INFORM Aims to Establish AI in Accordance with European Values

On June 14, the European Parliament adopted the AI Act. Now the vote with the EU Commission and European Council is starting, which is to be completed by the end of the year. If an agreement is found then, companies will have two years to adapt to the new rules. “It is true that there are reasons for regulation,” says Herbers. “However, I take a very critical view of what is happening at the EU level.” He goes on to explain why: The definitions of artificial intelligence and so-called high-risk AI systems are too vague, and the regulation goes far beyond learning systems with their specific risks and walks too close to regulating software in general. In addition, language models such as ChatGPT are to be subject to a type of regulation that will be impractical for providers. Herbers’ assessment: “This will lead to great legal uncertainty for European AI and will certainly not put the members of the EU and thus also Germany on the path to success.”

The fact that artificial intelligence is usually first associated with “risk” and “regulation” in Germany is not helpful, says Herbers. Regulations are important and necessary from an ethical point of view. “However, we also believe that we need to seize the opportunities that AI offers even more consistently. To put it simply, it would be nice if we built a stronger AI infrastructure in Europe and Germany that is worth regulating.”

“AI applications based on large language models will become a basic and infrastructure technology like the Internet once did,” says Herbers. “When it comes to infrastructure for cloud computing, mobile computing, or social media, we have largely left the field to the USA and China. However, we will only get infrastructures that comply with data protection regulations and are in line with European values ​​if we develop and offer these technologies and services ourselves.”

For Europe and Germany to catch up and then keep pace, their own infrastructure for large-language models is needed, for example. “The training of such models is extremely computationally time-consuming and expensive, but there are few companies in Europe that can afford such training. That’s why politics is needed here to create a framework and support initiatives for European language models.”

INFORM is not alone with Herbers‘ concrete demands on politics: The Aachen-based optimization specialist INFORM has joined forces with Germany’s AI Federal Association (KI-Bundesverband). With 400 companies, the association is considered the largest network for artificial intelligence (AI) in Germany. The association wants to “achieve digital self-determination and sovereignty in the European Union.” INFORM particularly supports the position paper that the Federal Association published in July. “We support the claim to make AI made in Germany a brand that stands internationally for quality, innovative power, and inventiveness,” emphasizes INFORM CEO Jörg Herbers.”