Fifty States, Fifty Algorithms: State-Level Variation in AI Governance and Its Implications for Public Trust

• In Event: AI Governance and Regulation: From State to Federal

Abstract

The United States does not currently have one comprehensive federal law governing the use of artificial intelligence. As a result, individual states have begun developing their own AI laws and policies which differ considerably. Some states such as Colorado and California have established relatively strong requirements concerning discrimination, transparency, risk assessment, human oversight and accountability. Other states have adopted narrower rules, focused mainly on particular technologies or policy areas, while many states have no AI-specific laws in several important sectors.
This paper examines how these differences create uneven protections for people across the United States. Using a theory-informed comparative analysis of state AI legislation, the study draws on a 50-state legislative tracking dataset covering AI and algorithm-related measures in employment, healthcare, education, credit and lending, criminal justice, government use, and elections. Each state’s approach is examined according to whether its laws provide safeguards such as notice that AI is being used, transparency, protection from discrimination, risk assessment, human review, accountability, and opportunities to question or appeal an AI assisted decision. The analysis also distinguishes between elected laws, proposed legislation, government policies, local rules, and states where no AI specific measure was identified.
In this paper, a direct measure (survey) of whether policy fragmentation has caused public trust to increase or decrease is not used. Instead, it draws on theories of procedural justice and institutional trust to examine the likely implications of unequal protections. These theories suggest that people are more likely to regard government institutions as fair and legitimate when decisions are made through consistent, transparent and accountable procedures. When people facing similar AI-assisted decisions receive different protection simply because they live in different states the resulting inequality may raise concerns about fairness and equal treatment.
The preliminary analysis shows that the United States is developing multiple state systems of AI governance rather than one consistent national framework. Residents of some states receive stronger, procedural protections against algorithmic harm, while residents of other states must rely mainly on general civil-rights, privacy, consumer protection, or administrative laws. This uneven system does not by itself prove the public trust has declined, but it does however, create conditions that theories of procedural justice associate with concerns about fairness, and institutional legitimacy.
The paper contributes to debate about American federalism, AI, accountability and equitable public policy. It considers whether a state-led regulation can provide adequate and reasonably consistent protection while allowing policy experimentation and technological innovation in the absence of comprehensive federal legislation.

Author

• Sonia Junior Arthur, University of Massachusetts at Boston
  Presenting Author