Bringing the Public Into AI Governance: A Representative Model for Consequential AI Systems
Anna Lenhart’s recent Federation of American Scientists policy memo, “The Federal Government Should Pilot a Decision Subject Representative Program for AI Systems,” offers an important governance proposal for an era in which artificial intelligence systems increasingly shape access to employment, housing, education, financial services, legal outcomes, and other consequential opportunities. The central insight of the memo is straightforward but significant: individuals affected by AI-enabled decisions should have a structured role in shaping how those systems are designed, procured, evaluated, and governed.
Lenhart argues that current approaches to AI fairness often rely too heavily on technical metrics, statistical testing, or post hoc assessments of bias. While these methods are important, they are incomplete. Fairness is not merely a computational problem. It is also a social, institutional, and contextual question. Different communities may experience the same automated decision system differently, and the harms that matter most to affected individuals may not be visible to developers, vendors, agencies, or technical auditors. For that reason, Lenhart contends that those who are subject to consequential AI decisions should be included in the governance process itself.
The article draws a useful analogy to the Food and Drug Administration’s Patient Representative Program. In that model, patients and caregivers are recruited, trained, and embedded into regulatory processes involving drug development, clinical trial design, risk-benefit analysis, product labeling, and advisory committee review. Lenhart does not suggest that AI systems and pharmaceutical products are identical. Rather, she uses the FDA example to show that lay expertise can be institutionalized in a serious, compensated, and procedurally meaningful way. Patient representatives do not replace scientific expertise; they complement it by introducing lived experience, practical judgment, and community perspective.
Lenhart proposes a similar “Decision Subject Representative Program” for consequential AI systems. Under this model, federal agencies would recruit and train individuals with direct experience of the contexts in which AI systems are deployed. For example, workers affected by automated hiring tools, students and parents affected by educational technology, or individuals affected by algorithmic financial or legal decisions could help agencies understand what risks, disclosures, appeal rights, benchmarks, and oversight mechanisms are most meaningful.
The memo offers three principal implementation pathways. First, the General Services Administration should pilot the use of Decision Subject Representatives in federal procurement of consequential AI systems. Because government procurement can shape private market behavior, embedding affected individuals into acquisition decisions could influence both public-sector and commercial AI practices. Second, the National Institute of Standards and Technology should include Decision Subject Representatives in future applied AI standards work, particularly where context-specific risks arise in hiring, education, finance, or criminal justice. Third, Congress should authorize flexible representative programs in future AI legislation so that agencies have clear statutory authority to involve decision subjects in regulatory design.
The proposal is especially valuable because it moves AI governance beyond abstract commitments to transparency and fairness. It asks how institutions can create actual mechanisms for participation, accountability, and contestability. Lenhart also recognizes the risk of superficial participation, often described as participation-washing. A representative program would need training, compensation, conflict-of-interest rules, transparency, and evaluation to ensure that participation is meaningful rather than symbolic.
For federal contractors, technology vendors, and agencies, the broader lesson is that AI governance is becoming less purely technical and more institutional. Procurement, standards, and regulation are likely to demand evidence not only that systems perform, but that their risks have been assessed in relation to the people most directly affected. Lenhart’s proposal provides a practical framework for making that shift credible.
Disclaimer: This post is for informational purposes only and does not constitute legal advice. The views summarized are based on Anna Lenhart’s article published by the Federation of American Scientists. Readers should consult appropriate legal, technical, or policy advisors before relying on any AI governance framework in a specific procurement, compliance, or regulatory context.