Modeling Dual-Task Concurrency and Effort in QN-ACTR and IMPRINT.

Abstract

Computational cognitive models have wide ranging applications from reducing the time and cost of task and interface analyses to the discovery of new human cognitive phenomena. We investigate the use and limitations of IMPRINT, a task network simulation tool, and develop an extension to improve the modeling of task component execution limits in multi-task performance under high workload. The extension is implemented as a Soar agent that moderates task execution akin to executive processes in EPIC. We show that an IMPRINT model of a UAV operation task with the extension exhibits qualitatively distinct workload management strategies also observed in human performance of the same task. Next, we develop QN-ACTR models of a concurrent addition and targeting task and collect empirical data of human performance on the tasks to validate the models' predictions of execution time and a time sharing concurrency metric. We also use the empirical data to validate an IMPRINT model of the addition and targeting tasks. Both QN-ACTR and IMPRINT models capture the primary effects of variable task difficulty parameters on execution time and concurrency. Model inaccuracy at the subtask level provides evidence for the use of visual spatial memory during complex addition. In a second experiment with similar tasks, we introduce an incentive to examine the effects of effort on execution time and concurrency in dual task performance. Incentive induced effort is found to increase performance on the rewarded dimension without an increase in the time sharing concurrency metric, suggesting that the performance improvements are not derived from an increase in task scheduling efficiency or resource sharing but from the same improvements found in single task conditions. The QN-ACTR task models are modified to account for the increased effort by adjusting base level parameters and are validated with the empirical data.