An Investigation to reduce Overreliance on Explainable AI (XAI) in light of Two System Theory

As technology is evolving, there is a rise in the use of AI systems. The increased use of AI systems has revealed issues of gender and racial biases. To address these issues, explainable AI (XAI) is introduced, but the use of XAI has triggered various kinds of biases leading to issues such as overreliance. In this study, we seek to devise interventions to mitigate the issue of overreliance on AI by better understanding cognitive biases and acknowledging that different users have different cognitive abilities, and we need to be mindful of that when we design XAI systems. We will conduct multiple experiments using the recidivism dataset collected by ProPublica and to develop a better understanding of and solutions to mitigate the issue of overreliance. The findings from this research will allow us to design XAI systems better, improving user trust in AI and further improving AI adoption.