The purpose of this article is to review where we stand with regard to modeling the kind of cognition involved in human-computer interaction. Card, Moran, and Newell's pioneering work on cognitive engineering models and explicit analyses of the knowledge people need to perform a procedure was a significant advance from the kind of modeling cognitive psychology offered at the time. Since then, coordinated bodies of research have both confirmed the basic set of parameters and advanced the number of parameters that account for the time of certain component activities. Formal modeling in grammars and production systems has provided an account for error production in some cases, as well as a basis for calculating how long a system will take to learn and how much savings there is from previous learning. Recently, we were given a new tool for modeling nonsequential component processes, adapting the "critical path analysis" from engineering to the specification of interacting processes and their consequent durations. Though these advances have helped, there are still significant gaps in our understanding of the whole process of interacting with computers. The cumulative nature of this empirical body and its associated modeling framework has further highlighted important issues central to research in cognitive psychology: how people move smoothly between skilled performance and problem solving, how people learn, how to design for consistent user interfaces, how people produce and manage errors, how we interpret visual displays for meaning, and what processes run concurrently and which depend on the completion of prior processes. In the bigger picture, cognitive modeling is a method that is useful in both initial design (it can narrow the design space and provide early analyses of design alternatives), evaluation, and training. But it does not extend to broader aspects of the context in which people use computers, partly because there are significant gaps in contemporary cognitive theory to inform the modeling and partly because it is the wrong form of model for certain kinds of more global questions in human-computer interaction. Notably, it fails to capture the user's fatigue, individual differences, or mental workload. And it is not the type of model that will aid the designer in designating the set of functions the software ought to contain, to assess the user's judgment of the acceptability of the software, or the change that could be expected in work life and the organization in which this work and person fits. Clearly, these kinds of considerations require modeling and tools of a different granularity and form.
展开▼
