Implementation of the AI Act raises difficult questions. What is the scope of the AI system defintion? Based on which criteria can the risk category of an algorithm be identified? Two dynamic questionnaires help deployers and producers with:

1. Identification of AI systems and high-impact algorithms;
2. Identification of risk category and prohibited applications.

Complex legal definitions from the AI Act and corresponding guidelines are translated into understandable language, such that the tool can be used by a wide audience.

Since many straightforward algorithms that impact people are not considered AI systems, the first questionnaire also identifies high-impact algorithms. This concept is used by the Dutch government to refer to simple algorithms that do not meet the definition of an AI system under the AI Act but still require control measures. More information can be found in the Algorithm Registry Guidance Document of the Dutch Ministry of the Interior.

The tool is developed open-source and can be used for free within your organisation to manage algorithms. All potential outcomes of the first questionnaire are shown in the figure below on this webpage. The flowcharts of both questionnaires can also be found below.

The outcomes of the first questionnaire are displayed in the below figure. The following categories are distinguished:

• Algorithms: fall outside the scope of the AI Act, no additional control measures are needed.
• High-impact algorithms: fall outside the scope of the AI Act, additional control measures are needed.
• AI systems: are in scope of the AI Act, only additional control measures are needed in case of high-risk AI systems.
• High risk AI systems: are in scope of the AI Act, additional control measures for high-risk AI systems are needed through harmonized standards.
• Prohibited AI systems: are in scope of the AI Act, usage of this type of AI systems is prohibited in the European Union. More information about this category is provided by the Dutch government.

The questions in the AI Act Implementation Tool are developed in collaboration with the municipaility of Amsterdam. The source code of the tool can be found on Github and can be (re-)used under the EUPL-1.2 license.

.

Considerations and choices made during development of the questionnaires regarding the legislative text of the AI Act, guidelines from the European Commission, and the Algorithm Registry Guidance Document are described in the white paper ‘Implementation of the AI Regulation – Definition of an AI System’. The policy briefing explains why the guidelines from the European Commission on the definition of an AI system contradict the legislative text. Both the white paper and the questionnaires have been adjusted following the publication of the guidelines by the European Commission.

Through examples the slides decks explain the intricacies of the AI system definition.

10 examples of (non) AI systems

Rule-based algorithms under the AI Act

Definition of an AI system under the AI Act