Toward Agents of Intelligence: Bridging the AI Expertise Gap in Domain Sciences

Artificial intelligence (AI) has the potential to accelerate scientific discovery, yet many domain scientists face significant barriers when attempting to incorporate AI into their research. These challenges include a lack of formal training, limited access to computational resources, difficulty selecting appropriate models, and fragmented workflows. Through interviews with eleven postdoctoral researchers across diverse scientific fields at UC San Diego, we identify common pain points and recurring questions that arise throughout the scientific research pipeline. In response, we propose a conceptual framework, called Agents of Intelligence: modular, context-aware AI agents designed to assist researchers with tasks such as model recommendation, data preparation, infrastructure coordination, and interpretation of results. By embedding AI expertise into composable workflows, these agents aim to make AI more accessible, systematic, and transferable. We describe how this framework can be instantiated within the National Data Platform (NDP) and provide a testbed for real-world integration and evaluation.