Brussels (dpa) - After tough negotiations, the EU has agreed on stricter rules for artificial intelligence (AI). These are the world's first rules for AI, the European Parliament and the EU member states announced in Brussels on Friday evening. The most important questions and answers:

What is AI and how does it work?

Artificial intelligence (AI) is the attempt to transfer human learning and thinking to a computer. The aim is to perform complex tasks that would normally require human intelligence. Despite all the progress made, general problem-solving machines (Artificial General Intelligence) are not yet in sight. However, more narrowly defined AI applications are already widely used in today's world: These include automatic translations, personalized recommendations for online shopping, facial recognition on cell phones, as well as intelligent thermostats and navigation systems. Generative AI applications such as the text robot ChatGPT also belong to the more narrowly defined AI applications.

Why do we need a law for this?

AI is considered a technology of the future. Experts believe that the technology could affect practically all aspects of the economy and everyday life in the future and that it will massively change the job market, for example: Some jobs will change, others may disappear altogether. However, AI is also a technology that harbors dangers. For example, the head of ChatGPT inventor OpenAI, Sam Altman, warned against misinformation with the help of artificial intelligence and therefore spoke out in favor of regulation. Photos and videos can be easily manipulated by AI. Another problem is that AIs are sometimes trained with distorted data sets and therefore discriminate against people. The use of AI in warfare is also considered possible.

What has the EU now agreed on?

The regulations now presented define obligations for AI based on their potential risks and effects. AI is classified as particularly risky if it has the potential to cause significant harm to health, democracy, the environment or security, for example.

Certain applications are completely banned, such as biometric categorization systems that use sensitive characteristics such as sexual orientation or religious beliefs. The untargeted reading of images from the internet or from surveillance footage for facial recognition databases will also not be permitted. However, there will be exceptions for biometric identification in public spaces in real time, for example in the event of a terrorist attack or the targeted search for victims of human trafficking. This point was the subject of intense debate; the EU Parliament actually wanted a complete ban.

Another point of contention was the regulation of so-called base models. These are very powerful AI models that have been trained with a broad set of data. They can form the basis for many other applications. This includes GPT, for example. Germany, France and Italy had previously demanded that only specific applications of AI should be regulated, but not the basic technology itself. The negotiators have now agreed on certain transparency obligations for these models.

What are the reactions?

Depending on the interest and perspective, the criticism is very different: either too strong, anti-innovation regulation or insufficient restrictions on AI applications. EU Commission President Ursula von der Leyen welcomed the agreement and described the law as a "global first".

According to the Federation of German Industries (BDI), Europe is now at risk of falling behind in the key technology of AI. "With the comprehensive regulation of basic AI models and AI applications, the AI Act is jeopardizing the ability to compete and innovate on both the manufacturer and user side," said BDI Managing Director Iris Plöger. The regulation is based on immature criteria that provide companies with less rather than more legal certainty.

Federal Consumer Protection Minister Steffi Lemke (Greens), on the other hand, sees the AI regulation as protection for consumers against the risks of the new technology. "During the negotiations, we were particularly committed to ensuring that AI systems are transparent, comprehensible and verifiable. In future, companies that offer the use of AI technologies will have to provide information about how their systems work and explain AI-based decisions," reported Lemke on Saturday. In the event of infringements, consumer associations could take legal action against this.

The European consumer protection organization Beuc criticized the EU for relying too much on the goodwill of companies to regulate themselves. "For example, virtual assistants or AI-controlled toys are not sufficiently regulated as they are not considered high-risk systems. Systems such as ChatGPT or Bard are also not given the necessary guardrails so that consumers can trust them," it said.

What happens now?

The EU member states and the European Parliament must first officially approve the project. However, this is considered a formality. The law should then apply two years after it comes into force.