A German initiative plans to build a European AI infrastructure with a high-performance computing center in Germany to participate in the disruptive potential of artificial intelligence.
The feasibility study of the LEAM initiative (Large European AI Models) was prepared by the German AI Association on behalf of the German Federal Ministry for Economic Affairs and Climate Action. It deals with the construction of a high-performance computing center for artificial intelligence. The aim is for Germany and Europe to develop their own foundation models, which, according to the study team, will have a "disruptive effect on the economy and society".
For example, a well-known foundation language model is GPT-3, which gave rise to the Codex AI model and the currently popular ChatGPT. It is time to "catch up," the study says. Otherwise, Europe's digital sovereignty is at risk.
AI foundation models will lead to disruptive applications based on artificial intelligence. This development will enable many new applications, platforms and business models in almost all areas of the economy and society in the near future. LEAM provides a roadmap on how Germany can participate in this future technology.
U.S. and language models dominate the AI market
According to the study, the U.S. dominates the market for foundation AI models, with 73 percent of the models coming from the U.S. and 15 percent from China. Development is being driven by "large technology companies with billions in investments," the study team said, and there is a risk that a paradigm shift will be missed. The German AI Association speaks of a "monopoly-like cluster".
"Digital sovereignty in Al and downstream applications is acutely threatened, and with it the competitiveness of the entire German economy." And further: "The importance of foundation models will increase in the future."
Two-thirds of the AI companies surveyed in Germany are already working with foundation models, 71 percent of them with language models. Eighty percent of the companies surveyed support the development of large AI models based on European values.
Lack of computing power puts the brakes on Europe's AI ambitions
According to the study's authors, Germany and Europe need a "high-performance supercomputing infrastructure" to remain competitive, which is currently lacking in Germany. The study puts the cost of building such an infrastructure at 350 to 400 million euros, and classifies it as "feasible in terms of hardware and software".
This sum could provide the necessary infrastructure, including services, for the development of trusted open source foundation models that are in line with European values. Industry could develop their own AI applications based on these models.
This "AI Service Center" could be successfully established and operated by a joint initiative of business, science, and politics in Germany, the study team writes. As a beacon, it could help keep AI specialists in Germany and train new ones.
The LEAM initiative says it is in exchange with OpenGPT-X, Aleph Alpha and Huggingface. The full study can be downloaded from LEAM AI, but it's currently only available in German.