Why Democratic Deliberation and Governance of AI Matters

public opinion and public policy on ai in the u.s

AI is not a single product or method but instead commonly refers to various models, applications, and tools that attempt to mimic and augment how we learn, make decisions, and express ourselves. Although the mathematical and computational foundations of AI were established as early as the 1940s, the recent development of generative AI models, including the watershed release of OpenAI’s GPT-3 model in 2020, has made AI much more powerful and accessible. Developers now market generative AI models and applications for broad public use. These tools respond to natural language prompts rather than code and quickly generate text, images, and audio that closely resemble content created by humans.

The proliferation of AI technology and its growing embeddedness in daily life raises widespread concerns. In late 2025, the Pew Research Center conducted a survey that found that nearly three-fourths of U.S. adults agreed that it was “extremely” or “very” important for the public to understand what AI is, and almost two-thirds responded that they “would like more control over how AI is used in their lives.” The survey also showed emerging differences in AI use and concerns by age, gender, race, and socioeconomic status.

OPEN IN VIEWER

The federal government’s use of facial recognition and other AI technology reveals a top-down, opaque approach to AI policy that threatens civil rights and liberties while leaving little room for public input and accountability.

iStock Photo // gorodenkoff

Current governmental efforts to regulate AI primarily occur through executive actions and state legislation. At the federal level, presidential administrations have increasingly relied on executive orders, agency policies and guidelines, and privately negotiated agreements with tech firms to govern AI development and use. For their part, nearly all U.S. states support AI regulation. According to the National Conference on State Legislatures, all 50 states introduced legislation on AI in 2025, with 38 states passing new laws on AI. These new laws now include California’s Transparency in Frontier AI Act and New York’s Responsible AI Safety and Education Act, which establish broad regulatory frameworks for AI development that emphasize transparency and accountability.

AI, then, is not just a technological means to a political end but also a source of power in and of itself.

Although state policies on AI can help, the sweeping effects of AI require ongoing federal leadership and coordination. In a 2024 article I co-authored with CUNY Graduate Center sociologist Leslie McCall, we argued that two approaches to AI policymaking are emerging: a safety-based approach and an equity-based approach. The safety-based approach defines risks as existential threats to humans and national security and focuses on managing AI-related risks through market-based solutions. This approach embodies what University of Michigan sociologist Elizabeth Popp Berman calls an economic style of reasoning, which has dominated public policy since the 1960s. The Biden administration partially implemented it, and tech firms strongly supported it at the time. In contrast, the equity-based approach focuses on creating equitable AI systems through affirmative protection of civil rights and liberties. Civil-society organizations advocated for this approach, and the Biden administration also partially enacted it through its landmark policy document, the Blueprint for an AI Bill of Rights, the culmination of efforts led by Institute for Advanced Study and Princeton sociologist Alondra Nelson.

Policy debates about AI during the Biden administration centered on these two approaches, which sought different outcomes and prioritized different targets for intervention. What they shared, however, was a common commitment to proactively regulating AI to keep Americans informed and protected. The second Trump administration has taken a different approach. On his first day in office, President Trump signed an order rescinding previous attempts by the Biden administration to regulate AI. He has since issued a “preemption” order that prohibits states from establishing regulations that his administration considers to be either overly onerous or ideologically biased, as well as an order that prohibits federal agencies from procuring “woke AI,” defined as AI models and applications that encode ideologies such as “diversity, equity, and inclusion.” Last year, Trump issued eight executive orders addressing AI alone.

The Trump administration’s approach is often labelled “accelerationist” because much of its rhetoric and policies focus on building and deregulating AI. However, the administration’s actions are more accurately understood as a hegemony-based approach to AI. This approach focuses on asserting unilateral control over AI development and use and rapidly building and using specific forms of AI technology to advance the administration’s political goals at home and abroad. AI, then, is not just a technological means to a political end but also a source of power in and of itself. The short-lived Department of Government Efficiency (DOGE), for instance, used AI to identify and eliminate federal jobs, contracts, and grants that did not align with the administration’s priorities. Immigration and Customs Enforcement (ICE) uses facial recognition and other AI technology to carry out the administration’s goals of identifying undocumented immigrants, as well as protesters. The recent conflict between the Pentagon and Anthropic provides perhaps the clearest illustration of the administration’s interest in controlling AI technology rather than simply accelerating its development. Indeed, all of these AI applications reveal a top-down, opaque approach to AI policy that threatens civil rights and liberties while leaving little room for public input and accountability.

bolstering democratic deliberation and governance of ai

Democratic governance of AI requires ongoing and robust deliberation. Fortunately, these conversations are alive and well among sociologists. A recent Socius special issue on AI, edited by Kelly Joyce and Taylor Cruz, examines how we understand and interact with AI technology in domains such as work and health. This essential research not only aids our understanding of the societal benefits and harms of AI but also identifies and amplifies public preferences, particularly those of marginalized groups who are often involuntarily exposed to AI technology.

Democratic governance of AI requires ongoing and robust deliberation.

Other new research directly investigates the role of leading AI developers in mediating the political engagement of citizens. Although all major companies advance their economic interests through political influence, leading tech firms can now mediate the very processes of political learning and mobilization during and beyond election cycles. We need to better understand this lesser-known aspect of tech firms’ political activities and the resulting implications for democracy in the digital age. In their book The Ordinal Society, Marion Fourcade and Kieran Healy explore how corporations with vested interests in tracking and measuring key aspects of our lives manipulate our norms about citizenship and social inclusion.

In addition, computational sociologists with technical expertise in AI continue to critically assess the models and tools tech firms promote for consumer and scholarly use while identifying crucial methodological and ethical concerns. A recent special issue of Sociological Methods & Research on generative AI, edited by Thomas Davidson and Daniel Karell, evaluates the possibilities and limitations of using generative AI models to analyze text and images and to augment survey data. AJ Alvero, Dustin Stoltz, Oscar Stuhler, and Marshall Taylor also recently fielded a survey to understand sociologists’ use and perceptions of generative AI. In this issue, Zhuofan Li’s article considers the role of computational social scientists in legitimating AI technologies in mass surveillance and other domains. Through this work, computational sociologists lead important conversations on whether and how AI should be integrated into science.

It can increasingly feel as if we are barreling toward a future in which, whether we like it or not, AI is ubiquitous and ungovernable. This is exactly the intended effect of what Princeton sociologist Ruha Benjamin calls “the gospel of AI evangelism,” a cultural and political framing rooted in ahistorical understandings of science and technology that presents AI as inevitable and indisputably good for humanity. “Trust us, because you have no choice,” the so-called gospel says. AI developers, and now the Trump administration, have advanced this framing of AI, effectively demobilizing everyday citizens from engaging in open debates and deliberate action on this important issue. But this framing, like AI technology itself, is just a tool and only as powerful as we allow it to be.