Abstract
Incorporating ethical principles of human-centered AI, such as fostering human autonomy and mindful decision-making, challenges the assumption that frictionless AI interactions are always preferable. Motivated by this tension, this study examines the effects of introducing interaction friction during Human-AI interactions to enhance human perceptions and decision-making. Using a Generative AI Assistant, we investigated the effects of Non-Friction AI (chatbot) versus Friction AI (chatbot plus summary or form) comparing them also to a control group. Participants’ perceptions of usefulness, confidence, trust, sense of agency, cognitive load, and performance were measured. Results showed that AI was perceived as more useful than no AI, but adding friction to chatbots did not significantly enhance perceived usefulness, cognitive load, or affect task performance. Cognitive load was significantly lower, while confidence and trust were higher, albeit not significantly, in AI-assisted modes.
Recommended Citation
Malaguti, Pauline; Karran, Alexander J.; Le, Di; Mortin, Hayley; Coursaris, Constantinos K.; Sénécal, Sylvain; and Léger, Pierre-Majorique, "Investigating Interaction Friction in Generative AI: Improving User Experience and Decision-Making" (2025). SIGHCI 2024 Proceedings. 26.
https://aisel.aisnet.org/sighci2024/26
