Cognitive Forcing for Better Decision-Making: Reducing Overreliance on AI Systems Through Partial Explanations
Authors: S de Jong, V Paananen, B Tag
Published: 2025
Publication: Proceedings of the ACM on ..., 2025 - dl.acm.org
Partial explanations encourage critical thinking and reduce user overreliance on incorrect AI suggestions, with performance varying based on individual need for cognition and task difficulty.
Methods: Two experiments were conducted: (1) participants identified shortest paths in weighted graphs, and (2) participants corrected spelling and grammar errors in text, with AI suggestions accompanied by no, partial, or full explanations.
Key Findings: Effectiveness of partial explanations in reducing overreliance on incorrect AI suggestions, and interaction of explanation type with task difficulty and user need for cognition.
Limitations: Effectiveness of partial explanations is task-dependent and may not outperform full explanations; task difficulty and individual differences were not exhaustively examined across broader demographics or AI systems.
Institution: Niels van Berkel: Aalborg University, Sander de Jong, Ville Paananen, Benjamin Tag: Monash University
Research Area: Cognitive Forcing, Human-AI Interaction, AI Explainability (XAI),Decision-Making in AI Systems.
Discipline: Human-Computer Interaction (HCI),Artificial Intelligence
Sample Size: 474 participants
Citations: 14
DOI: https://doi.org/10.1145/3710946