Auditing AI and Sociotechnical Systems
A workshop for people who study AI and algorithmic systems as they are actually built, deployed, and lived with.
A2S2 brings together researchers, practitioners, journalists, and affected communities working to audit sociotechnical systems. Our interest is not only in the models themselves, but also in how AI and algorithmic systems are integrated into products, platforms, organizations, and institutions, forming sociotechnical systems. What does it take to hold such systems accountable? Auditing questions can look very different after accounting for the interfaces, contextual social norms, incentives, data pipelines, and human decisions that surround a given model.
The workshop intentionally brings together academic and practitioner perspectives. Types of systems audited may include search algorithms, language models, ranking algorithms, recommender systems, computer vision systems, and more. We also hope to bring together people who "audit" under other names: AI assurance, independent verification, evaluation, red-teaming, compliance testing, investigative reporting, trust and safety, etc., alongside researchers developing methods for these activities. Recurring concerns include: how audit data is obtained, accessed, or generated; who is empowered to audit and on whose behalf; and what audits are actually able to change.