Samarbete mellan AI och människa och dess metakognitiva begränsningar
Jacob Uthardt · Aaltodoc (Aalto University) · 2026
Artificial intelligence is increasingly used to support problem-solving in different fields. While AI assistance often improves performance, recent research indicates it also affects users’ metacognition, that is, their ability to monitor and evaluate their own thinking and performance. This thesis reviews recent empirical and theoretical literature on metacognition in human-AI interaction with the aim of examining what effects AI has on its users’ metacognition, as well as how metacognitive processes affect joint human-AI performance. The findings suggest AI assistance can introduce several metacognitive constraints, such as overconfidence in performance and increased demands on metacognitive monitoring. The findings also indicate that metacognition plays a central role in the effectiveness of joint performance, as strong self-evaluation and understanding of the system support efficient delegation, while insufficient monitoring may contribute to overreliance on AI or accepting answers from AI without critical evaluation. Designing AI to support metacognition through transparency, uncertainty visualization and reflective questions may be one way to improve future human-AI collaboration outcomes.