Can LLMs boost urban causal inference?
Reimagining Urban Science: Scaling Causal Inference with Large Language Models
April 18, 2025
https://arxiv.org/pdf/2504.12345This paper proposes AutoUrbanCI, an LLM-powered framework for automating urban causal inference research. It addresses current research limitations like geographical bias, over-reliance on structured data, and methodological homogeneity by using modular agents specialized for hypothesis generation, data engineering, causal experimentation, and policy insights generation.
Key points relevant to LLM-based multi-agent systems:
- Modular Agents: AutoUrbanCI leverages specialized LLM agents, each with multiple operational modes to handle different research scenarios and data types. This enables flexibility and adaptability in causal inference workflows.
- Multimodal Data Integration: AutoUrbanCI is designed to integrate and process multimodal urban data (text, images, time series, etc.), addressing a current limitation in urban research.
- Automation of Causal Inference: The framework automates key steps in causal inference, from hypothesis generation to experimental execution and policy narrative generation, reducing manual effort and enhancing scalability.
- Human-AI Collaboration: AutoUrbanCI emphasizes a shift from AI as a tool to AI as a collaborator, with humans retaining oversight in defining research goals, validating outputs, and ensuring policy relevance.
- Multi-agent coordination: The paper highlights the challenge of maintaining context and consistency in multi-agent interaction and the need for robust mechanisms to ensure validity.
- Evaluation: The paper proposes a multi-dimensional evaluation framework considering methodological rigor, data quality, and ethical implications of AI-generated research, moving beyond typical LLM evaluations.