Theory-Based Design for Easily Learned Interfaces
by Polson, Peter G. & Lewis, Clayton H.

The problem adressed here is that of how to make walk-up-and-use applications for eg. bank teller-machines or airport information kiosks. Ease of use and ease of learning becomes crucial aspects. To reach these goals the authors propose a combination of several different models.

To begin with there is a need for a model that gives a representation of what knowledge that is needed to use an application effectively. For this purpose the GOMS method (Card et al., 1983) is chosen. The acronym stands for goals, operations, methods and selection rules. The model offers no quantification of the knowledge needed so that different tasks can be compared or training times can be predicted. For these purposes Kieras and Polson (1985) has proposed  an extention of GOMS called cognitive complexity theory (CCT). This extended model makes the assumptions that rules are cognitive units, that these are equally difficult to learn and that rules learned earlier can be transferred to a new task without any cost.

Knowing what knowledge and the amount of it that is needed is, however, not enough since the best design "is the one that minimizes the amount and complexity of the new knowledge necessary to use an application effectively", according to the authors. For this purpose the EXPL model (Lewis et al., 1987) is used. It breaks down the actions of the user and the responses of the system into smaller elements and tries to find causal connections between these links. By providing insight into the difficulty of specific aspects of a task EXPL provides a complement to GOMS and CCT, but EXPL has some limitations.

An EXPL analysis makes no use of goals which inhibits it from making certain connections that are obvious to a person. And it can not evaluate wether the prompts involved in the interaction adequately describes the appropriate actions. Finally the model uses a learnin-by-example heuristic and therefore needs examples to learn from. To achieve problem solving potential in unfamiliar domains the authors turns to classical problem-solving literature and import the concepts of problem space and search methods. So the key to designing easily learned interfaces is in facilitating the right problem-solving mechanism.

An integrated theory of exploratory learning of computer interfaces can now be put together using production representation of procedural knowledge from CCT, analysis of outcomes of actions from EXPL, the decision process from the puzzle-problem literature and the coordination of problem solving and learning from current cognitive architectures such as ACT* (Anderson, 1993). The resulting model is called CE+ and includes a problem-solving component that decides which action to take, a learning component to analyze the effects and store the results as rules and an execution component that coordinates execution of the rules with the problem-solving component. From this model some principles for design for successful guessing are derived (p 214):

1. Make the repertoire of availabe actions salient.
2. Use identity cues between actions and user goals as much as possible.
3. Use identity cues between system responses and user goals as much as possible.
4. Provide an obvious way to undo actions.
5. Make available actions easy to discriminate.
6. Offer few alternatives.
7. Tolerate at most one hard-to-understand action in a repertoire.
8. Require as few choices as possible.

These priciples are compared to the design principles put forward by Donald Norman (1988) with the conclusion tha CE+ offers "a conservative specialization of Norman's framework".

References

• Anderson, J. R. (1987). Skill acquisition: Compilation of weak-method solutions. Psychological Review, 94, 192-211.
• Card, S. K., Moran, T. P. & Newell, A. (1983). The psychology of human-computer interaction. Hillsdale, NJ: Lawrence Erlbaum Associates, Inc.
• Kieras, D. E. & Polson, P. G. (1985). An approach to the formal analysis of user complexity. International Journal of Man-Machine Studies, 22, 365-394.
• Lewis, C. H., Casner, S., Schoenberg, V. & Blake, M. (1987). Analysis-based learning in human-computer interaction. Proceedings of Interact '87, second IFIP Conference on Human-Computer Interaction, 275-280. Amsterdam: Elsevier.
• Norman, D. A. (1988). The psychology of everyday things. New York: Basic Books.