Diffusion of AI Policies across State Governments in the United States

• In Event: Digital Transformation in Government:  Policy Diffusion, Administrative Capacity, and Discretion

Abstract

The main research question of this paper is: how does technology innovation adoption diffuse across states? We ask this question in the context of how the emerging AI policies are rapidly diffusing across state governments in the United States. This is a critical question at the intersection of policy diffusion and public management literature, as Frances Berry had more broadly posed in her noted 2023 John Gaus lecture, “By what mechanism do policies diffuse?” Traditional innovation diffusion approaches have focused on learning, competition, emulation and coercion, which take into account the individual or organizational factors that influence policy diffusion. From public management perspective, the Agent Network Diffusion Model focuses on importance of microlevel individual agents who carry innovation knowledge. Even within the realm of information science and technology policies, diffusion of innovation approaches like technology acceptance model (TAM) and theory of planned behavior focus on individual perceptual attitudes. This paper shifts the focus to the broader ecological mechanisms that shape innovation adoption in the states’ technology policies. The main argument for the focus on broader ecological context is that diffusion could be conditional on the environmental conditions for diffusion to take place. Specifically, in the context of technology policy diffusion, states that have advanced in adopting one technology could hypothetically be better poised to adopt allied technology innovations in another arena. We test this hypothesis with how adoption of a set of disparate tech innovations (cybersecurity, privacy, blockchain, drones) have influenced the adoption of emerging technology of Artificial Intelligence at the state level. Methodologically, we have coded the innovation adoption for aforementioned technology policies at state level. These policies have evolved over the last 25 years in the American state governments. AI is a more recent rapidly evolving technology and only a few states have adopted policies. AI policies are measured on a three point Likert scale (AI prohibitions, no policy, AI enablers). The data for all policies were collected from the National Conference of State Legislatures. Logistic Regression (a machine learning method which allows to determine the impact of multiple independent variables) is used to predict the likelihood of AI policy adoption. Preliminary analysis of the data shows that while competition and early adoption of allied technology are determinant factors for leading states (i.e. AI enablers), lack of tech innovation adoption of allied technologies explains “AI prohibitions” or “no policy”. The findings provide new insights into how states’ managerial capacity and policy maturity evolve with each technology and reinforce the adoption of new emerging technologies.

Authors

• Susannah Bruns Ali, Florida International University
  Non-Presenting Co-Author

• Sukumar Ganapati, Florida International University
  Presenting Author