Raji, Ho et al: Audits Alone Won't Deliver AI Accountability
TL;DR
- Raji, Xu, Honigsberg and Ho argue existing algorithmic accountability policy has "neglected the lessons from non-algorithmic domains" on third-party participation.
- The paper surveys audit systems in financial, environmental and health regulation and finds their institutional design is "far from monolithic."
- Their conclusion: audits alone are "unlikely to achieve actual algorithmic accountability" without sustained institutional design work.
AI audits will not deliver accountability on their own without the institutional scaffolding that has grown up around audits in other regulated industries, a 2022 arXiv paper by Inioluwa Deborah Raji, Peggy Xu, Colleen Honigsberg and Daniel E. Ho argues.
The authors say existing algorithmic accountability policy approaches have "neglected the lessons from non-algorithmic domains," and survey audit systems in financial, environmental and health regulation to show that "the institutional design of such audits are far from monolithic." Their conclusion is blunt: "the turn toward audits alone is unlikely to achieve actual algorithmic accountability, and sustained focus on institutional design will be required for meaningful third party involvement."
No specific statute, agency or company is named in the abstract as the target of these recommendations.
Shared on Bluesky by 1 AI expert
Originally reported by arxiv.org
Read the original article →Original headline: Outsider Oversight: Designing a Third Party Audit Ecosystem for AI Governance