Complexity and excessive requirements of regulation may effectively hamper the development of intelligence in Europe, especially for small and medium companies. Experts point out, that the document in its present form hinders, rather than supports innovation, which threatens to weaken the position EU in the global technology race.
- AI Chamber representing Polish startups and companies AI criticized the Code AI project before the EU Bureau ds. Artificial Intelligence.
- The criticisms include the complexity of regulation, difficulties in implementation, controversial rules on the laws. rights author and excessive requirements for suppliers of models AI.
- Code, in the opinion of the Chamber, is too detailed and impractical, which may slow innovation and weaken competitiveness EU in the global market.
AI Chamber (Chamber AI), an institution that already associates some startups, companies, organizations and ngos AI in Poland, in the name of represented entities submitted to the EU Bureau of ds. Artificial Intelligence official position on the subject of the Second Draft General Purpose AI Code of Conduct (Second Draft General Purpose AI Code of Conduct). Code is the leading document for suppliers of models AI General Purpose in demonstrating compliance with the Act of (AI Act) in the whole cycle life of models.
There is few revisions – the final Code should be ready in April , because the provisions of the AI law will begin already in August 2025 year.
EU bureaucracy contra global competition
As indicates AI Chamber, the biggest problem of the second Code project is its excessive detail and creation of new obligations, which often exceed the requirements AI Act. The complexity and strictness of regulations may discourage companies from developing their own AI models, especially by small and medium enterprises (SMEs), which may recognize the cost and effort to comply with these regulations as too high. According to Isba, already there AI Act and the burdens it imposes on the development of art intelligence, put the EU in an unfavorable situation
in the “race of art intelligence”, especially regarding the US and China. Codes of conduct, such as
GPAI Code of Conduct, should facilitate compliance with AI Act, to make Europe an attractive place
to create models of AI and solutions based on this .technology.
Read more
Restrictions on-sales chips AI. The US says what with Poland
– Instead of supporting a dynamic and competitive environment for development AI, the EU risks creating from Code a “bureaucratic monster.” Although in comparison with the earlier version of Code some improvements have been introduced, the document in its current form still does not fulfill its basic purpose. From the perspective of European entrepreneurs the value added to the Code is significant, while the burden of its implementation – both financial, and administrative may effectively stifle innovation enthusiasm. – states Tomasz Snażyk, CEO AI Chamber.
According to AI Chamber the flaws in the current version of the Code may lead to slow innovation, a consequence still strong weaken competitiveness of EU in the field of intelligence in the international arena. Code in the current form is the contrary to that, which the EU needs now, to strengthen its economy and increase productivity.
Over- regulation stifles innovation
Code contains overly detailed KPI (Key Performance Indicators), which are difficult to implement and unambiguous. For example the requirements for documentation are comprehensive, but lack a clear justification
of their necessity. Requiring detailed information about parameters of models AI and data used for their training may violate
secrets commercial and rights of intellectual property, and also threaten public safety and discourage the creation of innovation. Although
the purpose of the Code is transparency, there are still concerns about the level detail of publishing parameters models, such as: data used for training, testing and validation, data for. resources billing, whether reporting the use of energy.
Read more
L’Oréal and IBM create the first model AI, which supports the production of cosmetics
GPAI contra AI Act and legislation author
As highlights AI Chamber, the draft Code introduces obligations, which are not provided for in AI Act. For example imposes super strict requirements for conformity with author rights of collections of data of third parties. Requiring that the internal rules on author’s rights be “consistent with the commitments of the Code”, is problematic, because the Code should not restrict the freedom of suppliers in the independent assessment of author right EU. Claiming that developers models AI “reasonable actions” for the purpose of evaluating conformity with author’s rights of data parties third will exceed the scope of law EU.
Also, as the Bureau stresses, the obligation to apply policy at each stage of model development is too wide and difficult to implement, a postulate monitoring the rights of other entities is impractical and interferes in commercial contracts. What more, the introduction in the Code of additional provisions of extraterritorial character raises serious legal doubts.
Size models AI does not define risks
In the Council’s view, the size model AI or the height power necessary for its training are not good indicators for assessing real risk, which the
new technologies create. While generally the Bureau supports the exemption of the MSM from the imposed regulations, because they lack the resources to
fully comply with the Code, sometimes such approaches can create lack of law, in which models of AI offered by small
and medium companies may have lower security standards and create significant risks of abusing specified in AI Act.
The Act on artificial intelligence has the purpose of ensuring the safety of all models of artificial intelligence, independently of the size vendor, therefore it is worth putting more emphasis on what is the real risk of breaking safety standards and what measures are taken by the model supplier.
Read more
Code built on non existing standards
Codebasedonfuturestandards,whichhavenot yetbeendefinedbytheEUAIOffice.ThiscreatesuncertaintyontheconcreterequirementsofcompliancewiththeAIlaw,whichsignificantlyhindersplanningandinvestmentindevelopmentofmodelsofartintelligence.CodefromitsbeginningshouldgiverealandconcreteguidanceoncompliancewithAIAct.
-Muchoftheguidelinesremainoverlyburdensomeandnormative,imposingadditionalobligationsonsuppliersofartificialintelligence systems.Thismakesthedocumentnonfunctional.NewemergingfirmsAI,whichfacethedecision,whethertodevelopitsownmodel-inconsideringsocomplicatedrulesofCode,theymightjustfindthatitisnotworthtakingthiseffort,losingthebenefitsofAI-emphasizesTomaszSnażyk.
