讲座信息
题目:Causality, Plausibility, and Backtracking Counterfactual Generation
报告人: 张冀寄 (香港中文大学)
时 间: 2026年2月4日 (周三) 19:30-20:30
联 系 人: 吴鹏(北京工商大学)
报告摘要
A popular approach in explainable artificial intelligence is to generate so-called counterfactual explanations for outputs of predictive models. In this talk, I attempt a philosophical critique of key aspects of this approach and argue for a tension between two constraints on a good counterfactual explanation in this context, often known as 'plausibility' and 'causality'. I suggest a way to mitigate this tension based on recent work on counterfactual generation that allows backtracking.