Human Error Modeling

Period of Performance: 07/02/2003 - 04/30/2004

$100K

Phase 1 SBIR

Recipient Firm

Micro Analysis and Design, Inc.
4949 Pearl East Circle, Suite 300
Boulder, CO 80301
Principal Investigator

Abstract

Human error can often be attributed to poor system design. Human factors and Human Computer Interaction (HCI) techniques can be used to evaluate a new system design with varying degrees of effectiveness and cost. Computer simulation offers a cost-effective alternative for evaluating system designs. Computational cognitive architectures like ACT-R now provide a way to model human performance as part of the evaluation of a proposed interface. Integrating cognitive modeling technologies with more traditional HCI modeling tools will allow designers to identify interface components or user interaction requirements that can lead to errors. This proposal describes the design and development plan for a software tool that will allow a user to specify the interface of a system design and, from that specification, generate an operator model without having to understand complex details of the underlying cognitive model. Once the interface and the human model are instantiated, the user can exercise the integrated models to detect potential errors that are inherent in the design. The tool we propose will also trace the errors that occur during the simulation to the underlying cognitive model so that the user can reconstruct psychologically principled explanations and consider possible design improvements suggested by the tool. The completion of the Phase I objectives will lay the foundation needed to develop a general tool that will provide an innovative means for identifying, predicting and explaining potential HCI errors early in the design cycle of a new interface. The proposed effort will also have implications for future research in cognitive modeling. Not only will this work extend recent efforts to provide a natural account of errors within an existing cognitive architecture, it will also demonstrate the extent to which a complex cognitive activity, like interface control, can be represented in terms of a handful of generic cognitive "modules." Such modularization would have significant research implications regarding the development and re-use of cognitive models generally.