The Effect of AI Accuracy and Type of Feedback on Human Decision Making

As the decisions we make become increasingly influenced by suggestions from Artificial Intelligence (AI), it is important to understand how the form and the accuracy of their suggestions affect our behavior. In this study, we investigated whether our decisions were affected by the accuracy of the AI used and the form of feedback. Ten participants evaluated whether facial images were real or fake under four conditions, namely, with suggestions given by a high or low accuracy AI and with or without the AI confidence level provided. The results suggest that, in the case of a high accuracy AI, the concordance rate between the decisions the participants made and the suggestions from AI were higher when confidence levels were provided compared to when the confidence levels were not show. However, in the case of a low accuracy AI, the concordance rate was lower when confidence levels were shown. Additionally, the accuracy of participant's answers increased when confidence level was provided, for both the high accuracy and low accuracy AI. In conclusion, providing confidence level information alongside AI suggestions may change the decisions made by human and the effects may be different depending on the confidence level of the AI.