A Three-Layer Framework for AI in Scientific Discovery

The paper presents a three-layer framework for AI in scientific discovery, emphasizing the importance of model formation through qualitative reasoning as the second layer, which is underdeveloped compared to search and execution layers. It argues that effective scientific discovery requires not only retrieval and optimization but also the ability to identify inadequacies in existing models and conceptualize new frameworks. The paper illustrates this with case studies that demonstrate how recognizing structural limitations can lead to breakthroughs in understanding complex problems.