With its AI regulation, the EU wants to create a legal framework for the development and use of artificial intelligence for the first time. Parliament is scheduled to vote on the hard-won bill today.
The European AI regulation is the world’s first attempt to introduce a comprehensive law regulating artificial intelligence. It therefore has a pioneering character. The EU Commission hopes that, like the data protection regulation, it will find imitators abroad.
However, formulating the regulation in a balanced way was a difficult balancing act for the legislators in Brussels. The responsible committees in the EU Parliament have been struggling for 18 months to find a position on the EU Commission’s proposal.
classification by risk
The regulation provides for artificial intelligence to be classified according to the risks of its intended use, from low-risk, limited risk, to risky and prohibited AI.
Artificial intelligence capable of suppressing people should be banned altogether. These include “social scoring” systems that evaluate people’s behavior, the automated recognition of emotions, for example when suspects are questioned – as well as comprehensive surveillance with biometric real-time data in public.
According to the current draft, however, the data can be accessed retrospectively by court order if serious crimes are involved.
Controversy over biometric data
In fact, shortly before the vote, there is still a dispute in Parliament as to the extent to which biometric data should be used in criminal prosecution. Within the Christian Democratic EPP group, people insist on changes to the draft.
Specifically, it is about exceptions when AI is used to search for missing children or when it comes to fighting terrorism. Some in the EPP Group hope that the countries will also increase the pressure on Parliament to take further action on this matter. Greens, leftists and pirates, on the other hand, warn against mass surveillance.
ChatGPT would remain allowed
In addition to the forbidden AI, there should be other categories of artificial intelligence. High, moderate and low risk, such as AI-powered toys, should generally be allowed. This also applies to so-called generative AI, such as the chatbot ChatGPT, which can independently write articles using information collected on the Internet.
However, the following principle applies to these applications: the riskier the AI, the higher the requirements. For example, manufacturers must assess the risks of their products and meet certain standards for training data. Audit authorities should check that. Among other things, the Social Democrats have criticized the fact that the companies carry out their risk analysis themselves and only then do the audit authorities check whether it is valid.
The ordinance is also intended to regulate that artificial intelligence does not use distorted data sets and could therefore discriminate against people, for example when AI checks creditworthiness or is used to recruit employees. Examples from the US, where black people are less likely to get credit, are seen as negative examples.
Expectations and fears are great
The expectations of the European AI law are high, as are the fears. Sam Altman, CEO of the company OpenAI, which developed ChatGPT, had warned of the existential dangers of unregulated AI. Smaller developers, on the other hand, fear that they will not be able to implement the many documentation requirements. Civil rights activists hope that the AI law will no longer perpetuate discrimination and disadvantages.
When Parliament decides its position at midday, negotiations with the EU member states can then begin. Above all, the rules on the use of artificial intelligence in law enforcement are likely to cause controversy.
If an agreement is reached before the European elections next year, the AI regulation could probably come into force in 2026. Until then, artificial intelligence will continue to develop. This indicates that further AI regulations will follow in the European Union.