Skip to Main content Skip to Navigation
Preprints, Working Papers, ...

Statistique & Règlement Européen des Systèmes d'IA

Abstract : Following the publication of the white paper for an approach to AI based on excellence and trust, the European Commission (EC) published numerous proposals for regulatory texts, including an Artificial Intelligence Act (AI Act) (2021) establishing harmonized rules on artificial intelligence (AI). What will be the consequences and impacts of the future adoption of this text from the point of view of a mathematician or rather statistician involved in the design of high-risk artificial intelligence (AI) systems? What tools and methods will make it possible to comply to the obligation of a rigorous and documented analysis of the data processed, the performances, robustness, resilience of the algorithm, its explainability, risks, for fundamental rights, of discriminatory biases? These questions are illustrated by a numerical example analogous to a credit score in search of a less bad compromise between all the constraints. We conclude on the advances and limitations of this regulation project for high-risk AI systems.
Complete list of metadata

https://hal.archives-ouvertes.fr/hal-03253111
Contributor : Philippe Besse <>
Submitted on : Tuesday, June 8, 2021 - 10:25:23 AM
Last modification on : Thursday, June 10, 2021 - 3:38:55 AM

File

statReglEuropIA.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : hal-03253111, version 1

Citation

Philippe Besse. Statistique & Règlement Européen des Systèmes d'IA. 2021. ⟨hal-03253111⟩

Share

Metrics

Record views

221

Files downloads

164