Frank E. Ritter
January 1999
In the next five years I expect to work on one main, multifaceted problem: creating a model computer user that learns. The main idea is to exploit and develop a unified theory of cognition by focusing on interaction with external environments using multiple data sources, multiple tasks, with a natural audience for the results. In order to be fully realized and self-sustaining, this will be a large project, and it will take time.
Usability results are often fragmented and hard to apply to interface design. Just putting them together for their own sake is not happening as much as in other fields. The lack of use and reuse is one reason&emdash;few unified theories or models created in them get reused enough to improve over time. One way to keep a web page current is to use it as a set of bookmarks. It has reuse and an audience, so it is kept up to date. Similar unified theories arise in electronics and static physics where the practical uses (electronics and bridge design) pull along and reuse the theory. A way to improve theories of cognition is to find an application that needs to use and reuse the models. A user model that can help evaluate interfaces or instructional material offers such an opportunity.
Progress on this task will require science and technology. Better theories of learning will be necessary. Current theories of learning don't usually take account of interaction. Including interaction changes learning models (Bass, Baxter, & Ritter, 1995; Jones & Ritter, 1998; Ritter & Bibby, 1997) . There also exist useful models by others that could be recruited (e.g., Howes & Young, 1996) . Integration of existing psychology theories realized as software will be necessary as well. Software engineering and code reuse may suggest lessons to help us build larger (information processing) theories in psychology, although the domain is different.
The technology problems are clear, and I have started work on several of them. In order to build and use a model user, we need better tools for building models (Ritter, Jones, & Baxter, in press; Ritter & Larkin, 1994) . We need the ability to tie models to computer interfaces routinely, so that they can interact with what subjects see. This is nearly in hand, at least to the extent that we understand what is necessary (Baxter, Ritter, Jones, & Young, submitted) . Future work here will focus on creating a more complete theory of vision and motor control, and how they influence interaction. Existing work here too can be used (Anderson & Lebiere, 1998; Meyer & Kieras, 1997) .
We will also need data. While there are large compendiums of regularities of human behavior (e.g. Boff & Lincoln, 1988) , the models that get created will need additional data to test that their internal components work together. We already have the ability to gather data routinely from experimental apparatus, that is, interfaces (Ritter, in preparation; Ritter et al., in press) .
We will need better analysis techniques, in order to understand what users do (e.g. Kuk, Arnold, & Ritter, in press) and software tools to analyze this data (Ritter, in preparation) . We have enough analysis techniques to know where the models can be improved, what we need are techniques that are more proscriptive, that tell us where and how to improve the models. Previous work on automated modeling suggest this might be possible (Langley & Ohlsson, 1989; Ritter, 1991) .
This approach will offer theoretical and practical payoffs. The theoretical payoff will be a better theory of users. Where this approach has been practiced to a limited extent, new and important questions have arisen about how people think and learn. Putting theory up against data, allowing a rapid testing cycle, is a feature that supports rapid progress, and we are developing an approach to do this for cognitive models (Gobet & Ritter, submitted 12/98) , and to use this approach to test theories of development (Jones & Ritter, 1998) and instruction.
There are several practical payoffs. Consider just two examples. One would be a model student that could help differentiate teaching strategies. This idea has been proposed and vetted for a long time (Ohlsson, 1992) but there just has not been the reuse that pushed a single theory forward. Another practical payoff that is likely to drive this work more is using the resulting model as part a design aid for interface designers. The model could not only provide general comments on good and bad aspects of the interface design like a critic, but it could base its comments on specific problems that it had, and provide estimated times and behaviors that could be compared across interface designs. This would increase the audience for HCI theories, and I believe would force us and allow us the resources, data, and motivation to improve our theories.
Who would fund this work? I've been discussing this approach with British Telecom, NASA, and the Australian Defense Department. ONR (US) has provided modest and peripheral support to this approach. DERA, roughly the UK equivalent of ONR, has provided support in the past. The role of cognitive models in design is one of the subjects of a recent book by the National Research Council (Pew & Mavor, 1998) , which suggests this will be a topic of funding in the future. This work includes many projects that are the right size for students, so students can not only be educated this way, but can help (and this has happened several times in the past).
As a step towards creating a model user that can interact with multiple interfaces, we are building a model eye and hand in Tcl/Tk. We have in hand 50 example phone interfaces. We can instrument these interfaces, so a rather large data/behavior comparison is possible. We will learn some real generalities about behavior and be a step closer to general user model.
Anderson, J. R., & Lebiere, C. (1998). The atomic components of thought. Mahwah, NJ: Lawrence Erlbaum.
Bass, E. J., Baxter, G. D., & Ritter, F. E. (1995). Using cognitive models to control simulations of complex systems. AISB Quarterly, 93, 18-25.
Baxter, G. D., Ritter, F. E., Jones, G., & Young, R. M. (submitted). Extending user interface management systems to support cognitive models as users. .
Boff, K. R., & Lincoln, J. E. (1988). Engineering data compendium: Human perception and performance. Wright-Patterson Air Force Base, OH: Harry G. Armstrong Aerospace Medical Research Laboratory.
Delaney, P. F., Reder, L. M., Staszewski, J. J., & Ritter, F. E. (1998). The strategy specific nature of improvement: The power law applies by strategy within task. Psychological Science, 9(1), 1-8.
Gobet, F., & Ritter, F. E. (submitted 12/98). Individual Data Analysis and Unified Theories of Cognition: A methodological proposal. .
Howes, A., & Young, R. M. (1996). Learning consistent, interactive, and meaningful task-action mappings: A computational model. Cognitive Science, 20(3), 301-356.
Jones, G., & Ritter, F. E. (1998). Initial explorations of simulating cognitive and perceptual development by modifying architectures. In Proceedings of the 20th Annual Conference of the Cognitive Science Society. 543-548. Mahwah, NJ: Lawrence Erlbaum.
Kuk, G., Arnold, M., & Ritter, F. E. (in press). Using event history analysis to model the impact of workload on an air traffic tactical controller's operations. Ergonomics.
Langley, P., & Ohlsson, S. (1989). Automated cognitive modeling. In AAAI-84. 193-197. Los Altos, CA: Morgan Kaufman.
Meyer, D. E., & Kieras, D. (1997). A computational theory of executive cognitive processes and multiple-task performance: Part 1. Basic mechanisms. Psychological Review, 104(1), 3-65.
Ohlsson, S. (1992). Artificial instruction: A method for relating
learning theory to instructional design. In M. Jones & P. H. Winne (Eds.), Adaptive learning environments: Foundations and frontiers. 55-83. Berlin: Springer-Verlag.
Pew, R. W., & Mavor, A. S. (Eds.). (1998). Modeling human and organizational behavior: Application to military simulations. Washington, DC: National Academy Press.
Ritter, F. E. (1991). Towards fair comparisons of connectionist algorithms through automatically generated parameter sets. In Proceedings of the Thirteenth Annual Conference of the Cognitive Science Society. 877-881. Hillsdale, NJ: Lawrence Erlbaum Associates, Inc.
Ritter, F. E. (in preparation). Dismal: A spreadsheet for sequential data analysis and HCI experimentation. .
Ritter, F. E., & Bibby, P. A. (1997). Modelling learning as it happens in a diagramatic reasoning task (Tech. Report No. 45). ESRC CREDIT, Dept. of Psychology, U. of Nottingham.
Ritter, F. E., Jones, R. M., & Baxter, G. D. (in press). Reusable models and graphical interfaces: Realising the potential of a unified theory of cognition. In U. Schmid, J. Krems, & F. Wysotzki (Eds.), Mind modeling - A cognitive science approach to reasoning, learning and discovery. Lengerich: Pabst Scientific Publishing.
Ritter, F. E., & Larkin, J. H. (1994). Using process models to summarize sequences of human actions. Human-Computer Interaction, 9(3), 345-383.