European Artificial Intelligence Act – European Regulation of AI is on its Way

The “Proposal for Regulation of the European Parliament and of the Council laying down harmonized rules on Artificial Intelligence” – in short, the Artificial Intelligence Act – of the EU aims at regulating AI in Europe. The document is interesting, although time-consuming, to read, as it includes both a risk-based classification of AI applications and a rather broad definition of AI. It also proposes certification requirements for providers of AI systems. 

The definition of AI provided in Annex I of the Act is rather general, including (quoted from the Act):

1. “Machine learning approaches, including supervised, unsupervised and reinforcement learning, using a wide variety of methods including deep learning;
2. Logic- and knowledge-based approaches, including knowledge representation, inductive (logic) programming, knowledge bases, inference and deductive engines, (symbolic) reasoning and expert systems;
3. Statistical approaches, Bayesian estimation, search and optimization methods.”

This definition is static, not accounting for the dynamic, fast-pace development of the field, and very generic, including also techniques such as statistical models and classical optimization methods from Operations Research. Defining those generally as being part of AI is debatable and might, through the AI Act, affect many systems using such methods that are successfully deployed and operating in industry since decades. 

The European Digital SME Alliance has analyzed the proposed Act from the perspective of innovative SMEs and provides a summary of the analysis (March 2022) as well as resulting proposals for modification. The seven key points of concern are:

1. The AI Act may stifle the innovativeness of Europe’s businesses.
2. The definition of AI provided in the proposal is too broad.
3. Real risks are not adequately addressed (to name a few: AI systems that may provide false information, misinformation, deep fakes; AI systems for deriving the psychological profile of natural persons from proprietary or public data sources; AI systems for assessing natural persons for obtaining access to or determining the fee to be paid for important insurances such as health insurances). 
4. Conformity assessments & compliance costs will be prohibitive for SMEs.
5. While standardization will play a key role in the AI Act, SMEs are largely underrepresented in key institutions.
6. The concepts of liability and “placing on the market” do not reflect how AI products are developed.
7. The proposed sandboxes are not designed to address the needs of SMEs.

The earlier position paper on the AI Act (September 2021) explains specific aspects and concerns from the SME perspective in more detail. The AI Act will be relevant for providers and users of AI systems, and the EU should consider the fact that SMEs are providing a high level of agility and innovation in this particularly dynamic field. A complex regulatory framework for AI could affect this SME ecosystem significantly.