Causal Modelling Agents - Augmenting Causal Discovery with Large Language Models

Explore a groundbreaking approach to causal discovery in this 30-minute talk by Ayodeji Ijishakin and Ahmed Abdulaal, Computer Science PhD students at University College London. Delve into the Causal Modelling Agent (CMA), an innovative framework that combines the metadata-based reasoning of Large Language Models (LLMs) with the data-driven modeling of Deep Structural Causal Models (DSCMs). Learn how this integration enhances scientific discovery by effectively merging cognitive operations with physical observations and experimentation. Discover the CMA's performance across various benchmarks and its real-world application in modeling the clinical and radiological phenotype of Alzheimer's Disease. Gain insights into how the CMA outperforms traditional data-driven and metadata-driven approaches to causal discovery, potentially revolutionizing our understanding of complex systems and diseases. This talk, part of the Data Science Festival Sandbox Sessions in 2023, offers a unique perspective on advancing causal modeling techniques and their practical implications in scientific research.

Causal Modelling Agents: Augmenting Causal Discovery with LLMs