Pragmatic Explainable Machine Learning

Abstract

My PhD thesis, "Pragmatic Explainable Machine Learning: Philosophy, Methods and Applications," develops a pragmatic, context-aware approach to explainable AI (XAI) and AI ethics for high-stakes settings. The extended abstract describes three interconnected research strands. First, a pragmatic-rhetorical account of XAI that evaluates explanation methods by how well they achieve context-specific goals within particular rhetorical situations. Second, technical and theoretical contributions to XAI research, including causal dependence plots (CDPs), theoretical limitations of non-causal XAI for fairness auditing, and fairness detectability as a quantitative diagnostic for evaluating explanations against causal baselines. Third, qualitative research on how AI ethics is shaped across societal contexts, including media discourse analysis and collaborative, value-driven responsible AI design with a nonprofit supporting survivors of gender-based violence.