The Quest for Explainable AI and the Role of Trust

The prize for recent successes in the field of Artificial Intelligence (AI) has been black-box models with lack of algorithmic transparency. Especially the use of deep learning algorithms for machine learning tasks have produced opaque systems, which escape explanation of how results are provided. Apparently, AI-developers have been more occupied with creating powerful AI systems than with ensuring explainability. On this backdrop, this work in progress paper presents preliminary reflections and argues that interpretability viewed as transparency wrt system functionality does not necessarily bring explanatory clarity, whereas interpretability in the shape of post hoc explainability may work by giving reasons for why the system reacted in a certain way without full reference to its inner workings and black-box models. Hence, interpretability understood as post hoc explanations may satisfy a request for epistemic transparency. Still, future work on the paper is needed to clarify the degree to which the demand for interpretability can be ethically satisfied in the light of epistemic opacity. Moreover, it is generally acknowledged that explainability is pivotal to foster trust in AI. Therefore, the notion of trust is clarified by pointing to a distinction between trust as a fundamental condition in all human interaction and reliance as reflecting rational confidence in others or artifacts.