BONNIE E. JOHN
Associate Professor, Human Computer Interaction, Psychology
www

My primary research interest is in creating engineering models of human performance to be used in the design of human-computer interaction. Such models seek to optimize several criteria that distinguish them from traditional cognitive models: the ability to make zero-parameter predictions, the ability to be taught to and be used by computer systems designers, coverage of total tasks, and approximation.

Engineering models must be able to make predictions in the absence of a working, or even simulated, system because the predictions are needed early in the design process, where they can be used to shape the specifications of the system. Any parameters used to make the predictions must be set during the development of the modeling technique, not when the technique is being applied to a new situation. This does not mean that parameters have to be fixed constants for every situation, only that they must be determinable a priori. Tables of parameters covering a wide range of tasks can be used to estimate values for a new situation and these are created from research done in the development of the modeling technique. Such graphs or tables build in basic psychology so that computer designers can use the models effectively.

The target users of HCI engineering models are computer system designers, not usually trained psychologists or human factors experts. Thus, the designer cannot be counted on to bring psychological expertise to the task and the basic psychology must be built into the models. Tables of parameter estimates have already been given as an example of building in this expertise. In addition, the procedures must be clearly defined and representative examples must be presented to allow the techniques to be taught. This does not mean that the procedures have to be as structured as recipes in a cook book. However, there must be guidelines and rules about what to do in many representative situations so that a style of analysis can be developed that leads to good predictions.

The activities people perform when interacting with computers are quite varied. These include reading, searching for information on the CRT screen, processing information relevant to task goals, forming plans, typing, pointing, learning, making and recovering from errors, and a myriad of other activities that overlap and interact with each other. Engineering models must do more than focus on each activity in isolation, they must allow prediction of the overlap and interaction between activities. Coverage of total tasks is the most difficult criteria for engineering models to satisfy.

Engineering models of human performance must produce useful predictions at different levels of approximation. Some design situations require gross predictions of performance, e.g., selecting a word processor based on how long it will take to complete some benchmark tasks. Other design situations require much finer grained analyses, e.g. predicting where a combat pilot's visual attention is directed at every moment during a tactical maneuver so the effects of a new visual display can be assessed. Engineering models meet different design needs by providing approximations to the mechanisms that produce behavior. Models of mechanisms and processes allow operations to be grouped, averaged over, or ignored, as appropriate to the design situation.

One engineering model of human performance has been developed over the last decade, GOMS, which forms the basis for my continuing work. GOMS stands for Goals, Operators, Methods, and Selection rules, which are the components of the user's knowledge and task requirements necessary to predict user performance. GOMS has received a decade of academic interest and laboratory research, and has been demonstrated to be a valuable real-world evaluation and design tool for task-specific workstations (e.g., workstations for telephone operators). A current GOMS project models the behavior of architects using complex CAD systems to increase their productivity. I am currently working within the Soar cognitive architecture to extend GOMS and produce more complete engineering models. The current focus of this work is in modeling rapid interaction with the environment. One current project investigates how expert programmers interleave planning and coding and use the information displayed by their programming environment to structure their work. Another project investigates the mechanisms for problem solving and learning in the externally-driven, fast-paced domain of video games. A third models learning in an air-traffic controller task. My secondary research interest is in understanding the usefulness and usability of predictive HCI assessment techniques other than engineering models. I have an ongoing project to apply other techniques (e.g., Heuristic Evaluation, Cognitive Walkthrough, User Action Notation, Claims Analysis) to the design of complex systems (e.g., programming environments, multi-media authoring tools) to determine what types of usability problems these techniques can identify, how much effort it is to learn and use them, to what types of systems and at what stage of development they can best be applied, and how their results compare to empirical methods.