Interpretive basis for referring to specific persons in AI accuracy provisions

Determine why the EU AI Act's accuracy provisions refer to specific persons, particularly given that providers must document compliance before a high-risk AI system is placed on the market or put into service.

Background

The EU AI Act requires information about accuracy and performance concerning specific persons or groups of persons on whom a high-risk AI system is intended to be used. The paper notes a tension between these individual-oriented requirements and the statistical, aggregate nature of machine-learning systems.

The authors explicitly state that the rationale for referring to specific persons is unresolved, especially because the relevant accuracy information may be required ex ante, before the system has been deployed and before evidence about particular individuals exists.