European Parliament votes in favor of introducing AI legislation
The European Parliament has approved the European Commission’s proposal on the Artificial Intelligence Act by a large majority. This legislation must ensure that AI systems used in Europe are safe and meet strict requirements.
On Wednesday, the European Parliament voted on the AI Act, a bill introduced by the European Commission in 2021. 499 votes were in favor of the law, 28 were against and 93 MEPs abstained, writes the EU institution. However, the EP has made a number of changes to the law. Later in the day, parliament will begin negotiations with the European Council on the final version of the law. The aim is to reach an agreement with the EU member states before the end of the year. In that case, the law cannot come into effect until 2026 at the earliest, because the AI Act has a transposition period of two years. Nevertheless, it may become the first comprehensive AI law.
When the AI Act comes into effect, AI systems operating in Europe will have to comply with different rules, depending on the risk level the EU classifies them as. Systems used for ‘social scoring’, such as security cameras with automatic facial recognition, are seen as an unacceptable risk and are therefore completely banned. This also includes ‘predictive policing systems that are based on profiling, location or a history of criminal behavior’.
Next come “high-risk” AI systems, such as “systems that can be used to influence voters and the outcome of elections.” This includes, for example, systems that use social media platforms with more than 45 million users to recommend content. Such AIs are only permitted under strict conditions under the Artificial Intelligence Act.
There will also be a category for ‘AI for general use’. This includes generative AI models such as ChatGPT. Initially, no rules were drawn up for such systems in the AI Act, but the EP wants such models to first be registered in an EU database after the law is introduced and that they ‘try to limit possible risks as much as possible’. They must also meet certain transparency requirements, such as disclosing that the content is generated by AI and “helping to distinguish deepfake images from real images.” Finally, they must make clear what copyrighted material is used in their training and provide safeguards against the generation of illegal content.
Because the law can only enter into force in 2.5 years, says MEP Brando Benifei in a press conference that consideration is being given to bringing into force an ‘earlier implementation’ of the AI law for certain high-risk systems, such as generative AI. However, he states that nothing further has been short-circuited about this. “Even if that doesn’t work, there are also initiatives underway for a voluntary code of conduct.” European Commissioner Margrethe Vestager said last month that she expects such a code of conduct to be ready within a few weeks.