A quality standard and certification for AI
Building trust. As artificial intelligence-based tools proliferate around the world and invade our daily lives, the fears associated with these new technologies are anything but virtual. ISO/IEC 42001, a voluntary international standard recently published by AFNOR Editions under the title ISO/IEC 42001 (the prefix IEC stands for International Electrotechnical Commission). We’re talking here about voluntary standardization, not regulation as envisaged by the European AI Act, even though the two go hand in hand.
One of the principles of the standard is to ensure that humans always retain control over machines,” explains Paul Houzé, who is in charge of standardization at Microsoft and a member of the French standardization committee headed by AFNOR. The text is based on the management systems model that is ISO’s strength (including the famous ISO 9001 on quality management), with a structure that takes account of the needs of all stakeholders, risks and modularity. ISO/IEC 42001 is adaptable to all contexts and all future innovations. “
ISO/IEC 42001: wherever certification is needed
Artificial intelligence is very different depending on whether it is intended for industrial, military or commercial use. Applicable to these different situations, the standard nevertheless lays down general principles concerning data protection, the specification of the information used to “feed” the AI, its robustness, as well as its transparency and explicability. It’s up to each player to make these factors their own, according to their needs and special characteristics. AFNOR Compétences training is already available here (in French) to perfect your skills.
“This is a certifiable standard. Independent auditors will therefore check a designer’s practices before awarding a sign of recognition, which will prove to all stakeholders (partners, legislators, customers, etc.) that the set principles have been complied with. It’s a tool that will create trust between market players,” adds Paul Houzé. “The spirit of this certification is close to that of quality certification for medical devices,” continues Brice Gilbert, who prepared the marketing of this AFAQ certification at AFNOR Certification. “An organization that achieves ISO 42001 certification demonstrates its commitment to managing artificial intelligence ethically and responsibly,” adds Mame Astou Ndiaye, who assisted Brice Gilbert in this work. Another aim is to prevent abuses and provide a framework for a sector that has been booming in recent years. You can ask for a quote here.
Using the standard to support innovation
At the same time, countries are arming themselves and legislating. The legal framework is about to get tougher, especially in France. But ISO/IEC 42001 is prepared for future developments. “The standard is designed to adapt to future regulatory requirements in whatever country. This is the strength of our management system, which provides a framework without being restrictive,” comments Paul Houzé. At a time when AI is developing at an astonishing pace, the aim of the standard is not to prevent innovation, but rather to support it by setting out shared principles that build consensus and point the way forward. “Read the article (in French) by a group of professionals in favor of the risk-based regulation of AI, including Patrick Bézombes, chairman of the AFNOR standardization commission, Laurence Devillers, a professor at the Sorbonne and member of this commission, and Franck Lebeugle, head of standardization activities at AFNOR.
As with any certifiable standard based on an NF norm, the certificate will be valid for three years, with follow-up and renewal audits. ISO/IEC 42001 could quickly become a must-have to reassure users and, more generally, a society worried about the place of technology and machines. Or how to make AI a tool for progress… and not for enslavement to incomprehensible algorithms. Want to know more? Join us in Paris on the morning of February 1, 2024 for a morning of discussion with the AFNOR group.