Mathematical decisions and non-causal elements of explainable AI

Explainability requires a common language across disciplines: The paper identifies a critical gap in responsible AI discussions—moral philosophers, technologists, and policymakers lack shared terminology and frameworks. Establishing common ground between these groups is essential for effective governance and ethical implementation of algorithmic systems.

Different types of explanations matter differently for accountability: The paper proposes a hierarchy of explanation types for algorithmic decisions, suggesting that not all technical explanations adequately address moral and social concerns. Understanding why a decision was made may require different explanations than understanding how it was made.

Technical explanations alone are insufficient for responsible AI: The framework distinguishes between mathematical/technical aspects and non-causal elements of explainability, implying that purely technical documentation of how algorithms work cannot fully satisfy ethical and social accountability requirements in sensitive decision-making contexts.

Philosophical analysis of algorithmic decision-making is underdeveloped: The paper argues that while empirical data from social scientists and psychologists is growing, rigorous philosophical examination of the social implications of algorithmic systems remains limited—a gap that must be filled for responsible AI governance.

A multi-faceted framework can bridge moral and technical concerns: The paper proposes that a structured hierarchy of explanations can connect technical realities with moral implications, potentially enabling more productive dialogue among stakeholders about AI governance and ethics.