Delegation to Virtual Agents in Critical Scenarios: Influencing Factors and Immersive Settings

Favored by the rapid advance of technologies such as artificial intelligence and computer graphics, virtual agents have been increasingly accessible, capable, and autonomous over the past decades. As a result of their growing technological prowess, interaction with virtual agents has been gradually evolving from a traditional user-tool relationship to one resembling interpersonal delegation, where users empower virtual agents to autonomously carry out specific tasks on their behalf. Forming a delegatory relationship with virtual agents can facilitate the user-agent interaction in numerous aspects, particularly regarding convenience and efficiency. Yet, it also comes with problems and challenges that may harm users drastically in critical scenarios and thus deserves extensive research. This thesis presents a thorough discussion of delegation to virtual agents based on a series of studies my colleagues and I conducted over the past four years. Several factors --including agent representation, theory of mind, rapport, and technological immersion-- are examined individually via empirical approaches to reveal their impacts on delegation to virtual agents. A conceptual model featuring three interrelated dimensions is proposed, constituting a theoretical framework to integrate the empirical findings. An overall evaluation of these works indicates that users' decisions on delegating critical tasks to virtual agents are mainly based on rational thinking. Performance-related factors have a significant impact on delegation, whereas affective cues --such as rapport, agent representation, and theory of mind-- are influential only to a limited extent. Furthermore, the usage of immersive media devices (e.g., head-mounted displays) has a marginal effect on users' delegatory decisions. Thus, it is advisable for developers to focus on performance-related aspects when designing virtual agents for critical tasks.