Counterfactual Target Cards: Operationalizing Target Specification Fairness for AI Act–Ready Systems

Tal (2023) argues many healthcare models optimize the wrong target (actual outcomes, not the counterfactuals decision-makers care about), a pervasive source of bias that persists even with perfect data. Meanwhile, Deck et al. (2024) note the EU AI Act pushes non-discrimination responsibilities into the design stage; Bahangulu and Owusu-Berko (2025) emphasize organizational governance and ongoing audits. We propose “Counterfactual Target Cards”—a standardized, auditable artifact and software toolkit to: (1) elicit the counterfactual target definition via structured vignettes and stakeholder workshops; (2) map it to real-world proxies, quantifying the mismatch as “target drift”; (3) run counterfactual stress-tests using causal methods (AFCR/AFCP workshops, 2021/2022; Mitchell et al., 2021) to estimate fairness and utility differences under alternative target operationalizations; and (4) generate AI Act–aligned documentation of choices, assumptions, and residual risks. We test this in two regulated domains with concrete legal stakes: automated hiring (Poe & El Mestari, 2024) and readmission prediction (Wang et al., 2023), and examine how compliance constraints (e.g., in EU non-discrimination law and domain-specific rules) shape feasible target choices. This goes beyond today’s fairness audits by making target-definition an explicit, auditable design decision with measurable consequences. The anticipated impact is twofold: better-aligned models (fewer harmful misallocations due to mis-specified targets), and governance artifacts that regulators and internal risk committees can actually act on.

References:

1. Evaluating Algorithmic Bias in 30-Day Hospital Readmission Models: Retrospective Analysis. H. E. Wang, Jonathan P. Weiner, S. Saria, Hadi Kharrazi (2023). Journal of Medical Internet Research.
2. Target specification bias, counterfactual prediction, and algorithmic fairness in healthcare. Eran Tal (2023). AAAI/ACM Conference on AI, Ethics, and Society.
3. Algorithmic bias, data ethics, and governance: Ensuring fairness, transparency and compliance in AI-powered business analytics applications. Julien Kiesse Bahangulu, Louis Owusu-Berko (2025). World Journal of Advanced Research and Reviews.
4. The Conflict Between Algorithmic Fairness and Non-Discrimination: An Analysis of Fair Automated Hiring. Robert Lee Poe, Soumia Zohra El Mestari (2024). Conference on Fairness, Accountability and Transparency.
5. Implications of the AI Act for Non-Discrimination Law and Algorithmic Fairness. Luca Deck, Jan-Laurin Müller, Conradin Braun, Domenique Zipperling, Niklas Kühl (2024). EWAF.
6. Algorithmic Fairness through the Lens of Causality and Privacy (AFCP) 2022. Awa Dieng, Miriam Rateike, G. Farnadi, Ferdinando Fioretto, Matt J. Kusner, Jessica Schrouff (2022). AFCP.
7. Algorithmic Fairness through the Lens of Causality and Robustness (AFCR) 2021. Jessica Schrouff, Awa Dieng, Miriam Rateike, Kweku Kwegyir-Aggrey, G. Farnadi (2021). AFCR.
8. Algorithmic Fairness: Choices, Assumptions, and Definitions. Shira Mitchell, E. Potash, Solon Barocas, A. D'Amour, K. Lum (2021). Annual Review of Statistics and Its Application.