On the Effect of Contextual Information on Human Delegation Behavior in Human-AI collaboration

Despite the remarkable capabilities of AI in specialized tasks such as image recognition and natural language processing, its integration into human workflows remains a complex challenge [11]. In particular, the effectiveness of AI is fundamentally linked to how people perceive, understand, and ultimately use these systems [11, 17, 46, 67]. One of the primary interaction forms in human-AI collaboration is the concept of delegation [2, 27, 35, 69]. For such computer-supported and cooperative work (CSCW) scenarios, it is important to design human-AI collaboration systems in a way that humans are supported to identify and delegate task instances to an AI for which it is capable of making a correct decision and that humans conduct the task instances themselves for which the AI is incorrect [27]. In AI-assisted decision-making domains, the need to discern delegation becomes particularly crucial. This is underscored by the human-in-the-loop concept [73], where human oversight is not only imperative for ensuring effective collaboration but may also be mandated by regulatory frameworks [18].