The EU has the advantage of being the first in the world to propose a set of rules for artificial intelligence (AI). However, without appropriate action, this emerging market could be left to big players to the detriment of smaller companies.
The Artificial Intelligence Act (AI Act) regulates AI based on its potential for harm, in line with the new AI legislation. It falls under product law. For AI systems that are considered high risk, the regulations introduce stricter requirements.
Controls prior to are generally used to assess the conformity of a product that will not change during its life cycle. However, the AI itself can evolve as it receives new data and “improves” the system through machine learning.
MEP Josianne Cutajar, who has followed the file for the European Parliament’s Transport Committee (TRAN), does not want to stop there. In her view, given the constant evolution of this new technology, policymakers should remain vigilant and ensure that small and medium-sized enterprises (SMEs) can benefit from these regulations without making compliance too costly.
“The impact assessment carried out by the Commission largely underestimated the potential costs of compliance”said Sebastiano Toffaletti, Secretary General of the European DIGITAL SME Alliance, which represents the interests of SMEs in the digital sector, at the same event.
The industry association has formed an AI think tank to discuss the potential effects of the new legislation with 150 leading AI SMEs. Mr Toffaletti believes the EU risks proposing legislation that could, without realizing it, cede the market to the biggest players to the detriment of SMEs.
An analysis that Kilian Gross, head of the European Commission’s unit responsible for AI, does not share. According to him, if the administrative burden weighs differently depending on the size of the AI provider, the risk remains the same.
However, according to Mr Toffaletti, the AI regulation’s approach could lead to cost inflation as once the new rules are in place, SMEs can only discuss how to comply. are private entities that have an interest in rising costs.
Another fundamental criticism concerns the fact that conformity assessment prior to tends to be efficient in the case of economies of scale. In contrast, Toffaletti pointed out that AI systems tend to be highly customized products, often aimed at enterprises.
“The question is always: what is the alternative? If we do not conduct conformity testing or certification before these products are released, how can we ensure safety and confidence in these products? » asked Miriam Buiten, assistant professor at the University of St. Gallen.
While the overall architecture of the AI Regulation is unlikely to change at this stage of the legislative process, Intellera Consulting, a consultancy specializing in technical assistance to European institutions, presented a self-funded study on how to optimize compliance costs for SMEs at the same event .
Massimo Pellegrino, one of the paper’s authors, suggested making a distinction between SMEs that incorporate AI into their end products and those that integrate AI systems into their internal products. In the latter case, SMEs should simply be seen as users, which would help to reduce administrative burdens and avoid regulatory bottlenecks.
Another way to reduce compliance costs is to use technical standards, which are based on EU rules by default. But also in this case special attention should be paid to SMEs, since the process of developing standards is usually carried out by large companies with more financial resources.
Finally, Mr. Pellegrino also suggested giving more power to European digital innovation hubs (European digital innovation hubs, EDIH) so that they can become test and experience centers or even act as conformity assessment bodies.
[Édité par Anne-Sophie Gayet]