Scientific Panel on Global Standards for AI Audits
The Problem
AI systems have brought many opportunities and risks across sectors and applications. While many of us do not fully understand the AI systems that shape our lives, we are increasingly impacted by their use. This has led governments, institutions, and standard-setting bodies across the world to turn to AI audits to evaluate if and how AI systems work. These audits evaluate AI systems’ algorithms or models, the training and testing data used in the system, and the organizational and social contexts in which these systems are designed, developed, and deployed, for disparities in their outcomes and impacts. There is a growing AI audit ecosystem with myriad approaches and aims. There is, however, no global standards for auditing AI systems, despite calls from industry and government to produce them.
The IPIE's Approach
To advance this global conversation, the Scientific Panel on Global Standards for AI Audits is helping define the critical, comprehensive, and reasonable characteristics of audits. It is working to identify a set of criteria, evidentiary documentation, methodologies, processes, and aims that must be a part of global standards for AI audits. This Panel draws from across the social, computer, and engineering sciences, as well as the humanities, to surface points of international expert consensus on how public-facing AI systems should be assessed for their risks and impacts. The Panel’s expert membership comes from academia and civil society and is in close conversation with industry and policy thinking. Reports published throughout the Panel’s lifetime will include assessments of the existing AI audit landscape and recommendations for future standardization and policy work. The Panel aims to foster and further global conversations around AI systems centered on the public good.
Reports Published by the Panel:
