Paper Number
ECIS2026-1727
Paper Type
SP
Abstract
Generative Artificial Intelligence (GAI) tools are increasingly adopted by knowledge workers, promising efficiency and quality gains, but they are also introducing perceptual risks. We present the design and results of a pre-study that investigates two such risks: the Illusion of Expertise, where professionals show inflated confidence in the quality of their GAI-assisted outputs, and the Erosion of Responsibility, where their felt responsibility decreases when assisted by GAI. An online experiment with knowledge workers was designed, and a respective pre-study was conducted. Participants completed a writing task under GAI-assisted and non-assisted conditions, followed by self-reporting confidence and responsibility perceptions, with initial findings indicating significant differences. The study was theoretically grounded in expectation effects and cognitive biases as well as attribution theory. As such, it aims to contribute to the information systems domain by conceptualising these perceptual risks as compounding phenomena. Furthermore, it aims to offer practical guidance on safeguarding quality assurance.
Recommended Citation
Mayer, Moritz; Schlögl, Stephan; and Castello, Alessio, "Generative Ai In Knowledge Work – A Breeding Ground For The Illusion Of Expertise And The Erosion Of Responsibility?" (2026). ECIS 2026 Proceedings. 13.
https://aisel.aisnet.org/ecis2026/cog_hbis/cog_hbis/13
Generative Ai In Knowledge Work – A Breeding Ground For The Illusion Of Expertise And The Erosion Of Responsibility?
Generative Artificial Intelligence (GAI) tools are increasingly adopted by knowledge workers, promising efficiency and quality gains, but they are also introducing perceptual risks. We present the design and results of a pre-study that investigates two such risks: the Illusion of Expertise, where professionals show inflated confidence in the quality of their GAI-assisted outputs, and the Erosion of Responsibility, where their felt responsibility decreases when assisted by GAI. An online experiment with knowledge workers was designed, and a respective pre-study was conducted. Participants completed a writing task under GAI-assisted and non-assisted conditions, followed by self-reporting confidence and responsibility perceptions, with initial findings indicating significant differences. The study was theoretically grounded in expectation effects and cognitive biases as well as attribution theory. As such, it aims to contribute to the information systems domain by conceptualising these perceptual risks as compounding phenomena. Furthermore, it aims to offer practical guidance on safeguarding quality assurance.