An executive-level computational model for human multitasking

Guoxi Zhang, Purdue University

Abstract

Current cognitive architectures are quite limited in their abilities to generate realistic human behavior. One reason for this is that cognitive architectures lack a model for the executive-level functions involved in human multitasking: prioritization and scheduling. Practitioners considering the use of a cognitive architecture in system design or usability testing would benefit greatly from an executive-level model since they would not have to write customized rules or functions to facilitate multitask performance. This research aims to build and test an executive-level computational model of human multitasking that could be adapted to any cognitive architecture. Rooted in control theory and integrating several motivational theories, empirical studies, and neuroscience research, a conceptual framework for human prioritization and scheduling is proposed. Borrowing from the manufacturing setting, shop floor control methodology (SFCM) is identified as an appropriate analogy to model human multitasking. Based on the conceptual framework and SFCM, an executive-level computational model was developed. Appropriate prioritization (dispatching) rules used in SFCM were selected and implemented in the computational model. The model was first tested by simulating a multitask scenario in which two tasks have fixed priorities and the good fit between the simulated and experimental data (R2 > 0.97) has been achieved. Then an empirical study manipulating valence, task processing time, and available time was conducted to further refine and validate the model. Given either two or four tasks to complete in a specified timeframe and a reward value associated with each task, subjects needed to prioritize tasks if they wanted to achieve maximum amount of award. The simulation results show that there is no significant difference between the simulated and empirical data with respect to the means of the total points earned, the means of the processing time for two-display tasks, the means of the processing time for five-display tasks, the first task picked by most subjects, and the task chosen sequence used by most subjects. Therefore, the computational model successfully replicated human behavior. The executive-level model of human multitasking can be fitted to any cognitive architecture and advances the state of the art of human performance modeling.

Degree

Ph.D.

Advisors

Feyen, Purdue University.

Subject Area

Industrial engineering

Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server
.

Share

COinS