The need for public participation in AI systems is acute. On October 30, 2023, President Joe Biden issued an Executive Order on Safe, Secure, and Trustworthy Artificial Intelligence (AI), calling for a coordinated, federal government-wide approach toward AI governance that requires “engagement with affected communities.” Exec. Order No. 14110, 88 Fed. Reg. 75191 (Nov. 1, 2023).
For instance, to prevent unlawful discrimination in federal programs using AI, the Order requires agencies to “consider opportunities to increase coordination, communication, and engagement about AI as appropriate with community-based organizations; civil-rights and civil-liberties organizations; academic institutions; industry; State, local, Tribal, and territorial governments; and other stakeholders.” Id. § 7.2(a).
The Office of Management and Budget (OMB) issued a proposed memorandum to implement the Order providing that “agencies must consult affected groups, including underserved communities, in the design, development, and use of the AI, and use such feedback to inform agency decision-making regarding the AI.” Off. of Mgmt. & Budget, Exec. Off. of the President, Proposed Memorandum for the Heads of Executive Departments and Agencies, Advancing Governance, Innovation, and Risk Management for Agency Use of Artificial Intelligence 19 (Nov. 1, 2023).
The OMB memorandum lists several mechanisms for soliciting ongoing public input, ranging from direct user testing to listening sessions to post-transaction customer feedback. And tellingly, the memorandum itself is open for public comment.
This emphasis on public participation echoes similar statements regarding AI in the White House’s AI Bill of Rights, the NIST AI Risk Management Framework, and Executive Order 14091 on Further Advancing Racial Equity and Support for Underserved Communities Through the Federal Government. It also accords with Executive Order 14094 on Modernizing Regulatory Review, requiring that all federal agency “regulatory actions … be informed by input from interested or affected communities.” Exec. Order No. 14094, 88 Fed. Reg. 21879
(Apr. 6, 2023).
AI poses serious risks to society that can be identified and countered with public participation. AI outcomes can be inaccurate, biased, and discriminatory. The scope and scale of AI mean that algorithmic failures and embedded biases have far-reaching effects, well beyond the discretion of a single government bureaucrat. Further, many AI systems are “black boxes” whose results are not easily explained or understood, even by their designers. Moreover, the developers who build AI systems and the agencies that adopt them often lack the perspective to understand AI’s real-world impacts.
To improve the government’s use and oversight of AI, federal agencies must ensure that public participation is meaningful rather than performative. See Michele Gilman, Beyond Window Dressing: Public Participation for Marginalized Communities in the Datafied Society, 91 Fordham L. Rev. 503 (2022). In its robust form, public participation consists of processes for people most likely to be affected by a given system to have influence on the system’s design and deployment, including decision-making power.
Public participation enhances the quality of decision-making by including a wider range of perspectives. It adds legitimacy to decisions because people gain trust from processes they understand and impact. It improves accountability by adding layers of scrutiny and discussion between the public and decision-makers.
Currently, initiatives for public participation tend to be more aspirational than operational. This is not unique to the United States; one global comparison of national strategies for AI governance concluded that they often referred to the value of public participation, but “were usually abstract and consistently overshadowed by other roles, values and policy concerns.”Christopher Wilson, Public Engagement and AI: A Values Analysis of National Strategies, 39 Gov’t. Info. Quarterly 1 (2022).
Fortunately, agencies implementing AI do not need to start from scratch in designing mechanisms for public participation. Environmental law has decades of experience with public participation mandates and is an apt analogy for AI. Both realms involve complex scientific and technical issues that are deeply intertwined with social and cultural values.
The National Environmental Policy Act of 1969 (NEPA) grants citizens the right to obtain government information about and provide public input in environmental decision-making. Over the decades, agencies have built upon the NEPA framework, spurred particularly by the environmental justice movement and its demands for equity. Executive Order 12898, issued by President Clinton in 1994, requires federal agencies to identify and address environmental impacts on minority and low-income populations, including with the involvement of the public. Exec. Order No. 12898, 59 Fed. Reg. 7629 (1994). This Order led to a government-wide infrastructure to effectuate environmental justice with public participation that can be a model for AI governance.
In the environmental realm, evidence shows that the benefits of participation can justify the costs. The public participation process results in less environmentally harmful decisions, as compared to initial proposals. The process may also impel agencies to abandon certain projects, knowing that they cannot survive the participatory gauntlet.
Some will object that AI is too complex for the public to participate in a useful way. However, people do not need to be able to code software to understand the ways that AI is impacting their lives. Further, where necessary, agencies can bring in technical experts to help educate participants and build their capacity, as has been done in various environmental and scientific participatory settings.
More serious qualms about public participation center on whether citizens will feel like they are engaged in a ritualistic process with no real opportunity to affect the outcome. In turn, this can generate distrust and alienate communities.
For all these reasons, public participation must be designed with intentionality and inclusivity, which means including impacted communities in the design of participatory mechanisms from the start. Michele Gilman, Democratizing AI: Principles for Meaningful Public Participation, Data & Society Research Institute (2023). Together, agencies and the public must establish the goals and mechanisms for participation, and these processes should be adapted throughout the life cycle of the AI system. Agencies must also dedicate financial and personnel resources to implement these processes.
Agencies must extend culturally appropriate outreach to impacted communities and remove logistical barriers so that less-advantaged citizens can participate. Internally, agencies need to foster a culture in which outside perspectives are valued, along with a willingness to shift course in line with input. Agencies should monitor and report on the methods and impacts of participation to refine participatory methods in line with emerging technologies. Finally, agencies should enact regulations codifying these participatory principles.
Computer scientists often explain faulty or biased technological outputs with the saying “garbage in, garbage out.” Public participation is a way to limit the garbage while expanding democracy.
“Public participation must be designed with intentionality
and inclusivity [of] impacted communities in the design of participatory mechanisms from the start.”
