These tools and metrics are designed to help AI actors develop and use trustworthy AI systems and applications that respect human rights and are fair, transparent, explainable, robust, secure and safe.
Peer-induced Fairness: A Causal Approach for Algorithmic Fairness Auditing
Peer-induced fairness is a framework for auditing the fairness of algorithmic decisions, based on causal reasoning. The EU AI Act, which entered into force on 1 August 2024, requires high-risk AI applications to meet strict standards of transparency and fairness. However, most current auditing methods only detect whether outcomes differ between groups. They do not take more complex situations into account. The framework aims to provide a scientific, reliable way to audit fairness, and was published in August 2024.
The framework combines two approaches. Counterfactual fairness asks whether a decision about a person would change if they belonged to a different group. Peer comparison compares each individual with similar peers from the other group. Together, these allow the framework to distinguish whether an unfavourable decision results from algorithmic discrimination or from the individual's own characteristics, such as a firm's financial weakness. This makes audit results more transparent. The framework is designed to work across domains and with data of varying quality, including skewed distributions.
The framework can be used by AI developers as a self-assessment tool, and by external auditors to check compliance with the EU AI Act. The authors demonstrate it on access to finance for small and medium-sized enterprises. They find that 41.51% of micro-firms face discrimination compared with larger firms.
Auto-discovered on 2026-02-11 by OECD Catalogue Automation
About the metric
You can click on the links to see the associated metrics
Metric type(s):
Objective(s):
Purpose(s):
Target sector(s):
Lifecycle stage(s):
Usage rights:
Target users:
Risk management stage(s):
Partnership on AI
