Know Thyself: The Relationship between Metacognition and Human-AI Collaboration

While collaboration between humans and artificial intelligence (AI) can yield enhanced performance due to complementarities of both entities, this potential is often not realized if human users manage the collaboration. This is because users need to be aware of their own capabilities and the capabilities of the AI. The users' ability to monitor their decision-making process is known as metacognition. In this short paper, we investigate the relationship between metacognition and human-AI collaboration. In an exploratory experimental study (n = 51), we quantify users' metacognitive efficiency and analyze how this affects their performance when receiving AI advice. We find that higher-performing subjects require higher levels of metacognitive efficiency to achieve improved collaborative performance with AI. We are the first to apply the metric of metacognitive efficiency to human-AI collaboration. Our findings have implications for designing AI advice shown to users based on both their individual capabilities and their metacognitive efficiency.