AI as Epistemic Mediator: Promises, Limits, and Critical Conditions for Knowledge Transfer Across Publics

Artificial intelligence is increasingly promoted as a tool for making knowledge more accessible and transferable across social and institutional boundaries. The question arising is whether generative AI (GenAI) can be harnessed to function as an epistemic mediator, i.e., a technology that shapes how knowledge moves between different stakeholders for different purposes. This panel contemplates GenAI as an epistemic mediator from four complementary perspectives. It balances four papers, two of which offer a critical look at GenAI’s limits and two proposing constructive ways to use it as a mediator between diverse socio-cultural groups. The first paper addresses the interface between scientists and general publics. Through an online blinded experiment, it provides critical empirical evidence regarding GenAI’s effectiveness in communicating science to publics. Via linear mixed-effects models, the findings demonstrate that GenAI-supported texts had no overall benefit for perceived content quality and engagement. Furthermore, GenAI-supported texts received lower ratings from audiences highly interested in science. The second paper explores the interface between experts and educators. This study uses a multi-layered temporal discourse analysis of an AI-supported multi-expertise workshop designed to connect teachers’ stories with pedagogical design principles . The study shows that AI can act as an epistemic mediator, fostering a bidirectional synthesis of theory-based knowledge and practice-based insights. The third paper examines the interface between regulators and educators. Using a GenAI chatbot that guides teachers through reflective conversations, combined with content analysis and policymakers' interviews, this work in progress will propose a model for integrating GenAI into policy development. The fourth paper examines interactions between laypeople and expert knowledge in the context of everyday health-related decision-making. Using a 2×2 design in an online survey, the findings show that exposure to conflicting GenAI responses actually supports informed decision-making, especially among laypeople who were unable to make informed decisions on their own. Taken together, these studies offer a nuanced perspective on AI-mediated knowledge transfer. They reveal that GenAI does not automatically improve communication or understanding. Instead, its epistemic value depends on the social contexts, interactional structures, and usage strategies within which GenAI operates. By examining both the promises and limitations of AI as an epistemic mediator, the panel contributes to broader discussions about knowledge, expertise, and communication in an increasingly AI-mediated society.