Task Delegability to AI: Evaluation of a Framework in a Knowledge Work Context Izabel Cvetkovic University of Hamburg cvetkovic@informatik.uni-hamburg.de Eva Bittner University of Hamburg bittner@informatik.uni-hamburg.de Abstract With the increased research focus on ways to use AI for augmentation rather than automation of knowledge-intensive work, a myriad of questions on how this should be accomplished arises. To break down the complexity of Human-AI collaboration, this paper pursues the identification of factors that contribute to the delegation of tasks to AI in such a setting, and consequently gain insights into requirements for meaningful task allocation. To address this research gap, we carried out an empirical study on an existing task delegability framework in a knowledge work context. We employed several statistical approaches such as confirmatory factor analysis, linear regression, and analysis of covariance. Results show that an adapted framework with fewer factors fits the data better. As for the framework factors, we show that the factor trust predicts delegability best. Furthermore, we find a significant impact of task on delegability decision. Finally, we derive theoretical and design implications. 1. Introduction Artificial intelligence (AI) is the science and engineering behind creating intelligent machines, particularly computer programs, which try to grasp and, to some extent, imitate human intelligence [1]. With the rise of AI, concerns about automation and consequent job loss have increased [2]. Galore research still focuses on finding ways to automate work with AI, without considering these consequences. However, in the last decade, researchers started giving rise to the importance of keeping the human-in-the-loop [3]. Furthermore, growing evidence of the advantages of Human-AI collaboration started appearing in the literature. This marks a shift to a collaborative rather than automation perspective of AI [4]. For example, Dellermann et al. [5], and Bittner et al. [6] argue that combining complementary strengths of human intelligence and AI leads to a better performance than each could achieve separately. Not only does this improve the outcome and group performance, but such a constellation also significantly contributes to mutual learning [7]. However, there is a long way to achieve the optimal collaboration for such socio-technical ensembles [8], and various research gaps to address in the first place. First of all, a realignment of the task allocation is necessary because the challenges of the modern world of work exceed the abilities of individuals [6]. In addition, the interplay between humans and tasks while collaborating with AI (assistants) as well as the outcomes of this collaboration necessitate further investigation. An AI assistant helps users achieve their tasks by interacting with them while using machine intelligence in form of, e.g., natural language processing, speech recognition, or machine learning [9]. AI capabilities are opening up new pathways for collaboration between knowledge workers and machines. Knowledge workers’ main attribute is knowledge. They apply this knowledge to develop products and services [10]. As knowledge work gains in complexity, it is becoming more and more challenging for individuals [11, 12]. Technological advances in the field of AI offer new design opportunities for the reorganization of knowledge work at the interface of humans and AI [13]. Many knowledge-based tasks can now be solved more effectively with AI technologies than with earlier technologies. For example, AI can be used to automate Q&A, enabling humans to focus on high-level interactions. But to take full advantage of the prospects of Human-AI collaboration in knowledge work, companies will have to redesign knowledge-work processes and jobs [14]. The existing potential for automation of tasks in knowledge work does not directly correspond to increased performance [15]. In contrast, AI assistants should be designed with the intention to augment, not replace human contributions [16]. Oeste-Reiss et al. [13] introduce Hybrid Knowledge Work Systems that continuously enable knowledge workers to acquire and transfer knowledge for the performance of their work tasks by means of hybrid intelligence [5]. Such systems Proceedings of the 55th Hawaii International Conference on System Sciences | 2022 Page 174 URI: https://hdl.handle.net/10125/79351 978-0-9981331-5-7 (CC BY-NC-ND 4.0)