AI systems increasingly shape consequential decisions in employment, lending, healthcare, and public services, making fairness an important problem for auditing and governance. Fairness assessment often requires more than identifying disparities in outcomes; it also requires understanding how those disparities arise. Causal approaches to algorithmic fairness address this explanatory problem by representing pathways through which protected attributes and other factors may influence outcomes. Yet causal reasoning in sociotechnical systems remains uncertain: relevant confounders may be unobserved, causal directions contested, and conclusions sensitive to modeling assumptions and representational choices. We introduce causal humility as a governance framework for managing the relationship between epistemic uncertainty and causal authority. Drawing on causal fairness research, epistemology, political philosophy, and science and technology studies, causal humility has four requirements: making consequential assumptions and omissions visible; preserving contestability across defensible causal accounts; recognizing indeterminacy as a legitimate analytical outcome; and maintaining proportionality between evidentiary support and institutional authority or action. We operationalize these requirements through Causal Cards, a documentation artifact for recording the causal basis, assumptions, omissions, model dependence, and governance implications of fairness claims. Using the UCI Adult dataset as a worked epistemic demonstration, we show how different causal discovery procedures applied to the same observational data render fairness-relevant pathways visible, absent, or unresolved. Causal humility treats such disagreement as information about the limits of what can responsibly be claimed and the authority those claims should carry.
Figures & tables
Component
Information recorded
Decision context and fairness claim
The decision setting, outcome of interest, fairness-relevant attributes, population, intended use of the analysis, and the specific causal fairness claim being evaluated.
Causal representation
The causal model or model set considered, including whether the structure is expert-specified, learned from data, partially identified, or constructed through multiple methods. Relevant substantive constraints on the model set should also be stated.
Consequential assumptions and omissions
Assumptions about causal structure, variable roles, confounding, measurement, directionality, and pathway permissibility that materially affect the fairness interpretation. Known omissions, suspected unobserved mechanisms, and relevant proxies should also be recorded.
Stability and contestability
Whether the fairness-relevant finding persists across plausible alternative models, sensitivity assumptions, variable interpretations, or pathway classifications, and which choices drive material disagreement.
Epistemic status
A characterization of the resulting claim as comparatively stable , assumption-sensitive , or unresolved under the alternatives considered. These categories describe evidentiary status rather than universal numerical thresholds.
Governance response
The action taken or recommended in light of the finding, its uncertainty, and the stakes of the decision, including where relevant additional review, mitigation, conditional authorization, monitoring, restricted use, or refusal to issue a categorical fairness determination.
Table 1: Core components of a Causal Card .
Figure 1: Causal structure learned from the UCI Adult dataset using PC. The fairness-relevant attributes (Race, Sex, Age, and Marital Status) and the outcome (Salary) are highlighted for emphasis.
Figure 2: Causal structures learned from the UCI Adult dataset using (a) FCI and (b) GES. The fairness-relevant attributes (Race, Sex, Age, and Marital Status) and the outcome (Salary) are highlighted for emphasis. Partially oriented relationships in FCI are treated as unresolved rather than as confirmed directed pathways.
Field
Illustrative entry
Decision context / claim
Illustrative assessment of whether fairness-relevant attributes have identifiable causal pathways to an income-threshold outcome in the UCI Adult data. The example is not used to justify real-world deployment or certification.
Outcome
Salary , represented as an income-threshold indicator.
Fairness-relevant attributes
Race , Sex , Age , and Marital Status .
Causal representations considered
Structures learned from the same observed feature set using PC, GES, and FCI. The outputs are treated as alternative representations for comparison, not as equally valid ground-truth causal models.
Key assumptions / analysis protocol
Causal structure is inferred from observational data. Candidate pathways are traced from fairness-relevant attributes to Salary . Only fully directed paths are interpreted as directed causal pathways; partially oriented FCI relationships are treated as unresolved.
Consequential omissions
The observed data do not directly represent mechanisms including discrimination, labor-market segmentation, inherited wealth, institutional exclusion, historical disadvantage, or differential access to educational and occupational opportunities.
Table 2: Illustrative Causal Card for the Adult income prediction example.
Algorithmic fairness studies how automated decisions distribute errors, opportunities, and harms across groups, and how to measure and reduce disparities.
Machine learning systems deployed for credit, hiring, and resource distribution are increasingly subject to regulatory oversight from policies such as the EU AI Act and GDPR. Current fairness governance practices rely on observational fairness metrics, post-hoc explainability, and immutable audit logs, but provide limited support for causal attribution and efficient evidentiary verification. We introduce Causal Evidentiary Governance (CEG), a framework in which regulated institutions commit to a versioned directed acyclic graph (DAG) that partitions causal pathways into allowable and disallowed groups. The Causal Harm Rate measures prediction variation attributable to disallowed causal pathways. Each decision is accompanied by a signed Decision-Evidence Packet (DEP), cryptographically binding the prediction to a digest of the published DAG and path-specific attributions. DEP digests can be appended to a Merkle tree to enable logarithmic-cost inclusion proofs. We validate CEG through a two-layer empirical methodology using demographic summaries from four years of PMA credit supervisory data to construct 10,000 synthetic credit applicants across four strategic DAG counterfactuals. Causal Harm Rate isolates injected causal effects more clearly than demographic parity or equalized odds. Cross-model validation and ablation studies assess robustness. Evaluation on the German Credit dataset shows that harm associated with specific causal pathways can be substantially understated by associational fairness metrics. Finally, a proof-of-concept implementation demonstrates operationally plausible throughput and highlights relevant performance tradeoffs.
Samah Kareem, Barış Çeliktaş
Department of Computer Engineering, Isik University, Istanbul, Türkiye. · Department of Computer Engineering, I¸sık University, Istanbul, Türkiye.
Causal discovery recovers directed structure from observational data and is increasingly used in clinical settings to support mechanism reasoning and fairness audits of predictive models. Path-specific counterfactual fairness asks whether a protected attribute influences an outcome through illegitimate pathways, but these estimands are defined relative to a supplied causal graph and therefore inherit whatever errors the discovery step introduces. Discovery methods are routinely scored on aggregate structural metrics that weight all edges equally, and no established evaluation asks whether the specific pathway an audit depends on survives discovery---or what the audit reports when that pathway is missing. Here we show that full-graph Graph-of-Thoughts reasoning yields acyclic discovered graphs that are structurally competitive with large language model (LLM) baselines, yet that structural fidelity alone does not guarantee fairness-faithful audits. We introduce GoT-CD, in which the reasoning unit is a complete candidate edge set: multiple graphs are generated in parallel, scored by a deterministic validity function, and merged under a hard union constraint that forbids invented edges, with greedy projection enforcing a DAG before commitment. GoT-CD returns a valid DAG on all five reported benchmarks and achieves the best DAG-valid F1 score among LLM methods on Asia, Alzheimer's, and COVID-Respiratory datasets. On an Alzheimer's benchmark with known unfair path, a post-hoc path-specific audit shows that five of eight discovered graphs recover no path from the sensitive attribute to the outcome and therefore report a null overall effect while mediated effects persist, necessitating downstream path-specific fairness analysis along with structural discovery.
Nitish Nagesh, Elahe Khatibi, Thomas Dean Hughes +3
Department of Computer Science, University of California, Irvine, Irvine, CA, USA · University of Orléans, Orléans, France
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 fY 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.