In light of this, Peltarion was invited by the Swedish Ministry of Infrastructure to comment on the EU's proposed AI Act. We submitted our response last week and chose to comment generally on selected issues that we think are particularly relevant to highlight at this stage of the conversation. A more in-depth and comprehensive assessment of the proposal requires additional time for analysis. Therefore, we wish to continue the discussion on these issues.
You can find the full response (in Swedish) submitted by Peltarion to the Ministry of Infrastructure here. Below is an extract of our response.
Peltarion welcomes the initiative to create a harmonized regulation that both promotes the positive aspects of AI and protects fundamental rights. However, in light of our comments below, we cannot support the current proposal as it is written.
Technology and innovation are crucial for achieving sustainable solutions to global societal challenges and for stimulating growth and strengthened welfare. To achieve the objectives of Agenda 2030, the legislator therefore needs to promote the development and the use of new technologies, such as AI. Thus, it is of great importance that the regulation is well-founded and fit for the purpose.
In order for the actors concerned to be able to assess the consequences of their actions, it is necessary that the regulation is formulated in accordance with the principle of legality’s requirements for predictability and clarity. In addition, the regulation is required to be up-to-date and flexible in order to follow the rapid changes that are taking place in this field.
It is a very big challenge to regulate AI as several methods and techniques fall within the definition and the areas of application differ greatly. The technology is constantly evolving, both at a general level as well as regarding specific AI models. Peltarion is doubtful about the proposal's technology-based approach, especially since the proposed definition of “AI systems” is inadequate. Peltarion believes that the proposed definition is far too broad (for example, the definition includes statistical methods) and risks leading to overregulation for a large number of applications, which could have serious detrimental effects on society. Peltarion therefore initially advocates for a narrower definition, which can later be adjusted and possibly expanded if necessary.
Furthermore, the definition of "high-risk systems", for which strict obligations apply under the proposal, poses an additional challenge. What is considered high risk can, and probably will, vary over time. The ambiguity may make companies and authorities hesitant to invest in new AI opportunities as there is uncertainty as to whether they will be classified as high-risk systems - now or within a few years’ time. In order not to unnecessarily hinder innovation and the use of AI technology, the concept of “high-risk systems” should be given a more narrow definition.
The proposed regulation would also entail a significantly increased administrative burden for the actors covered by the rules. Moreover, effective supervision is costly and resource-intensive both in terms of competence and of time, for all actors involved, i.e. the EU, Member States, and the companies and organizations developing and using AI technologies. It is doubtful whether sufficient knowledge and resources needed will be in place to apply such an ambitious regulation. To avoid the undesirable consequences of the introduction of the GDPR in this regard, Peltarion calls for a thorough analysis of the lessons that can and should be learned about ensuring the support system needed.
Further discussions are warranted as to whether the proposal is an adequate way to achieve the behavioral changes that are wanted. Such an intrusive regulation as the one at hand risks adversely affecting Europe's competitiveness and must therefore be preceded by a very careful impact assessment, including an in-depth analysis of the needs and its proportionality. Peltarion considers that the proposal lacks such an analysis and calls for a more detailed account of the reasons as to why less intrusive means of achieving the aims sought have been ruled out.
Finally, Peltarion believes that a broader discussion is needed on what the proposed regulation may entail. We who work with and develop AI solutions on a daily basis, clearly see that this has a political dimension that is being largely overlooked. AI will affect everyone, but studies show that many people have little to no knowledge of AI. This is problematic, and ultimately a matter of equality and democracy.