Causal Bias Detection in Generative Artifical Intelligence

2026-05-11

AI Summary

This work addresses the challenge that generative artificial intelligence may implicitly encode causal mechanisms that amplify demographic biases related to race, gender, and other attributes, rendering existing causal fairness methods inadequate. We present the first formalization of causal fairness in generative AI and introduce a unified theoretical framework that enables fine-grained quantification of bias through causal decomposition along distinct pathways and mechanism-replacement dimensions. By integrating structural causal models, causal inference, and generative modeling, we establish identifiability conditions and develop efficient estimators for computing causal fairness metrics. Experiments on multiple datasets with large language models demonstrate that our approach effectively detects and quantifies racial and gender biases, confirming its validity and practical utility.

Abstract

Automated systems built on artificial intelligence (AI) are increasingly deployed across high-stakes domains, raising critical concerns about fairness and the perpetuation of demographic disparities that exist in the world. In this context, causal inference provides a principled framework for reasoning about fairness, as it links observed disparities to underlying mechanisms and aligns naturally with human intuition and legal notions of discrimination. Prior work on causal fairness primarily focuses on the standard machine learning setting, where a decision-maker constructs a single predictive mechanism $f_{\widehat Y}$ for an outcome variable $Y$, while inheriting the causal mechanisms of all other covariates from the real world. The generative AI setting, however, is markedly more complex: generative models can sample from arbitrary conditionals over any set of variables, implicitly constructing their own beliefs about all causal mechanisms rather than learning a single predictive function. This fundamental difference requires new developments in causal fairness methodology. We formalize the problem of causal fairness in generative AI and unify it with the standard ML setting under a common theoretical framework. We then derive new causal decomposition results that enable granular quantification of fairness impacts along both (a) different causal pathways and (b) the replacement of real-world mechanisms by the generative model's mechanisms. We establish identification conditions and introduce efficient estimators for causal quantities of interest, and demonstrate the value of our methodology by analyzing race and gender bias in large language models across different datasets.

Problem

• causal fairness
• generative AI
• causal bias
• demographic disparities
• large language models

Innovation

• causal fairness
• generative AI
• causal decomposition
• bias detection
• large language models