The working group of developers, evaluators and users of artificial intelligence algorithms launched by LNE to create a certification system for artificial intelligence has just published a first version of the standard. This text, applicable to the processes of design, development, evaluation and maintenance in operational conditions of AI, is now proposed in a public call for comments open until May 21, 2021.
As the Artificial Intelligence (AI) market has exploded in recent years, affecting all industries, the prospects for development are ever increasing.
AI solutions are becoming ubiquitous, but the levels of maturity and performance are very uneven and it is not easy to ensure that your supplier has respected its requirements in terms of ethics, transparency or confidentiality. Today, it is complex for a company wishing to integrate AI into its activities to choose between several solutions due to the lack of a common reference.
Indeed, to be able to compare AI systems, it is essential to have criteria shared by all, especially concerning the definition of the field of use of the AI functionality which must be precise, unambiguous and communicated to the customer.
Given the strong development of this market and the growing need for confidence in the performance of AI systems, and its dual expertise in AI evaluation and certification, LNE has developed a certification standard in collaboration with various AI developers and users, to provide this confidence.
The objective of this certification is to demonstrate that the AI supplier has a real know-how and a mastery of all the stages of the AI life cycle to achieve a result in line with the expectations defined with the customer, in terms of performance but also regulations, ethics, transparency or others.
In this standard, the choice was made to aim for a certification of the processes that frame the AI life cycle: the design, development, evaluation and maintenance of AI systems in operational conditions. From their proper application, intelligent systems will perform well. The requirements in the standard are therefore applicable to companies providing all types of AI based on machine learning, critical or not, regardless of the sector of activity.
A first version of the repository is proposed today in the context of a public call for comments. It covers the processes of design, development, evaluation and maintenance in operational conditions of AI.
Observations, comments and proposals can be submitted until May 21, 2021.
LNE and the members of the working group thank in advance those who will respond to this call for comments.
Once comments have been collected and taken into account, LNE will publish a first applicable version of the standard at the end of June 2021 and will be able to carry out the first audits with a view to certification in the meantime. LNE would also like to point out that no company has yet been certified or can claim a guarantee of future certification under this standard.
Learn more about the evaluation of intelligence systems