Regulation of artificial intelligence: EU Parliament votes on AI law

The vote in the European Parliament on the European Commission’s proposal for a regulation on artificial intelligence (also known as the “AI law”) has taken place. This clears the way for negotiations with the EU member states on the final form of the regulation.

Unclear risk classification

Specifically, the regulation follows a risk-based approach, according to which AI systems are assigned to a risk class and high-risk systems must meet stricter requirements than AI systems in a low risk class. In this context, a current study by the “appliedAI” platform warns against slowing down innovation . According to this, for about 40% of the 106 applications examined, it was currently not possible to classify them unequivocally. In conclusion, almost 60% of future applications would involve considerable additional financial and human resources.

Risk management systems for basic models

An example of such additional work is the regulations for the introduction of risk management systems. These should not only apply to high-risk AI systems, but also be mandatory for developers of so-called foundation models. A basic model, so it reads in the final version of the draft, means “an AI model that has been extensively trained on a wide data base and is designed for general output and adaptable to a wide range of different tasks.” The definition includes language models such as GPT-4 and Google Bard as well as Image generators like Midjourney or Stable Diffusion. Since basic models by definition have an (unmanageable) multitude of possible applications, a meaningful risk assessment is almost impossible.

Over-regulation threatens to slow down start-ups and SMEs

In addition, the development of risk management systems costs one thing above all: a lot of money. Costs that large companies like Google and Microsoft can handle much more easily than small and medium-sized companies and even more so as start-ups. Ultimately, the AI ​​law could become a stumbling block for many companies.

practical tip

Companies that use AI-based technologies should familiarize themselves early on with the principles of the risk classification laid out in the draft of the AI ​​law and observe the negotiations in Brussels in order to prepare specific decisions for their own company and the products to be developed by them, if necessary to be able to