Welfare and AI: An experiment in co-learning for local governments

The digital era has profoundly transformed social policies, pushing towards a new welfare model. This approach as pires to innovations in offered services and influences policy management (Campedelli and Vesan, 2023). In this context, artificial intelligence (AI) is viewed as a solution to reduce repetitive work and provide new tools for policy makers (Toll et al., 2019). However, rapid advancements in AI have raised concerns about the possibility of crucial de cisions for social security escaping direct human control. To address and discuss this tension between the benefits and threats of generative AI in welfare, we present preliminary results from a research and intervention program called Govern-AI, focused on the regional welfare of Campania (Italy). The main goal of Govern-AI is to understand the inter action between institutional actors and new technologies through the introduction of an AI chatbot for local policy making. The construction of the chatbot follows a partici patory research approach, as data provided by participat ing institutions not only constitute the chatbot’s learning foundation but also significant empirical evidence. Before construction, preliminary focus groups identified relevant aspects of institutional imagination regarding AI, generat ing opinions and representations that translate into actual “policy frames” (Rein e Schön, 1996). Govern-AI not only provides a practical tool for decision assistance but also acts as a catalyst for co-learning between humans and the AI system, relying on information from human actors while providing outputs used by them