An Analysis of the New EU AI Act and A Proposed Standardization Framework for Machine Learning Fairness

The EU AI Act lacks specific fairness metrics: The paper identifies a critical gap in the new EU regulation—it provides no quantifiable fairness metrics for AI systems, making it difficult for companies to know exactly what they need to measure and report to comply with the law.

Confusing terminology creates legal and business risks: The EU AI Act uses "transparency," "explainability," and "interpretability" interchangeably without clear definitions, and notably omits any requirement for transparency about ethical compliance itself. This ambiguity creates substantial liability risk that could discourage investment in AI development.

Industry standardization is needed alongside regulation: The paper argues that broad regulations alone cannot provide the level of detail needed for practical implementation. The authors recommend developing standardized industry best practices that work alongside regulations to ensure compliance without stifling innovation.

A public assessment framework could help: The researchers propose creating a public system to evaluate whether AI systems are actually fair and transparent, using examples from automatic speech recognition (ASR) and speech synthesizers to demonstrate how such a framework could work in practice.

Fairness transparency should be a strategic priority: The paper emphasizes that making AI fairness transparent—not just the AI systems themselves—is strategically important for building trust and ensuring accountability in regulated AI systems.