"The Diagram is like Guardrails": Structuring GenAI-assisted Hypotheses Exploration with an Interactive Shared Representation

TL;DR

This work tackles conceptual hypothesis exploration in GenAI-assisted data analysis by proposing an ordered node-link diagram as a shared representation augmented with AI-generated visual and textual hints. Through a design probe with 22 participants, the study demonstrates that the diagram provides guardrails, maintains overviews, and supports non-linear, breadth–depth exploration, while visual hypotheses ground ideas in data and reduce cognitive load. The findings suggest that such interfaces can enhance human–AI collaboration in generating and refining hypotheses, and may generalize to guiding decision making and coordinating complex workflows. Overall, the work offers a concrete, multimodal interface pattern that blends structure with exploratory flexibility to advance high-level analytic tasks.

Abstract

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 investigates the design of an ordered node-link tree interface augmented with AI-generated information hints and visualizations, as a potential shared representation for hypothesis exploration. Through a design probe (n=22), participants generated diagrams averaging 21.82 hypotheses. Our findings showed that the node-link diagram acts as "guardrails" for hypothesis exploration, facilitating structured workflows, providing comprehensive overviews, and enabling efficient backtracking. The AI-generated information hints, particularly visualizations, aided users in transforming abstract ideas into data-backed concepts while reducing cognitive load. 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.

Figures (9)

• Figure 1: The wireframes depict the initial design for the formative study, showcasing synchronizing a chatbot with a node-link diagram based on weideleEmpiricalEvidenceConversational2024. In the upper figure, after the user uploads a dataset and inputs the analysis intent "inequality in education and income" in the chatbot, the system auto-generates five hypotheses (e.g., "correlation between income and education level") in the node-link diagram. The lower figure demonstrates interactivity: when a user clicks a hypothesis node, such as "correlation between income and education level," the diagram shows the hypothesis background and an interestingness rating, while simultaneously updating the chatbot with a message "let me check the hypothesis for correlation between income and education level."
• Figure 2: System architecture illustration of the AI-powered hypothesis exploration system.
• Figure 3: Walk-through of hypothesis exploration functions and P5's behaviors.
• Figure 4: Count of generated nodes by level, starting from level 2 given that level 0 always has 1 analysis intent node and level 1 always has 5 initial hypothesis nodes. As the depth increases, the count of generated nodes per level progressively decreases, reflecting a more focused analysis at deeper levels of the tree structure.
• Figure 5: Counts of total number of nodes explored (number of nodes clicked + number of new hypothesis branch generated, a new node will be displayed when a new branch is generated), number of nodes clicked, number of new hypothesis branch generated.