William B. Rouse

Human-Computer Interaction in the Control of Dynamic Systems

Modes of human-computer interaction in the control of dynamicsystems are discussed, and the problem of allocating tasks betweenhuman and computer considered. Models of human performance in avariety of tasks associated with the control of dynamic systems arereviewed. These models are evaluated in the context of a designexample involving human-computer interaction in aircraftoperations. Other examples include power plants, chemical plants,and ships.

Design of man—Computer interfaces for on-line interactive systems

An attempt is made to integrate a wide range of material into a conceptual structure for the design of man-computer interfaces for on-line interactive systems. Typical roles for the human in man-computer systems are considered. Suggestions for the design of systems are developed in discussions of displays and input devices, visual information processing, and mathematical models of human behavior. Possible developments and avenues of research in man-computer systems are suggested.

MATHEMATICAL CONCEPTS FOR MODELING HUMAN BEHAVIOR IN COMPLEX MAN-MACHINE SYSTEMS

Many useful mathematical models for manual control, monitoring, and decision-making tasks in man-machine systems have been designed and successfully applied. However, critical comments have occasionally been made, mainly by practitioners concerned with the design of complex man-machine systems. They especially blame models which seem to explain only data from abstract subtask experiments designed particularly for these models. In this review paper, an initial approach to bridging the gap between these two perspectives of models is presented. From the manifold of possible human tasks, a very popular baseline scenario has been chosen, namely car driving. A hierarchy of human activities is derived by analyzing this task in general terms. A structural description leads to a block diagram and a timesharing computer analogy. The range of applicability of existing mathematical models is considered with respect to the hierarchy of human activities in real complex tasks. Also, other mathematical tools so far not often applied to man-machine systems are discussed. The mathematical descriptions at least briefly considered here include utility, estimation, control, queueing, and fuzzy set theory as well as artificial intelligence techniques. Some thoughts are given as to how these methods might be integrated and how further work might be pursued.

Problem Solving Performance of Maintenance Trainees in a Fault Diagnosis Task

Forty trainees in an FAA certificate program participated in an experimental study of trouble-shooting of graphically displayed networks. The effects of network size, computer aiding, and training were considered. It was found that performance degraded as network size increased, improved with the use of computer aiding, and that skills developed with computer aiding were transferred to the unaided situation.

Adaptive Allocation of Decision Making Responsibility between Supervisor and Computer

Multi-task situations where supervisor and computer have intersecting decision making responsibilities are discussed. Adaptive allocation of task responsibility is espoused and formulated as a multi-queue, multi-server situation with a pre-emptive but non-competitive service discipline. Average delay in task performance and percent of decisions performed by the computer are predicted via simulation as a function of number of tasks, human-computer speed mismatch, and probabilities of various types of computer error. Prerequisites to the real-world realization of adaptive human-computer multi-task systems are considered and two laboratory investigations in this area are discussed.

Problem-Solving Skills of Maintenance Trainees in Diagnosing Faults in Simulated Powerplants

Eighty-six maintenance trainees in an FAA certificate program participated in two experimental studies of context-free and context-specific problem solving skills. The context-free training method included two previously reported tasks that involve the troubleshooting of graphically displayed networks. The context-specific task involved automobile and aircraft powerplants simulated on a new computer-based system called FAULT (Framework for Aiding the Understanding of Logical Troubleshooting). Results of the study indicated that it is possible to develop context-free diagnostic skills to be used in context-specific problems and that suboptimal diagnostic performance is largely due to not fully utilizing the information present in the structure of the problem.

Problem Solving Performance of First Semester Maintenance Trainees in Two Fault Diagnosis Tasks

Forty-eight first semester trainees in an FAA certificate program participated in an experimental study of trouble-shooting of two different types of graphically displayed networks. The effects of network size, redundancy, feedback, computer aiding, and training were considered. It was found that performance degraded as network size increased, degraded as the level of feedback was reduced, improved with the use of computer aiding, and that skills developed with computer aiding in one task were transferred to the other task.

Training Maintenance Technicians for Troubleshooting: Two Experiments with Computer Simulations

Aviation maintenance trainees participated in two experiments designed to assess the relative effectiveness of traditional instruction versus two types of computer simulation in the context of aircraft power-plant troubleshooting. Simulations ranged in nature from abstract, context-free problems to those involving specific aircraft power plants. Traditional instruction included reading assignments, television programs tailored to aircraft power-plant troubleshooting, and on-line quizzes. The first experiment compared the three training methods, and the second considered a mixture of the two computer simulations versus traditional instruction. The primary conclusion was that an appropriate combination of low- and moderate-fidelity computer simulations can provide sufficient problem-solving experience to be competitive with the more traditional lectureldemonstration form of instruction.

Experimental Studies and Mathematical Models of Human Problem Solving Performance in Fault Diagnosis Tasks

One of the reasons often given for employing humans in systems is their supposed abilities to react appropriately and flexibly in failure situations (Johnson, Rouse, and Rouse, 1980). On the other hand, we seem to hear increasingly about incidents of “human error”. The apparent inconsistency of these two observations can cause one to wonder what role the human should actually play. This question has led us to pursue a series of investigations of human problem solving performance in fault diagnosis tasks. Using three different fault diagnosis scenarios, we have studied several hundred subjects, mostly maintenance trainees, who have solved many thousands of problems. The results of these studies have led to the development of three mathematical models of problem solving behavior. The three tasks, results of the eight experiments, and the three models will be reviewed in this paper.

A Theory of Human Decisionmaking in Stochastic Estimation Tasks

Human decisionmaking in stochastic estimation tasks is considered. Theoretical and experimental results from mathematical psychology are reviewed. Concepts from stochastic estimation theory are used to develop a theory of human decisionmaking that employs optimal stochastic estimators with a short-term memory model, at least one long-term model, and a method of trading off estimates derived from each model. Approaches to testing the theory as well as the theorys implications are discussed. Also, the limitations of a linear theory are considered.