Brief

A new paper proposes a three-layer framework for AI in scientific discovery, emphasizing model formation through qualitative reasoning as the most critical and underdeveloped aspect. The framework distinguishes between AI's search capabilities (Layer 1), its ability to form new conceptual models (Layer 2), and its execution functions (Layer 3). The authors illustrate Layer 2 reasoning with case studies, including OpenAI's 2026 disproof of the Erdos unit distance conjecture, highlighting the recognition of inadequate frameworks and the identification of missing conceptual objects. AI