---
title: Robust agents learn causal world models
url: https://www.ml-quant.com/papers/arxiv/2402.10877/
site: ML-Quant (https://www.ml-quant.com)
updated: 2026-09-26
license: Summaries CC BY 4.0; links go to the original sources
index: https://www.ml-quant.com/llms.txt
identifier: arXiv:2402.10877
source_url: https://arxiv.org/abs/2402.10877
featured: 2024-02-21
citations: 102
topic: Econometrics & Forecasting
---


# Robust agents learn causal world models

The research suggests that intelligent agents must learn an approximate causal model to generalize to new domains, impacting fields like transfer learning and causal inference.

- Source: https://arxiv.org/abs/2402.10877
- Identifier: arXiv:2402.10877
- Released: 2024-02-16
- First featured: Quant Letter No. 38 (2024-02-21): https://www.ml-quant.com/issues/2024-02-21/
- Citations (Semantic Scholar): 102
- Published in: International Conference on Learning Representations
- Topic: Econometrics & Forecasting

## Related

- [Expert-aided causal discovery of ancestral graphs](https://www.ml-quant.com/papers/arxiv/2309.12032/): A novel method for causal inference uses ancestral graph sampling and expert feedback to refine causal discovery, providing uncertainty estimates and accounting for unobserved confounders.
- [Exploring the heterogeneous impacts of Indonesia’s conditional cash transfer scheme (PKH) on maternal health care utilisation using instrumental causal forests](https://www.ml-quant.com/papers/arxiv/2501.12803/): The research uses machine learning to study the effects of Indonesia's conditional cash transfer scheme on maternal health care, finding significant variations based on supply-side factors and poverty indicators.
- [The impact of extracurricular education on socioeconomic mobility in Japan: an application of causal machine learning](https://www.ml-quant.com/papers/arxiv/2506.07421/): A machine learning study found that private tutoring in Japan can have positive socioeconomic impacts, but these are undermined by economic disparities among households.
- [Machine Learning for Causal Inference: Is a Nonlinear First Stage Really Forbidden in 2SLS?](https://www.ml-quant.com/papers/ssrn/4772060/): The paper shows that the bias in the two-stage least squares estimator can be split into an observable and unobservable bias, without needing to specify the first stage's functional form or validate the instrumental variable.
- [Double/Debiased ML in Stata](https://www.ml-quant.com/papers/repec/tsj-stataj-v-24-y-2024-i-1-p-3-45/): The article presents a package named ddml for double/debiased machine learning in Stata, supporting estimators of causal parameters for five econometric models and is compatible with various supervised machine learning programs.
- [Epistemic Limits of Empirical Finance: Causal Reductionism](https://www.ml-quant.com/papers/ssrn/4646664/): Causal Reductionism: The research criticizes the use of one-way causation in capital market studies, suggesting that current quantitative finance tools may only be suitable for after-the-fact causal inference.