Position: In Defense of Post-hoc Explainability

Recent discourse in AI ethics emphasises the importance of truthful and complete explanations, particularly in high-stakes scenarios with significant real-world consequences. This stance stems from concerns about the potential incompleteness or misleading nature of post-hoc explanations, which may not fully capture an AI system's decision-making process. However, this position paper argues that incomplete post-hoc explanations can be epistemologically justifiable in such scenarios. We introduce Computational Interpretabilism (CI), a philosophical stance that advocates for interpretability whilst giving equal epistemological weight to intrinsically interpretable models and post-hoc methods. CI defends post-hoc explanations, proposing a balanced framework for enhancing AI transparency and trustworthiness.