AIGC and Causal Inference: Mutual Empowerment and Applications with YLearn

The presentation introduces the theme "AIGC and Causal Inference Mutual Empowerment" and outlines four main parts: (1) using AIGC for structured data synthesis, (2) how synthetic data aids causal inference, (3) applying causal inference to improve Agent‑Based Modeling (ABM), and (4) an overview of the YLearn causal learning platform.

AIGC, driven by large‑scale language models, excels at generating unstructured content but can also be extended to produce structured data through synthetic data generation, which in turn provides high‑quality inputs for causal discovery and effect estimation.

Industry reports predict synthetic data will surpass real data by 2030; synthetic data addresses common enterprise challenges such as cost‑limited data collection, privacy restrictions, and scarcity of labeled samples. Two main synthetic‑data generation approaches are data‑driven (GANs, VAEs, Bayesian networks) and process‑driven (ABM, discrete‑event simulation, Monte‑Carlo). ABM is highlighted as an effective bridge because it can simulate autonomous agents, generate rich feature sets, and produce counterfactual scenarios.

In causal inference, tasks such as causal discovery, effect estimation, and counterfactual prediction benefit from synthetic data. ABM can generate counterfactual samples, provide complete causal graphs, and capture full feature information, enabling more accurate evaluation metrics (e.g., MSE, RMSE) beyond traditional AUUC or Qini.

The YLearn framework is introduced as a one‑stop solution for causal learning, covering causal discovery, graph construction, effect estimation (Meta‑Learner, Causal Forest, etc.), policy learning, interpretation, and counterfactual prediction, all through a unified API.

Overall, the talk demonstrates that AIGC‑generated structured data and ABM‑based simulations can substantially enrich causal inference research, while causal methods can also enhance AIGC pipelines, creating a virtuous cycle for AI development.