The EU AI Act is currently still in the negotiation phase, or trilogue, in Brussels. If passed, it will have a significant impact on the M&A sector. The EU is encouraging European companies to become more active in the artificial intelligence (AI) space, but there is still considerable uncertainty as to which systems and applications will be covered by the new law.
For example, while the European Parliament (EP) would like there to be a broad list of banned AI systems, including software that processes facial images from the internet, the EU Council prefers a narrower list.
Importantly, there is no agreement on the definition of AI itself.
Numerous studies on the definition of AI have been conducted in Europe. Many experts are concerned that the definition’s lack of clarity will complicate the application of the AI Act. Industry is concerned that the AI Act’s definition of AI may be too broad and may include applications such as simple calculations in a spreadsheet. On the other side of the spectrum, some fear that too precise a definition will undermine the AI Act in the face of rapid technological change.
EU INSTITUTIONS’ POSITIONS
The European Commission’s proposed definition includes a number of techniques and applications listed in the Annex to the Regulation.
However, both the EU Council and the EP moved their definitions to the main body of the AI Act. The EP attempted to align the definition of an “AI system” in Article 3 with the AI definition developed by the Organisation for Economic Co-operation and Development (OECD).
EU member states represented in the EU Council are more concerned that the AI Act is too broad and will cover traditional software applications. The Council is trying to limit the definition to systems that use machine learning, logic, or knowledge-based approaches.
PROPOSED WORDING FOR ARTICLE 3
EU Commission: “[An] ‘artificial intelligence system’ (AI system) means software that is developed with one or more of the techniques and approaches listed in Annex I and can, for a given set of human-defined objectives, generate outputs such as content, predictions, recommendations, or decisions influencing the environments they interact with.”
EP: “[An] ‘artificial intelligence system’ (AI system) means a machine-based system that is designed to operate with varying levels of autonomy and that can, for explicit or implicit objectives, generate outputs such as predictions, recommendations, or decisions, that influence physical or virtual environments.”
EU Council: As noted, the Council is trying to limit the definition to systems that use machine learning, logic, or knowledge-based approaches.
The approaches of the EP and the EU Commission have so far diverged considerably.
In its proposal, the EU Commission emphasizes that the system can “generate outputs for a given set of human-defined objectives.” This reference to human-defined objectives does not exist in the EP proposal.
The Commission’s proposal refers to Annex 1, where “AI systems” are defined in more detail. This reference makes it difficult to predict what the definition will include. The EP version also contains some vagueness; for example, with the term “different levels of autonomy.”
These are not the only contentious definitions in the AI Act. The EP has introduced other new definitions in its proposal, including calling users of AI systems “users” and adding new definitions for “data subjects,” “basic model,” and “general purpose AI system.”
The general problem is that the AI Act will not come into force until 2025 at the earliest. Either the AI Act remains vague in its definitions to capture new developments, or it misses its intended targets because of the rapid pace of development.
By way of comparison, the US approach to AI regulation is even more vague: the definition in the Blueprint for an AI Bill of Rights reads, “This framework applies to (1) automated systems that (2) have the potential to meaningfully impact the American public’s rights, opportunities, or access to critical resources or services.”
What would be desirable is a joint US–EU approach to what should be defined and regulated as AI by the EU AI Act and new US regulations. But that is probably wishful thinking.