Bridging AI Research and Policy: A Co-Production Framework and the Case for Policy Transfer Offices

Abstract

Artificial intelligence systems now shape governance, labor, healthcare, and public life at scale, yet there remains a persistent gap between technical knowledge and the policy processes meant to govern AI's use. We argue that this gap is not a communication failure but an institutional design problem: the absence of sustained, equitable mechanisms that make researcher-policymaker collaboration feasible. Drawing on theories of science-policy co-production, we present empirical evidence from a survey of AI researchers (n=44) that identifies the dominant structural barriers as time costs (66%), pathway opacity (57%), and weak institutional support, not lack of access to policymakers. We introduce a co-production framework centered on the novel proposal for Policy Transfer Offices (PTOs): institutional structures, analogous to technology transfer offices, that broker relationships, provide translation templates, and legitimize policy work within academic evaluation systems. We further expand on the proposal, in the dimensions of placement, governance, funding, and failure modes, and ground it in two in-depth case studies (algorithmic collusion and AI in education) that illustrate when technical expertise fails to reach governance in time, and what institutionalized co-production enables. We further analyze equity in participation, showing how current ad hoc models concentrate policy influence among senior researchers at elite institutions, and propose reforms for broadening access. The paper offers empirical evidence, an actionable institutional mechanism, and equity-centered design principles to align AI research with democratic governance.