Towards the engineering of trustworthy AI applications for critical systems - The Confiance.ai program - Second Edition - Confiance.ai
Autre Publication Scientifique Année : 2024

Towards the engineering of trustworthy AI applications for critical systems - The Confiance.ai program - Second Edition

Michel Morvan
  • Fonction : Directeur de publication
  • PersonId : 828776

Résumé

This white paper sums up the journey and findings of the program Confiance.ai, the technological pillar of the Grand Défi “Securing, certifying and enhancing the reliability of systems based on artificial intelligence” launched by the Innovation Council of the French Administration. The two other pillars of this state initiative focus on standardization (norms, standards and regulation toward certification) and application evaluation. The active collaboration of over 50 partners including large-scale multi-sector industrial partners and research centers, for over four years, has addressed numerous challenges on the topic of engineering Trustworthy AI for critical systems as it aimed at the convergence of solvability of current industrial challenges and applicability of innovative research developments. As the largest yet technological research program in the national AI strategy, Confiance.ai began in 2021 by a first year dedicated to covering the state of the art and pre-existing tools related to the integration and evaluation of data-driven AI. The following years focused on characterizing industrial use cases, developing technological components for assessing trustworthiness, and constructing numerous guidelines and an End-to-End method for the trustworthy design, integration, and evaluation of Machine learning (ML) components. The previous white paper in 2022, provided initial results of the program including the first steps toward engineering trustworthy AI, use cases, a first version of a pipeline, a taxonomy and key attributes to characterize AI trustworthiness. As the program evolved, so did the initiatives toward regulation including the AI Act, making the program a bidirectional partaker on the process; this is: ensuring the production of methodological guidelines and digital components that incorporate state of the art developments and envisaged European constraints, as well as the contributing to these initiatives technologically, methodologically and in support of standards. As the AI Act, another element of rising interest during the course of the program is the topic of generative AI. Even though the subject itself was not addressed in the program, some results still hold in this field (e.g. image generation through diffusion models and experience on an NLP use case) and motivate the pursuit of this research in the initiatives ensuring the continuation of Confiance.ai.
Fichier principal
Vignette du fichier
L0703_Livret_Livre-Blanc_v4.0.pdf (7.25 Mo) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-04709737 , version 1 (25-09-2024)

Licence

Copyright (Tous droits réservés)

Identifiants

  • HAL Id : hal-04709737 , version 1

Citer

Michel Morvan. Towards the engineering of trustworthy AI applications for critical systems - The Confiance.ai program - Second Edition. 2024. ⟨hal-04709737⟩
57 Consultations
8 Téléchargements

Partager

More