Ethics-based AI auditing: A systematic literature review on conceptualizations of ethical principles and knowledge contributions to stakeholders

Abstract

This systematic literature review synthesizes the conceptualizations of ethical principles in AI auditing literature and the knowledge contributions to the stakeholders of AI auditing. We explain how the literature discusses fairness, transparency, non-maleficence, responsibility, privacy, trust, beneficence, and freedom/autonomy. Conceptualizations vary along social/technical- and process/outcome-oriented dimensions. The main stakeholders of ethics-based AI auditing are system developers and deployers, the wider public, researchers, auditors, AI system users, and regulators. AI auditing provides three types of knowledge contributions to stakeholders: 1) guidance; 2) methods, tools, and frameworks; and 3) awareness and empowerment.