The symposium, held virtually, was the third of a four-part summer series exploring the use of Artificial Intelligence (AI) and Machine Learning (ML) in federal government agencies.

Chai Feldblum, Partner at Morgan Lewis & Bockius, former Commissioner of the U.S. Equal Employment Opportunity Commission, and Public Member of ACUS, moderated the panel. She welcomed panelists:

• David Super, Carmack Waterhouse Professor of Law and Economics, Georgetown University Law Center
• Kristin Johnson, McGlinchey Stafford Professor of Law, Tulane University Law School
• Alex Givens, President and Chief Executive Officer, Center for Democracy & Technology

The panel discussed how AI systems can encode implicit bias and how federal program managers can protect against bias when they rely on AI systems. Panelists also discussed ways in which government can use AI to detect and counter biased behavior by regulated entities.

Each panelist presented a case study in which bias in government AI systems significantly impacted vulnerable populations within the United States. Professor Johnson began by discussing the ways in which the use of alternative data by AI in consumer credit-scoring platforms may advantage or disadvantage certain groups of people, leading to disparate impacts toward legally protected classes. Alex Givens highlighted the ways in which simple design and data entry errors in AI systems can result in devastating human impacts among Medicaid recipients. Similarly, Professor Super discussed how AI has been used to deny Supplemental Nutrition Assistance Program (SNAP) benefits to some of the United States’ most vulnerable populations. Together, these case studies highlighted the challenges associated with government’s reliance on AI and the need for appropriate checks and balances.