Thread by @h_nilforoshan on Thread Reader App
Short summary
A new ICML paper argues that shifting fair machine learning toward causal reasoning actually harms policy objectives like diversity and equity, contrary to what researchers expected. The authors show that causal methods can produce worse outcomes than simpler approaches like removing sensitive features entirely.
- Causal fairness frameworks, despite seeming like improvements over traditional methods, can actively damage diversity and equity efforts in practice
- Removing sensitive attributes (the traditional "blinding" approach) sometimes outperforms causal reasoning for achieving policy goals
- The counterintuitive finding challenges the current direction of fair ML research, which has largely embraced causal thinking as more rigorous and principled
Full summary
A research team won an outstanding paper award at ICML 2022 for work questioning whether causal reasoning actually improves fairness in machine learning. The paper, authored by researchers including Sonja Gershman, Ravi Shroff, and Sharad Gershoff, challenges a major shift in the field toward causal approaches to fairness. Their counterintuitive finding: moving toward causal methods can actually harm common policy goals like diversity and equity efforts.
The researchers highlight a specific contrast. Traditional fairness approaches in ML often use blinding—simply removing demographic features like race or gender from decision-making algorithms. This seems crude, but causal methods, while theoretically more sophisticated, can produce worse real-world outcomes for the very groups these policies aim to protect. The paper suggests that what looks like methodological progress on paper may backfire when applied to actual policy problems.
The work matters because it exposes a gap between technical sophistication and practical results. Causal reasoning sounds better than crude feature removal, yet the research indicates it can undermine diversity and equity initiatives. This forces a reckoning in the fair ML community about whether adopting more complex causal frameworks actually serves the policy objectives that motivated fairness research in the first place.