Schalast | Understanding and Definition

Depending on context, the term “AI system” may refer to a wide variety of various types of technology. The EU Commission’s proposal for an AI Regulation is criticized for being too wide in scope, and this is true even within the area of computer science, where there is no commonly recognized definition.

When discussing AI systems, the term “machine learning” (ML) is frequently used. Using this method, software may be taught to identify patterns and rules in data, extend those rules, and then apply them in novel contexts. This has resulted in the development of a fresh body of law pertaining to AI.

Article 3(1) Draft AI Regulation defines an ‘artificial intelligence system’ (AI system) as software that is developed with one or more of the techniques and approaches listed in Annex I and which can, for a given set of human-defined objectives, generate outputs such as content, predictions, recommendations, or decisions influencing the environments they interact with. In light of this, it has been argued (with a certain amount of criticism) that the phrase “AI system” in the Regulation should be synonymous with “software” for the sake of clarity.

Interestingly, the principle on which the proposal is built is itself dynamic: Articles 4, 73 Draft AI Regulation state that the EU Commission will be empowered to adopt delegated acts to amend Annex I where necessary and therefore update the Regulation to technical progress. The German Data Ethics Commission also advocated for a more precise definition of AI, defining it in its final report as a “catch-all term for technologies and related applications based on digital methods which involve the machine processing of potentially very large and heterogeneous data sets in a complex procedure that mimics human intelligence.”

However, it also advocated for more comprehensive regulation that covers all “types of algorithmic systems.” Since the proposed definition of an AI system in the Draft AI Regulation is so wide, it also encompasses “‘simple’ control systems that are based on algorithms (=rules) determined ‘manually’ by experts,” which is a nod to this requirement. On the following pages, we will highlight some of the numerous legal concerns that arise from this scenario.