Civic Capacity or Technical Competence: Can AI Literacy Democratize AI Agenda-Setting?

• In Event: Participation and Procedural Justice in AI Governance

Abstract

Public participation is widely deemed crucial for shaping AI governance, and central to the democratic process of policymaking. However, previous studies on public participation in AI policy suggest that the agenda-setting process remains expert dominated, and that public understanding of AI remains relatively nascent. In response, large technology companies, universities, civil society organizations, and even governments are implementing large-scale AI literacy interventions with stated goals of building public understanding, trust in AI, and civic capacity. While many programs are primarily designed to build technical competence, drive workforce readiness, and promote AI adoption, the policy rationales often explicitly invokes democratic participation in both government and industry decision-making—advocating literacy as a precursor for meaningful public engagement by shaping who can define problems, select solutions, and signal priorities in AI governance, a domain where the policy agenda is still being formed. These AI literacy programs, fairly unprecedented in their scale and rapid adoption, thus constitute an innovative development in this nascent policy domain, potentially restructuring participation by affecting both who engages in AI development and governance, and how this engagement unfolds.

Yet, the instrumental logic underpinning this assumed link between literacy and democratic participation remains untested: do these programs actually promote public participation in AI policy agenda-setting, or do they primarily serve as tools of economic logic such as enabling workforce readiness and promoting consumer adoption? We empirically evaluate the instrumental logics of this process innovation through a survey experiment that simulates the core mechanism: exposure to AI literacy content. Drawing from Long and Magerko’s (2020) seventeen competencies of AI literacy alongside additional literacy frameworks and prominent training materials, we distinguish two forms of AI literacy—technical (design and development of AI) and socio-ethical (implications of AI on individuals and society)—which reflect distinct design logics within these programs. Utilizing a representative sample of 2,000 American adults recruited via Prolific, the study randomly assigns participants to specific competencies administered through training videos across three treatment groups—technical, socio-ethical, and a third group comprising both technical and socio-ethical competencies—alongside two control groups (pure control and non-AI technology training). Using an instrumental variable and quasi-pre-post regression design, we examine whether technical and socio-technical competencies translate into political participation, evaluating two outcome types: macro-political participation (electoral, civic, and administrative participation and micro-political behavior in AI policy. We additionally examine how these outcomes are mediated through AI self-efficacy, trust in AI, and perceived egotropic and sociotropic relevance.

Our findings suggest that these programs significantly enhance technical competencies while enhancing socio-ethical competencies to a lesser degree. However, neither type of literacy meaningfully improves public macro- or micro-political participation. Instead, we find that extant limitations of internal and external political self-efficacy dominate public willingness to participate in agenda-setting. Large-scale AI literacy programs—in practice typically focused even more on technical knowledge than tested here—may therefore not be sufficient to overcome political capacity deficits and democratize the AI policy agenda.

Author

• Indira Dhananjaya Patil, Purdue University
  Presenting Author