AAAI Workshop (2026)
Juhyun Lyu, Junghee Kim, Jinseok Yang
Abstract
Causal agents have recently emerged as promising tools for automating causal analysis and facilitating user collaboration. However, existing causal agent systems are often limited to a single task (e.g., causal discovery (CD) or causal effect estimation (CEE)), and they accept only numerical data as input, which prevents the integration of domain knowledge expressed in natural language. To overcome these limitations, we propose OrcheCause, a collaborative causal analysis agent. Specifically, OrcheCause is designed to orchestrate a sequence of interrelated causal tasks in response to user queries. Furthermore, OrcheCause supports diverse data types---numerical as well as textual---allowing the extraction of cause-effect (CE) pairs from raw text to improve causal discovery. OrcheCause also introduces a practical framework for hyperparameter optimization in causal discovery, employing BIC-based evaluation when ground truth is not available.