How Will AI Affect Social Inequality?
By Neelanjana Gautam
Over the past few years, artificial intelligence (AI) has emerged as an important topic of discussion among social scientists, particularly in terms of its usefulness as a research tool, ethical implications and potential effects on society. For the most part, these conversations have been speculative. Now, several years after the release of OpenAI’s GPT-3 model — and as generative AI models and tools have become more integrated in everyday life — researchers are moving beyond prediction to examine how AI is currently reshaping labor markets, health care and education. Because these areas are highly consequential for social and economic opportunity, researchers are eager to understand AI’s impact on social inequality.
A survey by the Pew Research Center in 2025 examined how Americans think of AI and its impacts on society. While Americans widely consider AI literacy essential, 61% say they want greater control over how AI is used in their lives. Notably, concerns vary across age groups, gender, race and socioeconomic status — suggesting that AI’s social impacts may not be experienced equally.
Recently, the UC Davis Center for Poverty and Inequality Research convened a crosssector conference to examine the current impacts of AI in our lives — and how researchers, policymakers and industry leaders can respond. The conference also included keynote talks by Genevieve Smith, founding director of the Responsible AI Initiative at the UC Berkeley Artificial Intelligence Research Lab and fellow at the Clayman Institute Stanford University, and Julianne McCall, chief executive officer of the California Council on Science and Technology, state government’s independent, nonpartisan advisor for science and technology policy. Conference organizers Jacob Hibel and Tina Law, faculty experts from the College of Letters and Science broke down the day’s discussion into several core themes.
Will AI increase or decrease social inequality?
Smith addressed how AI systems encode existing inequities. She argued how data inevitably reflects the social inequities of the world it is drawn from –– and the choices that shape AI systems, from what data to include to what outcomes to optimize and what counts as fair, are never neutral. By default, technologies tend to encode dominant norms and power hierarchies, making inequitable outcomes a predictable result. But design can be done differently: starting with people, centering communities, reframing toward equity, and building collaboratively –– these are approaches that offer real pathways to more inclusive systems.
As AI is rapidly integrated into daily life, its capacity to either deepen or reduce inequality depends not only on developers and institutions, but also on how communities choose to shape, question and govern these technologies. McCall emphasized the dual role AI can play in reducing or exacerbating social inequality, underscoring that AI is never neutral –– it reflects the data, design choices and institutional structures that produce it. These dynamics are already visible across labor markets, healthcare and education, making cross‑sector collaboration essential for understanding and addressing their impacts.
How can we measure whether AI is increasing or reducing inequality over time?
Hibel underscored the critical role of research in identifying and tracking these patterns. “We need baseline data — early studies on employment, career trajectories and wage patterns. These will be essential for tracking long‑term effects,” Hibel said. Law drew a pointed analogy, comparing the challenge to the development of the Gini coefficient, now a standard measure of economic inequality. “It took years of collaboration to develop and adopt that metric,” she said. “Measuring AI’s impact will require the same kind of interdisciplinary effort.”
What does research reveal about the role of AI in hiring and work?
Several studies at the conference examined how AI is affecting hiring and work. A research team led by the University of Oxford’s Chenxi Li examined gender bias in AI‑mediated hiring, with mixed results. Some HR evaluators rated female‑coded applicants who used AI as less competent, yet automated resume screening showed less dramatic gender disparities than anticipated. “The evidence for really dramatic gender bias didn’t seem to show up –– there’s been progress,” noted Hibel.
Another striking finding came from UCLA sociology doctoral student Maureen Cowhey, who, with UCLA Associate Professor of Sociology Natasha Quadlin, examined how gender shapes evaluations of AI use at work. Regardless of gender and job types, they found that disclosure itself carries a penalty –– workers who reveal AI use are consistently rated more negatively than those who do not, raising concerns about transparency norms and workplace stigma. Law argued this stigma may compound existing wage gaps. “It raises questions about transparency — who feels safe admitting they use AI, and who gets penalized for it,” Law explained. As AI becomes embedded in everyday work, the politics of disclosure may emerge as a new focal point of workplace inequality.
In addition, a study by Harvard sociology doctoral student Nayun Eom and Harvard Professor of Social Policy and Sociology Daniel Schneider found that workers feel AI systems exert significant control over their tasks while they have almost no say in how those systems are designed or implemented. “That lack of agency is a big deal,” said Law, “Workers want a voice in how AI is introduced, and they’re not getting it.”
That absence of input raises a harder question about who is most exposed and most vulnerable to displacement when things go wrong? The experts pointed out that current AI anxieties are part of a decades‑long automation trend, reframing AI not as a sudden rupture but as an acceleration of long‑term labor market restructuring.
What unique challenges does AI introduce in the healthcare setting?
Hibel pointed to a foundational problem that clinical research has historically relied on homogeneous study populations, leaving people of color underrepresented and women outnumbered by men in trials. AI tools trained on that data inherit those gaps. “If the AI tool doesn’t capture the full range of human experience, it will then limit the tool’s ability,” Hibel said –– and with federal funding cuts threatening the research pipelines that might correct those imbalances, the problem may deepen before it improves.
Law raised a parallel concern about access. AI could increase healthcare reach in underserved areas through chatbots and automated triage, but access to an AI tool is not the same as access to a physician. The risk is a two‑tier system: underserved communities receive AI-mediated care while others get human care.
Gul Seckin, associate professor of medical sociology at the University of North Texas, presented a study at the conference that examined how artificial intelligence in healthcare contributes to social inequality in the United States, introducing the idea of algorithmic stratification to show how AI used in diagnosis, treatment and risk prediction can deepen disparities in trust, access and outcomes. The findings highlight several policy challenges: lower trust in AI-driven healthcare among Black and Latino patients, digital and socioeconomic divides shaping who benefits from AI‑enabled care, age‑related discomfort with AI‑driven decisions and broad ambivalence about whether AI reduces or reinforces bias. Together, these patterns align with wider research showing that without strong oversight, AI is more likely to mirror existing structural inequities in the U.S. healthcare system than to challenge it.
Who should shape AI policy and toward what end?
In 2024, Law and CUNY Graduate Center sociologist Leslie McCall published Artificial Intelligence Policymaking: An Agenda for Sociological Research, which identified two emerging approaches to AI policymaking: a safety-based approach and an equity-based approach. The safety-based approach frames risk as existential threats to humans and national security, addressing them primarily through market-based solutions. The equity-based approach, by contrast, centers the affirmative protection of civil rights and liberties as the basis for building equitable AI systems. In a new article, Law identified the recent emergence of a third approach: a hegemony-based approach where “AI, then, is not just a technological means to a political end but also a source of power in and of itself.”
Law noted that although there are now several competing approaches to AI policymaking, “there’s still no federal regulation on AI.” She also pointed to a structural problem shaping how policymakers fill that gap by relying heavily on private AI companies (OpenAI, Anthropic, Google, Meta) for expertise. She called this “the definition of a conflict of interest,” and argued for broader representation of expertise in the legislative process, namely social scientists, computer scientists, public health researchers and other academics without direct financial stakes in the outcomes.
What role can public institutions, including universities, play in shaping AI governance?
California is in large part where AI is being developed, and the UC system is the largest public research institution in the country. Importantly, UC Davis is uniquely poised at the intersection of education, health, agriculture and food — sectors undergoing some of the most profound AI‑driven change. “We felt hosting the conference was important to bring researchers and policymakers together in Sacramento, right next to the Capitol, to talk about real data on AI’s consequences for labor markets, healthcare, education and more,” said Hibel. “To design the best, most impactful research on these topics, it is important for researchers to understand policymakers’ goals and concerns.”
McCall shared that policymakers are being asked to make high-stakes decisions about governance, safety and accountability — often without access to the full breadth of technical and social science expertise. For instance, with the Transparency in Frontier Artificial Intelligence Act (SB 53) having gone into effect in January, policymakers are currently navigating the practicalities of oversight. By bringing researchers, policymakers and practitioners together we are able to create an opportunity to align evidence with action at a moment when choices will shape equity and inclusion in health and AI for a long time. She highlighted how the CCST supports evidence‑based governance through initiatives such as the AI Academy, Science Advisor program, and Science & Tech Policy Fellowship, all designed to equip decision‑makers to govern AI responsibly.
“As social scientists, we work hard to conduct research that we hope will shape public policy for the greater good, but often those aspirations take a very long time to make it out of the pages of peer-reviewed journals, if at all. Through the ideas generated in the conference, we hope we can help improve both research and policy on AI,” signed off Law.
Media Contacts:
Aaron Cheline, UC Davis Office of Research, [email protected]
Neelanjana Gautam, UC Davis Office of Research, [email protected]
