical cases;results from a computer simulation (operations research) model, the Quantified Judgment Model (QJM). The QJM takes information about a combat situation and generates a prediction (based on many variables) of the victor. It also predicts advance rates, casualty rates, equipment loss and recovery rates, plus many other factors.[1] It is being used as a preprocessor for the expert system.Another important feature concerns the status of each of the knowledge components, i.e., whether the overall system is best envisioned as a collection of more or less autonomous expert systems governed by a controlling expert system, or whether the knowledge collection can be organized in some principled way to allow the multiple sources to be handled within a more homogeneous setting, the limit being a single expert system with accessibility to a (relational) database and other resources, especially simulation results. (A simulation capability is considered by some researchers [2,3] to be an important part of an expert system designed to provide task planning.) Consideration is being given to the implementation of the simulation portion on a parallel processing machine, specifically the Sequent 21000.Important human factors exist for many kinds of expert systems and, especially so, for the application study of the paper. [4] Included among these factors are: mode of the system, defining the attitude of the system with regard to the user, e.g., consultation, critique, advisory, alert, advocacy; [5]conversational style;architecture of the system and its relation to user stress;causal assessment and support of good human decision making in expert systems;avoidance of consequence buffering or transfer of responsibility from the user to the expert system.Constrained for a number of reasons to a microcomputer environment for the expert system portion of the prototype, we have chosen to utilize the Texas Instruments Personal Consultant Plus development tool to achieve such goals as: rapid development time, ease of explanation generation, availability of both forward- and backward-chaining control mechanisms, built-in functions for online help and other explanatory features, convenient knowledge base segmentation through use of frames, consistent user interface between development and user environments.