Efficient multitasking: parallel versus serial processing of multiple tasks

Abstract

In the context of performance optimizations in multitasking, a central debate has unfolded in multitasking research around whether cognitive processes related to different tasks proceed only sequentially (one at a time), or can operate in parallel (simultaneously). This review features a discussion of theoretical considerations and empirical evidence regarding parallel versus serial task processing in multitasking. In addition, we highlight how methodological differences and theoretical conceptions determine the extent to which parallel processing in multitasking can be detected, to guide their employment in future research. Parallel and serial processing of multiple tasks are not mutually exclusive. Therefore, questions focusing exclusively on either task-processing mode are too simplified. We review empirical evidence and demonstrate that shifting between more parallel and more serial task processing critically depends on the conditions under which multiple tasks are performed. We conclude that efficient multitasking is reflected by the ability of individuals to adjust multitasking performance to environmental demands by flexibly shifting between different processing strategies of multiple task-component scheduling.

Keywords: bottleneck; cognitive control; crosstalk; dual-tasking; functional bottleneck; multitasking; parallel versus serial processing; psychological refractory period (PRP).

FIGURE 1

Schematic illustration of serial task processing (A) and different forms of parallel processing (B,C) of two tasks in the framework of an assumed capacity-limited central processing stage. Dashed lines illustrate the changes in result patterns when assuming different forms of parallel processing. Note that although theoretical models are explained in terms of response time (RT) pattern, the same logic also applies to error rates. (A) Illustration of the response-selection bottleneck (RSB) model as explanation for severe dual-task processing limitations (Pashler, 1994). Each task consists of different processing stages (i.e., P, perception; RS, response selection; MR, motor response). Processing in some stages can occur in parallel (in white). Processing of other critical stages cannot occur simultaneously (shaded), because they rely on the same capacity-limited processing channel. When both tasks overlap substantially (e.g., short stimulus onset asynchrony, SOA), Task 2 (T2) processing is interrupted, because RS2 processing has to wait until RS1 processing is completed (psychological refractory period, PRP). At long SOA, no interruption occurs, as critical stages do not overlap. This results in the typical pattern of performance decrements in T2 at short SOA (high dual-task load) compared to long SOA (low dual-task load). Task 1 (T1) processing is only little affected by temporal task overlap. (B) Crosstalk refers to the observation that T2 processing impacts on T1 processing, which has been taken as evidence for parallel processing despite an assumed RSB. Crosstalk effects are typically measured in response latency in T1 (RT1). The impact of T2 processing on central stage processing in T1 can be both beneficial or costly with decreasing or increasing RT1, respectively (e.g., Koch and Prinz, 2002). Importantly, any influence of T2 processing on T1, shortening or prolonging RT1, will back-propagate onto T2 (Ferreira and Pashler, 2002; Miller and Reynolds, 2003; Schubert et al., 2008). Changes in RT1 due to crosstalk should thus also be obtainable in response latency in T2 (RT2). Theoretically, crosstalk effects are not compatible with classical models of a single-channel theory (e.g., RSB model) and favor explanations in terms of capacity sharing (see C). However, assumptions of serial processing according to the RSB model can be preserved when assuming that different sub-components of RS2 can operate in parallel. Some authors thus distinguish response activation (RA) processes from more classical response-selection processes as the basis for interacting central components between two tasks (Hommel, 1998; Lien and Proctor, 2002; Schubert et al., 2008). (C) Capacity models assume that the central bottleneck is not immutable but flexible. The processing limitation arises, because two central processes require access to the same cognitive resources. Available resources are divided between the two tasks for the period during which both central stages overlap. The allocation of resources to the tasks at hand depends on task factors (e.g., instruction, incentives). Extreme forms can mimic a central bottleneck, with 100% resources allocated to T1 and 0% to T2. The more resources are shared between the two tasks (e.g., 70/30 or 50/50), the higher the RT1 increase and RT2 decrease at short SOA. This resource allocation is assumed to be realized by mechanisms of cognitive control (for details, see text).