Cary Coglianese (U Penn Law) and Nabil Shaikh (U Penn Law) have posted “Management-Based Oversight of the Automated State: Emerging Standards for AI Impact Assessment and Auditing in the Public Sector” in Yaghmaei, et al., eds., Global Perspectives on AI Impact Assessment (Oxford University Press, forthcoming 2024) on SSRN. Here is the abstract:
This paper focuses on the role for algorithmic auditing and impact assessment as a management-based approach to governing uses of artificial intelligence (AI) by government agencies. Because these uses can vary widely, as can their purposes, contexts, designs, and impacts, the responsible use of AI almost never can depend on compliance with a set of formulaic rules governing specific actions that agencies must take or outcomes to be avoided. Rather, AI governance will depend on adherence to a set of management-based standards that call for measures involving testing, validation, and monitoring throughout the lifecycle of AI design, development, and deployment. These measures of impact assessment and auditing will necessarily play a key role in helping to ensure that AI uses accord with principles of responsible AI. Already, governmental auditors from around the world have developed a series of frameworks and standards for auditing governmental use of AI. This paper describes and compares the main elements of such an approach by reference to standards issued in the United States, Canada, and Europe. The paper offers a synthesis of the main elements of AI impact assessment in the public sector and explores what is realistic to expect from the use of impact assessment and auditing as a tool for governing public sector AI.