Submitted text
The Federal Council is instructed to create the legal basis for a declaration obligation for artificial intelligence applications and automated decision-making systems. This declaration obligation should apply to applications in both the public and private sectors. This ensures uniform standards and thus creates trust for both citizens and companies that use AI technologies.
Justification
Algorithmic systems and artificial intelligence are already being used by state and private actors. However, it is not always clear to all those affected that these are being used. Transparency and traceability are among the measures recommended by both experts and representatives of the industries concerned. The declaration obligation is intended to ensure that all applications and systems based on artificial intelligence that are used in Switzerland are obliged to disclose their AI components. This includes information about the algorithms used, the data sources, the training data and the evaluation methods. Such disclosure makes it possible to better understand the functioning and potential of AI applications and to identify possible prejudices, discrimination or undesirable effects. The introduction of a declaration obligation for artificial intelligence strengthens Switzerland as a pioneer in the field of ethical AI and positions us as a responsible country that takes the protection of privacy, the fight against discrimination and the promotion of transparency in technology development seriously.
Opinion of the Federal Council of 30.8.2023
The issue of creating transparency in the use of AI systems is currently being discussed both as part of the EU’s work on the “AI Act” and in the negotiations for a binding agreement on AI in the Council of Europe under the Swiss Presidency. Transparency and traceability as basic principles are important components of both sets of rules. At the same time, both sets of regulations are based on a risk approach that provides for correspondingly different regulatory requirements for applications with different risks. Depending on the context, a graduated and differentiated regulatory approach should be applied. The EU currently does not provide for a declaration obligation for applications that are only assigned minimal risks. How exactly the Council of Europe formulates such a graduated approach is currently still the subject of negotiations. However, it can also be assumed that agreement will be reached on a declaration obligation for applications above a certain risk level (but not for all AI applications).
When the new Data Protection Act (nDSG) comes into force on September 1, 2023, an information obligation will already apply in Switzerland for decisions that are based exclusively on automated data processing (Art. 21 nDSG). This provision covers decisions that are of a certain complexity and that have legal consequences for the data subject or significantly affect them. The provision applies to both the public and private sectors. According to Article 25 paragraph 2 letter f nDSG, the person responsible for the processing is also obliged to inform the data subject of the logic on which the automated individual decision is based as part of the right of access. In addition, the data subject must also be informed of the amount and type of information used and its weighting. Based on this data, the data subject should be able to understand the decision and, if necessary, contest it.
In his response to the Postulate Dobler (23.3201) the Federal Council has already announced that it will draw up a political overview with possible options for sectoral and, where necessary, horizontal regulatory measures in the area of AI by the end of 2024. The relevant analyses will be carried out within the existing bodies (in particular the Tripartite Platform and its Administrative Committee, the interdepartmental EU Digital Policy Coordination Group, the AI Guideline Monitoring and the AI Competence Network (CNAI)) and with the involvement of all federal agencies concerned. The interpretative analysis will also address the question of the extent to which a regulation regarding the obligation to declare AI systems is appropriate in Switzerland that goes beyond the regulations already provided for in the nDSG.