AI know it when AI see it: EU tries to clear up what an AI system is

The European Commission has issued guidance on what systems are artificial intelligence under the EU’s AI Act. Understanding the scope of “AI system” is the starting point for companies assessing the impact of the Act on their business.

The Commission’s guidelines aim to clarify what falls in scope of “AI system” under the Act. They follow a 2024 consultation which sought feedback on the seven main elements in the definition and examples of simple software systems that should be out of scope.

The seven elements of the “AI system” definition are:

1. a machine-based system
2. that is designed to operate with varying levels of autonomy
3. that may exhibit adaptiveness after deployment
4. and that, for explicit or implicit objectives
5. infers, from the input it receives, how to generate outputs
6. such as predictions, content, recommendations, or decisions
7. that can influence physical or virtual environments

Points made by the guidelines include:

1. Machine-based: The Commission confirms that this covers a wide variety of computational systems
2. Autonomy: Some degree of independence is necessary but systems that are designed to operate solely with full manual human involvement and intervention (directly or indirectly) are not in scope
3. Adaptiveness: The Commission says that this is not a decisive condition for determining whether a system is AI
4. AI system objectives: These objectives are internal to the system, referring to the tasks to be performed and their results, rather than the intended purpose of how the system is used
5. Inference: Arguably the key element to distinguish AI from other types of system, the Commission suggests that basic data processing and systems based on predefined rules or algorithms are not AI because they do not learn, reason or model like AI
6. Outputs: The Commission tries to draw a distinction between AI systems and other technology in that the former can handle complex relationships and patterns in data to generate more nuanced outputs
7. Interaction with the environment: AI systems are not passive but rather have an active impact on e.g. data flows or software ecosystems

The guidelines do not try to provide an exhaustive list of what is (or is not) an AI system but they should at least give some assurance that many types of “traditional” computing systems are not intended to be caught. Businesses will need to continue to consider the systems they deploy, provide and distribute in the EU against the AI Act’s key concepts and obligations.

See also: Guidance on “AI systems” - Does an elephantine definition matter?