Abstract.Humans have the ability to efficiently decode human and human-like cues. We explore whether a virtual agent's facial expressions and gaze can be used to guide attention and elicit amplified processing of task-related cues. We argue that an emphasis on information processing will support future development of assistance systems, for example by reducing task load and creating a sense of reliability for such systems. A pilot study indicates subjects' propensity to respond to the agent's cues, most importantly gaze, but to not yet rely on them completely, possibly leading to a decreased performance.Keywords: Virtual embodied assistant; attention; task-switching; social information; cognitive processes 1
Introduction and ConceptHuman-machine interaction can be very complex, especially when it involves executive functions (attention allocation, priority setting, response scheduling, working memory, etc.) operating on multiple tasks. For example, in car-driving, operating industrial facilities, or air traffic control, humans often have to perform multiple tasks with time-varying demands. While Cognitive Scientists have investigated mechanisms of switching between multiple tasks and, e.g., how task interferences result in costs and errors [1], fields like human factors, software ergonomics, and humancomputer interaction strive to prevent multi-task situations or, since they are often unavoidable, to provide users with support mechanisms [2]. In this paper we study how a virtual agent can provide support in such multi-task scenarios by guiding attention with social signals, and what effects the presence of such an agent has on the user and the respective tasks. Interface agents are far from being a novel idea (see e.g. [3]). However, although positive social effects of virtual agents on processes such as learning have been confirmed [4], only a little research has investigated the detailed cognitive effects of an agent's presence on users that have to fulfill given tasks. We focus on situations where users have to carry out multiple tasks simultaneously (see Fig. 1, A). In most cases, these tasks interfere in some way and automatic performance is hard to establish. This means users have to employ a task-switching strategy that requires to manage attention accordingly: they would focus on one task temporarily, but have to monitor the other to decide whether a switching is necessary (Fig. 1, B). This split attention as well as frequent task switching is likely to hamper performance in both tasks. We explore whether a virtual agent can assist in this task-switching by guiding attention to those tasks that need to be attended to (Fig. 1, C). Fig. 1. Schematic rationale of our virtual assistant. In most situations, performing multiple tasks truly simultaneously is virtually impossible (A). Thus the user has to switch between the tasks, based on interlaced observation of both tasks which may hamper the performed task (B). We envision an agent to socially signal the need to switch tasks in resource-sparing ways (C).Our...