Structuring GenAI-assisted Hypotheses Exploration with an Interactive Shared Representation

Data analysis encompasses a spectrum of tasks, from high-level conceptual reasoning to lower-level execution. While AI-powered tools increasingly support execution tasks, there remains a need for intelligent assistance in conceptual tasks. This paper proposes an ordered node-link tree interface augmented with AI-generated information hints and visualizations, as a shared representation for hypothesis exploration. We further discuss how node-link diagrams can support both parallel exploration and iterative refinement in hypothesis formulation, potentially enhancing the breadth and depth of human-AI collaborative data analysis.