Do the Details Matter? Comparing Performance Forecasts from Two Computational Theories of Fatigue
Abstract
We have been developing a computational theory of the effects of fatigue (especially sleep-related fluctuations in alertness) on the human cognitive system, implemented through mechanisms that impact existing components of the ACT-R architecture. These mechanisms include the suppression of activation in the declarative knowledge system, as well as brief breakdowns in the central production execution cycle, which we call micro-lapses. Through an iterative series of mechanistic architectural modifications, model implementations, and goodness-of-fit evaluations in task contexts like the Psychomotor Vigilance Test (PVT Dinges & Powell, 1985) and the Walter Reed Serial Addition/Subtraction Task (SAST Thorne, Genser, Sing, & Hegge, 1985), the theory has evolved to a state in which we have some confidence in its appropriateness. In other words, we feel increasingly confident that the mechanisms we are using to replicate and explain relevant empirical results are both sufficient and necessary for that purpose (Estes, 2002). This gives us a measure of confidence that it is reasonable, perhaps even advisable, to use the theory to make novel performance predictions in task contexts beyond those used for originally developing and evaluating the theory. So far, we have promising results from fatigued performance predictions in both the context of dual-tasking (Gunzelmann, Byrne, Gluck, & Moore, 2009) and also in the context of simulated driving (Gunzelmann, Moore, Salvucci, & Gluck, submitted).
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 01, 2009
- Accession Number
- ADA548488
Entities
People
- Glenn Gunzelmann
- Kevin A. Gluck
- Michael A. Krusmark
Organizations
- Air Force Research Laboratory