Lauriane Aufrant

• Towards accurate, transparent, and explainable systems in AI and NLP, in support of the AI Act

Artificial Intelligence

While it is feared that a large number of standards associated with the AI Act could create an excessive burden for SMEs to understand and implement the new requirements, on the other hand, if there remain gaps in the standards’ coverage of the AI landscape, this will create huge challenges for SMEs whose products sit precisely in such areas (inability to comply).

My work serves as support to the preservation of consumer rights, by enabling more transparency, comparability, and clarity on the actual performance of AI systems in the market.

• Supporting the AI Act with standards for trustworthy systems and datasets in AI and NLP

There is growing concern regarding the impact of the upcoming AI Act on the activity of SMEs in that field. While it is feared that a large number of standards associated to the AI Act could create an excessive burden for SMEs to understand and implement the new requirements, on the other hand if there remain gaps in the standards’ coverage of the AI landscape, this will create huge challenges for SMEs whose products sit precisely in such areas (inability to comply). I have started to make increased efforts to reach out to SMEs and collect their views and needs in those regards, and I am leveraging those inputs throughout my contributions in the various standards as well as in coordination.

My efforts are currently focused on answering the standardisation request that CEN-CENELEC JTC21 received from the European Commission in relation to the AI Act. As the target dates grow closer, it seems that the current pool of experts is not sufficient to answer the diverse needs expressed in the request. I have thus initiated reinforced outreach efforts, as well as onboarding sessions for new experts.

• R in ISO/IEC AWI 24970 on AI logging, in JTC 21’s work item JT021024 on AI Risk Management, JT021029 cybersecurity of AI systems, or JT021039 quality management system for regulatory purposes.

Artificial Intelligence

There is growing concern regarding the impact of the upcoming AI Act on the activity of SMEs in that field. While it is feared that a large number of standards associated to the AI Act could create an excessive burden for SMEs to understand and implement the new requirements, on the other hand if there remain gaps in the standards’ coverage of the AI landscape, this will create huge challenges for SMEs whose products sit precisely in such areas. I have started to make increased efforts to reach out to SMEs and collect their views and needs in those regards, and I am leveraging those inputs throughout my contributions in the various standards as well as in coordination. In parallel, and beyond the considerations for the AI Act, I am also including in my work continuous considerations for interoperability aspects, which appear to be key to enable easier entry into the market for European SMEs.

• Fast delivery of initial contents for a range of standards across the AI Act standardisation request

Artificial Intelligence

Harmonised standards are the preferred way for SMEs to comply with regulations under the New Legislative Framework (as the AI Act is) as they enable easy compliance with legal certainty and avoiding the need to resort to costly third-party legal support. The work is therefore to be developed with SMEs as one of the important targets. Furthermore, in the particular case of the AI Act, there are extra requirements in the EC standardisation request to consider the needs of SMEs/startups, and the AI Act Article about the quality management system includes special provisions for adaption of the obligations to the specificities of SMEs. In that context, I have pushed, first through comments in ballot then in the working group during comment resolution, for initiating dedicated discussions in that standard on those specificities with key stakeholders, which has proved beneficial even if still limited by the overall under-representation of SMEs in the standardisation work.

My work serves as support to the preservation of consumer rights, by enabling more transparency, comparability, and clarity on the actual performance of AI systems in the market. Its societal impact also encompasses ethical aspects of AI such as human agency, for which appropriate use of explainability methods is a key enabler. More generally, it benefits the society at large through its interplay with the upcoming AI Act that will impact daily lives in Europe.