ABSTRACT Early AI policymaking in the United States appeared bipartisan, but subsequent developments raise the question of whether AI policy will become more polarized over time. To examine how partisanship takes root around novel policy issues, we perform a mixed‐methods study, analyzing survey data from 129 state legislators in 44 states and performing four case studies featuring legislative debates over enacted AI regulation in Idaho, Colorado, and Illinois. We articulate four potential partisan triggers that shape the emergence of issue polarization: (1) competing problem definitions, (2) preferred policy tools, (3) stakeholder participation, and (4) issue placement in policy subsystems, and we introduce the concept of subsystem shopping. Flexibility and compromise in subnational AI policymaking were initially common due to uncertain issue ownership over diverse problem definitions, the presence of disrupted industries, and deference to nonpartisan expert voices, which enabled subsystem shopping towards less controversial policymaking. However, bipartisan windows narrowed as AI became anchored in party‐owned domains, lobbying became more coordinated, and “soft” areas of initial legislative consensus hardened into disputes over more robust policy tools. These developments offer insights into how nascent policy domains can initially sustain bipartisan cooperation, and why it may erode unless actors work to preserve it.
Jacobson et al. (Wed,) studied this question.