l3s/causality-for-trustworthy-ai

:books: Curated list for Causality and AI

A Review of the Role of Causality in Developing Trustworthy AI Systems – Datasets and Packages

Special thanks to our colleague Alina Nekrasova for designing this great figure!

This repository is a curated list of datasets used for recent Causal Machine Learning (ML) publications we covered in our survey. It also includes an overview of useful causal and non-causal tools and packages to assess different characteristics of ML models (e.g., robustness or fairness) and for use in healthcare. We also provide slides that summarize the use of Causality for each discussed aspect of Trustworthy AI (see here for more information).

Feedback is welcome, so don't hesitate to reach out if you have any questions or comments. Simply send an e-mail to [email protected]

If you find this overview helpful, please cite our corresponding survey:

@misc{Ganguly_Causality_2023,
      author = {Ganguly, Niloy and Fazlija, Dren and Badar, Maryam and Fisichella, Marco and Sikdar, Sandipan and Schrader, Johanna and Wallat, Jonas and Rudra, Koustav and Koubarakis, Manolis and Patro, Gourab K. and Zai El Amri, Wadhah and Nejdl, Wolfgang},
      title = {A Review of the Role of Causality in Developing Trustworthy AI Systems},
      month = feb,
      year = 2023,
      publisher = {arXiv},
      doi = {10.48550/ARXIV.2302.06975},
      url = {https://arxiv.org/abs/2302.06975},
}

Introduction

As a result of our literature review on causality-based solutions for Trustworthy AI, a need for an extensive overview of relevant datasets and packages was observed. To make causal machine learning (ML) more accessible and to facilitate comparisons to non-causal methods, we created a curated list of datasets used for recent Causal ML publications. This appendix also includes an overview of useful causal and non-causal tools and packages to assess different trustworthy aspects of ML models (interpretability, fairness, robustness, privacy, safety and accountability). We also provide a similar overview for the healthcare domain. Each aspect has its dedicated section that is structured as follows:

1. An overview of publicly available real-world datasets used in cited publications of our main survey

2. Some benchmarks and packages for Causal Machine Learning that researchers could utilize

3. A number of well-established tools, that allow for a better comparison to non-causal machine learning

We want to clarify that this curated list does not (and cannot) aim for completeness. Instead, we want to provide researchers interested in working on a selection of aspects of Trustworthy AI with a concise overview of exciting avenues for experimenting with causal machine learning. We highly encourage readers interested to seek additional related reading material, such as Chapter 9 of [2] or the two Github repositories for datasets¹ and algorithms² resulting from [1].

Aspects of Trustworthy AI and Application Domain

Interpretability

Fairness

Robustness

Privacy

Auditing (Safety and Accountability)

Healthcare

References

[1] Guo, Ruocheng, Lu Cheng, Jundong Li, P Richard Hahn, and Huan Liu. 2020. “A Survey of Learning Causality with Data: Problems and Methods.” ACM Computing Surveys (CSUR) 53 (4): 1–37.

[2] Kaddour, Jean, Aengus Lynch, Qi Liu, Matt J Kusner, and Ricardo Silva. 2022. “Causal Machine Learning: A Survey and Open Problems.” arXiv Preprint arXiv:2206.15475.

Acknowledgements

This work has received funding from the European Union's Horizon 2020 research and innovation program under Marie Sklodowska-Curie Action "NoBIAS - Artificial Intelligence without Bias" (grant agreement number 860630) and Network of Excellence "TAILOR - A Network for Trustworthy Artificial Intelligence" (grant agreement number 952215), the Lower Saxony Ministry of Science and Culture under grant number ZN3492 within the Lower Saxony "Vorab "of the Volkswagen Foundation and supported by the Center for Digital Innovations (ZDIN), and the Federal Ministry of Education and Research (BMBF), Germany under the project "LeibnizKILabor" with grant No. 01DD20003.

NoBIAS	TAILOR	Leibniz AILab

Future Lab "Work and Society"	L3S Research Center

Footnotes

1. https://github.com/rguo12/awesome-causality-data ↩

2. https://github.com/rguo12/awesome-causality-algorithms ↩