Towards a constraint analysis of human multitasking

When people conduct multiple tasks in tandem, they often interleave the various operators of each task. Just how these basic cognitive, perceptual and motor processes are ordered generally affords a range of possible multitasking strategies. We briefly outline how a cognitive constraint approach can potentially be used to explicitly explore a range of multitasking strategies, within the theorized constraints that operate on the human cognitive architecture. The power of this approach lies in the task description language, which allows higher-level task performance to be constrained by information requirements and resource demands of lowerlevel tasks. In general, this approach could provide an a priori method for identifying possible multitasking strategies. Consider while you are driving in your car, it is sometimes not too difficult to direct your attention away from the road, in order to complete a secondary task, such as dialing a number on a cell phone. In this example, there are obvious tensions between the two tasks; suspending attention from the primary task of driving for too long a time period might result in a collision, but completing the secondary task in a rapid and timely manner is probably also important. We briefly outline how an approach called Cognitive Constraint Modeling (CCM: Howes et al., 2004), can be used in a multitasking context to identify the optimal points at which to interleave a primary task, such as driving, in order to complete a secondary task, such as dialing a number on a cell phone. One of the aims of the cognitive modeling community is to provide an account of human performance on complex real-world tasks. Cognitive architectures (e.g., ACT-R: Anderson et al., 2004) allow models to be developed within a unified framework that integrate assumptions about the time course and information processing constraints that operate on the human system. For multitasking scenarios, like that described above, most previous models have tended to rely on a customized executive, which strategically controls the interleaving of the various task operators (see Salvucci, 2005, pp. 458-460). In response, Salvucci (2005) has proposed a general executive for controlling multitasking in the ACT-R cognitive architecture. The general executive assumes that control between two or more primary tasks is passed through a queuing mechanism. The queuing mechanism allows for the interleaving of the various operators that make up each primary tasks. In other words, the multitasking general executive provides a domain independent mechanism for integrating separate ACT-R task models. Salvucci (2005) has applied the multitasking general executive to the problem of integrating the control and monitoring required for driving, with the completion of secondary in-car tasks, such as dialing a cell phone number. The model was able to account for the increase in dialing time required while driving compared to baseline, and also the degraded steering that resulted from the introduction of the secondary dialing task. The multitasking general executive accounted for these results by assuming that a central cognitive bottleneck operates to limit performance, and that cognitive control must be sequentially ceded between the two tasks. However, a limitation of this approach is that the modeler has to make additional assumptions regarding the possible range of points in a task that control could be ceded. In other words, the precise operators in a task, at which control can be temporarily given up to a secondary task, must be specified by the modeler. This is a problem because performing one or more complex tasks in tandem affords the cognitive architecture a range of possible strategies with which to order the basic cognitive, perceptual and motor processes required for each task. Here, we briefly outline how an alternative approach, called CCM (Howes et al., 2004), might be used explicitly explore a range of possible strategies for multitasking.