Mitigating belief projection in explainable artificial intelligence via Bayesian Teaching
This work is significant for improving human understanding and prediction of AI behavior in XAI, particularly for mitigating the common cognitive bias of belief projection.
This paper addresses the problem of belief projection in explainable AI (XAI) where humans assume an AI's decisions will match their own. By explicitly modeling the human explainee via Bayesian Teaching, the authors demonstrate an improved ability for participants to predict AI judgments, moving them away from their prior belief.
State-of-the-art deep-learning systems use decision rules that are challenging for humans to model. Explainable AI (XAI) attempts to improve human understanding but rarely accounts for how people typically reason about unfamiliar agents. We propose explicitly modeling the human explainee via Bayesian Teaching, which evaluates explanations by how much they shift explainees' inferences toward a desired goal. We assess Bayesian Teaching in a binary image classification task across a variety of contexts. Absent intervention, participants predict that the AI's classifications will match their own, but explanations generated by Bayesian Teaching improve their ability to predict the AI's judgements by moving them away from this prior belief. Bayesian Teaching further allows each case to be broken down into sub-examples (here saliency maps). These sub-examples complement whole examples by improving error detection for familiar categories, whereas whole examples help predict correct AI judgements of unfamiliar cases.