The Effect of AI Advice on Human Confidence in Decision-Making

As artificial intelligence advances, it can increasingly be applied in collaboration with humans in decision-making contexts. However, questions on the design of different collaborative environments remain open. In the context of AI-advised human decision-making processes, we analyze the influence of AI advice on human confidence in the final decision. In a laboratory experiment, 458 subjects performed an image classification task. We compare their confidence over three treatments: i) a baseline case where subjects do not receive any AI advice; ii) where subjects receive AI advice; and iii) in addition to AI advice subjects also see the certainty of AI for its choice. Our results suggest that while AI advice can increase human overconfidence, this effect can be mitigated by augmenting the AI advice with its certainty. Our results not only contribute to the growing literature of human-AI collaboration, but also bear important practical implications for the design of collaborative systems.