Technical Framework
Overview of the Technical Framework
Positive AI aims to assess the maturity level of companies of all sizes and sectors in terms of responsible AI, to provide a tool for identifying ways to improve, and finally to enable them to obtain a label after an independent audit. The framework is built on the key principles of responsible AI defined by the European Commission.
From the 109 initial questions of the AI Trustworthy framework, Positive AI derived around forty more precise evaluation criteria. About twenty of them examine the company’s approach and AI governance, and another twenty focus on evaluating the algorithms themselves. Positive AI centers its evaluation on three areas: fairness and equity, transparency and explainability, and human intervention. Technically, it involves implementing some form of explainability — at least interpretability for the most sensitive algorithms.
The framework aims to enable positioning with as little subjectivity as possible concerning these questions. It was developed by business experts and data scientists working for the founding members of Positive AI. The association was then supported by the consulting and auditing firm EY to make this framework auditable.
An external and independent expert committee subsequently reviewed the framework. The three specialists involved are Raja Chatila, Emeritus Professor of Robotics, AI, and Ethics at Sorbonne University; Caroline Lequesne, Senior Lecturer in Public Law and head of the Master’s in Algorithmic Law and Data Governance at the University of Côte d’Azur; and Bertrand Braunschweig, Scientific Coordinator of the Confiance.ai program.
EY will conduct audits for companies seeking to obtain the Positive AI label, offering three gradual levels of certification depending on the company’s maturity.