Why Social Science Matters for AI Governance

Artificial intelligence (AI) now shapes all or most parts of human life. AI is involved in decisions about employment, healthcare, education, public benefits, policing, and democratic participation. Social media platforms, powered by AI, shape public discourse, amplify misinformation, and deepen political and social divides. Access to AI and smart devices is key to being able to participate in society. Yet, most AI systems are built without meaningful input from social scientists, human rights experts, or communities who are affected by AI harms such as bias, exclusion, and surveillance. 

Policymakers face urgent questions such as: How do we ensure AI is fair, transparent, and accountable? How do we protect democratic rights, civil liberties, and public trust? Social science and human rights research provide a critical, and currently missing, foundation for responsible AI governance. Solutions depend on understanding how AI systems shape people’s lives, rights, and opportunities. Social science evidence shows who is harmed and left out, and which policies actually reduce bias and improve accountability. Policymakers should strengthen transparency and auditing for AI tools, allocate funding for community-led testing and review of these technologies, and ensure that AI regulations align with existing civil rights laws.

Research Shows that AI Policies Are Lacking

Social scientists, including qualitative researchers, political scientists, human rights scholars, and digital researchers, study how people communicate, how institutions function, and how power operates. With regards to technology, they have examined how AI shapes society, what political and social systems are strengthened or weakened, and how AI influences accountability, democracy, and governance. These insights reveal several core issues:

• AI frequently misreads and represses marginalized communities. AI systems decide, through their training data, whose voices and vocabularies matter. Research shows that emotion-recognition, language processing, and biometric tools perform worse for Black, Indigenous, and Latinx communities, disabled people, gender-diverse people, and speakers of languages other than English. When algorithms are used for governance decisions, they reproduce structural inequities in housing allocation, policing decisions, and public transport that are baked into the historical data that trained the AI. Hiring algorithms have been shown to downgrade applicants based on their race, sex, age and residential postal codes. Facial recognition systems misidentify darker-skinned faces at far higher rates. Risk assessment tools in criminal justice label marginalized defendants as higher risk based on proxies for poverty and race. These errors can lead to misdiagnoses, unfair hiring decisions, biased policing, and misinterpretation of testimony by affected communities.

• Many harms arise from context, not code. AI systems often fail because they ignore social context. For example, language differences are mistaken for “negative affect”, disability-related facial or vocal patterns treated as “errors”, cultural expressions are interpreted through biased training data. Biases are built into the system, disproportionately affecting minority populations, thereby deepening existing divides and leading to further marginalization in the digital sphere. Social science research explains these contexts, helping policymakers and technologists understand why certain groups are disproportionately harmed.
• Communities want control over how AI affects them. Human rights and participatory research show that people most affected by algorithmic decisions want a role in how AI systems are designed, evaluated, and governed. When communities can shape the categories, data, and assumptions that systems rely on, the result is not only more accurate technology but also processes that reflect their lived realities. Community involvement strengthens trust, ensures that systems account for cultural and linguistic diversity, and reduces the risk of harm or misinterpretation. It also shifts power: rather than being acted upon by AI systems, communities become active partners in their creation and oversight.
• Effective AI governance requires understanding power. Political science and policy research demonstrate that AI systems redistribute decision-making power between governments, companies, and citizens. Without oversight, automated systems can centralize authority in private companies, reduce the discretion of frontline workers, and weaken democratic mechanisms that traditionally allow citizens to challenge or appeal decisions. Without strong oversight, AI can expand opaque decision-making, shift power to tech corporations, and create new forms of dependency on proprietary technologies. Understanding these power shifts is essential for designing regulations and accountability structures that protect democratic participation, civil liberties, and the public’s ability to understand and, if applicable, contest algorithmic decisions.

How to Build Responsible AI Systems

Social science research points to practical steps policymakers can take to build responsible, equitable AI systems:

• Strengthen transparency and auditing. Require clear documentation of data sources, demographic performance gaps, and known limitations. Fund independent audits by third party and university researchers, not only private vendors. Require social impact assessments before federal or state agencies adopt AI.
• Support community-based impact evaluation. Allocate federal and state funding for community participation in AI testing and review. Fund community-engaged AI research that includes marginalized groups as co-designers, not just data sources. Require agencies to partner with affected communities before deploying algorithmic tools.
• Prioritize civil and human rights protections. Ensure AI regulation aligns with the Civil Rights Act, Americans with Disabilities Act, voting rights protections, and international human rights norms. Mandate accessibility standards and disability-inclusive design across all public-sector AI tools. Require any AI used in hiring, healthcare, policing, or public benefits to show how well it works for different groups of people to ensure fair treatment. Establish clear guardrails ensuring that AI systems cannot replace human decision-making where rights may be violated.
• Invest in public-interest research infrastructure. Support interdisciplinary research centers that unite social science, human rights, and computer science. Expand programs at Historically Black Colleges and Universities, Tribal Colleges, regional public universities, and community colleges to increase AI literacy and technical capacity. Create state and federal advisory boards that include social scientists, human rights experts, disability advocates, and affected community members.

AI governance is not only a technical challenge, it is a societal and democratic one. Without insights from social science, human rights, and public policy research, policymakers risk adopting tools that reinforce inequality and weaken public trust. With these perspectives at the table, the United States can build AI systems that are transparent, accountable, and aligned with the rights and dignity of all people.