Artificial Intelligence Policy under Uncertainty: An Analysis of U.S. State Legislation
Author(s)
Kim, Eunji Emily
Advisor(s)
Editor(s)
Collections
Supplementary to:
Permanent Link
Abstract
This dissertation examines how U.S. state legislatures are shaping the governance of artificial intelligence (AI) through legislative action, focusing on the emergence, alignment, and content of AI-related bills across 50 states from 2017 to 2025. In the absence of comprehensive federal regulation, states have increasingly taken on the role of early experimenters in regulating emerging technologies. This study investigates the institutional, political, and rhetorical dynamics underlying that subnational activity, offering a detailed analysis of legislative proposals not just as binary policy outcomes but as strategic texts shaped by procedural constraints, partisan coalitions, and policy uncertainty.
Drawing on an original dataset of over 600 full-text state-level AI bills, the dissertation employs a mixed-methods approach combining text similarity analysis, structural topic modeling (STM), and logistic regression. The analysis addresses three key research questions: (1) what patterns of textual similarity emerge in state-level AI legislation and what do they reveal about institutional routines and strategic reuse; (2) how is artificial intelligence framed in legislative language across partisan and institutional contexts; and (3) what political, institutional, and structural factors are associated with the passage of AI-related bills.
Text similarity analysis identifies four distinct patterns of legislative alignment: strategic reintroduction within the same state across sessions, symbolic duplication within a single session, rhetorical convergence across states, and limited cross-session, cross-state reuse suggesting cautious or selective diffusion. These findings underscore that legislative similarity often reflects institutional routines and symbolic signaling rather than direct diffusion. Structural topic modeling reveals ten dominant themes in AI bills, ranging from AI system regulation and data governance to education, healthcare, economic development, and content moderation. These topics vary by sponsor party: Democratic-sponsored bills focus on oversight and risk mitigation, while Republican-sponsored bills emphasize education curriculum, safety, and content regulation. Bipartisan bills tend to center on data privacy, health workforce, and administrative implementation, pointing to zones of cross-party consensus.
Finally, logistic regression models show that actor-level political alignment—not just partisan identity—plays a key role in bill success. Bills sponsored by legislators aligned with both the state legislature and governor are significantly more likely to pass, while broader state partisan indicators (e.g., Republican presidential vote share) are negatively associated with passage. Coalition size (number of co-sponsors), state-level technology employment, and institutional capacity are also positively correlated with enactment. Conversely, higher education levels and larger GDP are negatively associated with success, potentially due to heightened stakeholder complexity and procedural rigor.
The findings contribute to policy studies by showing that subnational AI governance is unfolding not through uniform diffusion or federal delegation, but through an adaptive, iterative process shaped by institutional constraints, rhetorical strategies, and actor alignment. The dissertation offers practical insights for policymakers by identifying where and how AI policy gains traction, and theoretical insights for scholars by demonstrating the value of text-as-data methods for understanding policymaking under uncertainty. As the regulatory landscape continues to evolve, this study highlights the central role of state legislatures in constructing the frameworks through which emerging technologies are defined, contested, and governed.
Sponsor
Date
2025-08-21
Extent
Resource Type
Text
Resource Subtype
Dissertation (PhD)