Translated with DeepL
After just three years, more than half of the Swiss population is already using generative AI tools such as ChatGPT. Never before has a technology spread so quickly in Switzerland as artificial intelligence. The Swiss economy also has high hopes for it. Generative AI could increase GDP by up to 11 per cent in the coming years.
However, this rapid adoption of AI also brings with it risks such as increased fraud through AI-generated content or challenges in the labour market. These negative effects can be countered with regulation. However, it takes time to draw up new laws. Meanwhile, the technology is constantly evolving.
With the AI Act, the European Commission brought the world's first comprehensive AI regulation into force on 1 August 2024. It classifies and regulates AI systems according to risk levels. Its development took several years. One of the stumbling blocks was the speed at which AI is developing. Systems such as ChatGPT, which was launched in November 2022, were barely covered in the first draft of the law. The Commission had to go back to the drawing board.
Drafting the AI Act was an enormous effort. But that's not the end of it. In order to be able to apply it, so-called harmonised standards are also needed: technical standards that help to implement the requirements of the legislation in concrete terms. The development of such standards is a consensus-based process that takes time. Time that the Commission is running out of. This is because binding deadlines are associated with the entry into force of the AI Act.
The technical committee JTC 21 Artificial Intelligence has been tasked with drafting these standards. Because it is behind schedule, the Commission is putting pressure on the European standardisation organisations (SDOs), the European Committee for Standardisation (CEN) and the European Committee for Electrotechnical Standardisation (CENELEC), to which JTC 21 is affiliated. The technical board of both SDOs has reacted and recently decided to shorten the usual development time so that the standards are available in time for the end of 2026.
This decision was sharply criticised by the members of JTC 21. They believe it jeopardises the principle of consensus. The development of broad-based harmonised standards would no longer be possible.
The experience with the EU AI Act should be a lesson to us. Switzerland should not attempt to regulate artificial intelligence comprehensively, but rather on a sector-specific basis as far as possible. This reduces complexity, effort and the risk of being overtaken by ongoing developments.
Any Swiss AI law should be based on established international standards. It makes little sense to enact regulation that cannot be implemented in practice. We should not put ourselves under unnecessary pressure and risk regulation undermining the very principles it is supposed to protect: Transparency, inclusion and trust.
This article was written by the author and does not represent the official opinion of the Swiss Academy of Engineering Sciences SATW.
