Assessing the Assessment: Comparing Algorithmic Impact Assessments and AI Audits

Abstract

As algorithmic systems play an increasingly influential role in addressing both public and private needs, lawmakers increasingly require a means through which they can ensure the public interest is protected. Advocates and policy makers have turned to audits and assessments of AI systems as potential mitigation strategies and transparency techniques. AI audits and assessments generally are intended to manage the effects of AI systems in ways that make them more accountable and safer, and also serve to make their effects on the world less unpredictable. Although they are often mentioned in the same breath as each other and are sometimes considered to be different sides of the same coin, AI audits and algorithmic impact assessments differ in important ways that have crucial consequences for managing the effects of AI systems. Audit and impact assessment are complementary governance practices, and regulators and practitioners must consider how these practices exist alongside each other, how their findings complement each other, and how they produce different forms of accountability from each other.

Keywords: audit, assessment, ai, artificial intelligence, accountability