Abstract

Excerpted From: Spencer Overton, Overcoming Racial Harms to Democracy from Artificial Intelligence, 110 Iowa Law Review 805 (January, 2025) (348 Footnotes) (Full Document)

By the year 2050 there will be no majority ethnic group in the United States. Although the nation has made great progress toward becoming a racially inclusive democracy, key obstacles remain. As the nation has become more diverse, racial polarization and cultural anxiety have increased. Unfortunately, another major but to date underappreciated threat has emerged at the intersection of race and democracy: artificial intelligence (“AI”). This Article addresses that threat.

While synthetic video and audio (“deepfakes”) receive the bulk of popular attention, generative AI and related technologies are transforming all aspects of our electoral system and pose significant challenges to the future of a racially inclusive democracy. AI could empower politicians to iterate billions of datapoints to find hidden patterns and develop customized messages-- including those seemingly unrelated to voting--designed to covertly manipulate the cultural and political identity of individual voters of color or reduce voter turnout in communities of color. AI will empower hackers to deploy cyberattacks and swarms of frivolous open-record requests targeted to incapacitate local election offices that serve large populations of voters of color. Local law enforcement's warrantless deployment of AI to surveil social media and mobile phone data will continue to chill demonstrations for police reform by Black Lives Matter activists.

Even absent intentional discrimination, bias and other flaws embedded in AI datasets used to create content, moderate content, detect deepfakes, maintain voter rolls, verify mail-in ballot signatures, provide adequate language assistance in voting, and perform a host of other tasks could replicate disadvantages and fortify racial and cultural hierarchy in elections and policymaking well into the future.

The challenges arise both from the bias of AI datasets and the limitations in the design of AI to adequately serve all in a pluralistic, diverse society. Because AI models typically address diversity in mathematical ways by defaulting to averages or dominant patterns, they often fail to recognize different perspectives and needs--including those of communities of color in the context of voting and democracy. The current design of many AI models effectively automates homogenization and reifies racial stereotypes, which is inconsistent with respect for and the coexistence of both factual information and diverse interests and viewpoints in a liberal democracy. Technologists have only recently started to acknowledge these design flaws.

The harms of AI to deliberation, participation, and representation are often more difficult to detect and regulate when they disproportionately affect groups with less social, political, and economic influence--including many communities of color. The homogeneity of those who develop the tools and govern tech companies and the failure to prioritize the unique ways in which many communities of color experience AI technologies only compound the antidemocratic nature of the harms.

This Article does not take the position that AI inevitably spells the end of democracy. Rather than ignore AI's racial and cultural implications for democracy, we must anticipate and mitigate the threats it poses. A central goal of AI and the law could be--and should be--to facilitate our transition to a well-functioning, inclusive, pluralistic democracy--one that respects both identity and individual autonomy and enables cross-group engagement, coalition building, and collective well-being. Artificial intelligence and future laws that regulate it should encourage incumbent politicians to respond to demographic change through fair representation rather than entrenching power by stoking racial animosity and manipulating racial turnout.

Even though race is the most significant demographic factor that shapes voting patterns in the United States, this Article is the first to comprehensively examine the racial challenges of AI for democracy. Other scholars have started to analyze the implications of AI for democratic deliberation broadly and the dangers to elections of AI-enabled deepfakes. Recognizing the growing significance of AI, demographic change, cultural anxiety, antidemocratic sentiment, and a U.S. Supreme Court increasingly hostile to traditional voting rights protections, this Article provides an essential first step in developing legal structures and private-sector practices to secure representative democracy for future generations in the United States.

Technology is evolving rapidly, and we may not be able to completely eliminate every challenge. But by acting now, advocates, policymakers, and the private sector can contain the most severe harms from these new tools. This Article introduces a framework to approach AI in the context of our increasingly diverse democracy, catalogs key challenges posed by AI, and crafts principles to mitigate these challenges. Future scholarship will isolate and analyze individual problems and propose detailed legal solutions. Recognizing that AI can also be used to mobilize underrepresented voters, expand the capacity of underfunded community organizations and candidates of color, and facilitate cross-racial coalition building, future scholarship will also explore how law can incentivize the development and use of AI applications to enhance racially inclusive democracy.

This Article does not grapple with the currently unknown challenges from future generations of artificial intelligence that could become independent of human control. Their implications for multiracial democracy are speculative at best. At this moment, the priority should be identifying, understanding, and taking action to contain the challenges to multiracial democracy that are most likely to arise from existing and emerging AI technologies.

Part I of this Article describes how the United States has made great strides over the past six decades in becoming a racially inclusive democracy. Still, significant challenges remain, including increasing racial polarization, cultural anxiety, and antidemocratic attitudes. Part II provides a brief overview of artificial intelligence and emerging applications of the technology in the context of democracy. Part III maps many of the ways in which AI may be used to cause racial harms to democracy. Recognizing that existing legal protections are unlikely to address these challenges, Part IV proposes four key principles for those who develop, deploy, and regulate AI to tackle the challenges and support an inclusive democracy that reflects the will of our increasingly diverse population.

[. . .]

AI technologies pose unique opportunities and challenges for the future of racially inclusive democracy in the United States. As our nation becomes more diverse, an explicit goal of AI and the law should be to facilitate our transition to a well-functioning, pluralistic democracy that respects both identity and individual autonomy, facilitates cross-group coalition building, and incentivizes political operatives to respond to demographic change through inclusive representation. Those who develop and deploy AI technologies should mitigate the risks of racial harms to democracy from their applications, and policymakers should ensure that AI regulation contemplates and prevents racial harms to democracy.

Patricia Roberts Harris Research Professor of Law and Faculty Director of the GW Law School's Multiracial Democracy Project, George Washington University.