CausalChat: Interactive Causal Model Development and Refinement Using Large Language Model

Abstract: Causal networks are widely used in many fields to model the complex relationships between variables. A recent approach has sought to construct causal networks by leveraging the wisdom of crowds through the collective participation of humans. While this can yield detailed causal networks that model the underlying phenomena quite well, it requires a large number of individuals with domain understanding. We adopt a different approach: leveraging the causal knowledge that large language models, such as OpenAI’s GPT-4, have learned by ingesting massive amounts of literature. Within a dedicated visual analytics interface, called CausalChat, users explore single variables or variable pairs recursively to identify causal relations, latent variables, confounders, and mediators, constructing detailed causal networks through conversation. Each probing interaction is translated into a tailored GPT-4 prompt and the response is conveyed through visual representations which are linked to the generated text for explanations. We demonstrate the functionality of CausalChat across diverse data contexts and conduct user studies involving both domain experts and laypersons.

Teaser: The CausalChat Dashboard analyzing the AutoMPG dataset. It is a simple and well-tested dataset but CausalChat enabled a significant update from the 1980s cars to include modern technology -- see the newly added variables connected as dotted lines, serving as causal colliders (or alternatives).

The dashboard includes (A) the Control Panel allowing users to specify fundamental parameters and manage the model tree for variations of the causal model, (B) the Causal Graph Panel for interactive refinement of the graphical model, (C) the Causal Debate Chart for resolving causal directions and inclinations, (D) the Causal Justification Panel offering a rationale for each hypothetical causal statement, covering latent factors, potential confounders, and mediators, and (E) the Causal Relation Environment Chart suggesting potential latent variables, confounders, and mediators for specific causal relations and variables. In this figure the confounder/mediator chart is shown.

Video: Watch it to get a quick overview how a user would construct comprehensive causal networks with our system:

Paper: Y. Zhang, A. Kota, E. Papenhausen, K. Mueller, “CausalChat: Interactive Causal Model Development and Refinement Using Large Language Models,” IEEE Transactions on Visualization and Computer Graphics, 2025. PDF

Funding: CDC Cooperative Agreement 1 NH25PS005202-01-00, American Public Health Association (APHA) award # 2023-0004, and NSF grant CNS 1900706