The Munich University of Applied Sciences is considered a pioneer in the distribution of tools based on generative artificial intelligence (AI) to researchers and students. A year ago, the university made the “new miracle tool” ChatGPT (generative pre-trained transformer) available for writing term papers and exam papers – and for free. Then, on July 30, 2024, came the shock: the university closed access to ChatGPT. The reason: it became too expensive to offer the AI for free.
The emergency stop in Munich is indicative of the way large language models (LLM) are handled at Germany’s universities. So far, there are many small projects, but no nationwide low-threshold access to LLM for students. According to a request, the German Rectors’ Conference has no overview. The Federal Ministry of Education and Research (BMBF) does not feel responsible, a spokeswoman said. And the lively AI Campus, a learning platform for artificial intelligence, does offer a variety of free training courses and collaborative workshops. But the project, which is funded by the BMBF and developed by the Stifterverband and numerous other partners, cannot provide its own AI access.
AI pioneers in Lower Saxony
Universities are far from providing their almost 3.5 million researchers and students with basic generative AI. What academia is drawing on are three projects in Lower Saxony, as research for research and teaching shows. They are currently the most important addresses for AI access for German universities: the “AI Service Center for Sensitive and Critical Infrastructures”, KISSKI for short, in Göttingen; the HAWK art college in Hildesheim, Holzminden and Göttingen; and finally a private model called Educa-AI, which is de facto a spin-off from the universities of Göttingen and Clausthal-Zellerfeld. In short: the Chief Information Officers (CIO) of German universities are all in Lower Saxony.
These three projects offer direct (sometimes free) access to various LLMs (KISSKI), implementable interfaces (HAWK) or embedding in learning management systems such as Moodle (Educa) for the whole of Germany. All other projects, such as the attempt to build a bwGPT for Baden-Württemberg, the BayernGPT or the AI project “KI:edu.nrw” based in Bochum, either cooperate with Lower Saxony or are still a long way from actually usable AI access.
A look into the laboratories in Lower Saxony reveals a German dilemma: There are quite a few AI pioneers at universities. But the university’s organization, which is dominated by committees, is slowing down the absorption of the technology.
One prototype replaces 100 committee meetings
“One prototype replaces 100 committee meetings,” says Stefan Wölwer from the Design Department at the University of Applied Sciences and Arts (HAWK), describing the appeal of “just doing it.” It is the basic idea of the “Interaction Design Lab” that Wölwer founded. It is there that the AI interface was created that any university can copy.
At the same time, employees of all three pilot projects are observing slowing effects: “Everything just has to go through the bureaucratic process,” says one of the AI employees. “People aren’t making any progress, even though they’ve been working on the implementation for three or four months – because so many committees have to approve it first.”
Julian Kunkel heads the KISSKI project, which is run by the High Performance Computing Center at the University of Göttingen and the Max Planck Society. “We had a prototype pretty quickly – but it took half a year until the data protection and IT security issues were resolved,” reports the professor of high performance computing. “Universities are typically slow to adapt new technologies because there are many bodies and complicated control systems.”
“We had a prototype pretty quickly – but it took half a year until the data protection and IT security issues were resolved.”
Professor Julian Kunkel, KISSKI project leader
All of the AI pioneers interviewed have two minds. They believe it is important to involve all members of the universities through their bodies. But they also point out that the rapid pace of AI development does not take committees into account. “None of us expected that the performance of the large language models would increase so quickly,” says Julian Kunkel, who heads the largest of the AI projects. Large language models, initially seen as a technical gimmick, are now considered critical infrastructure for universities in the country of the world export champion.
13,000 versus 100 million unique users
No other digital tool has reached as many people as quickly as ChatGPT – 100 million in three months. At German universities, the process is more like a long, calm river. KISSKI has 130 institutional customers, including 50 universities, the rest are colleges and research institutes. HAWK, in turn, has over 30 colleges that have downloaded its interface from GitHub. They also come from all over Germany.
If you take the total numbers as a benchmark, you can see how far there is still to go. KISSKI, for example, has registered 13,000 individual users – out of three million students. Students therefore have to buy the new technology themselves. Two thirds of students say in Surveysthat they already use the tool for their studies. They present themselves as law-abiding citizens. In personal conversations it becomes clear that they use the fixed text generators for pretty much every writing and research process. At many universities there are no rules at all, say four out of ten students.
“Generative AI is also changing a lot in terms of teaching.”
Vincent Timm, Leiter Interaction Lab, HAWK
Vincent Timm knows how problematic this is. He heads the Interaction Lab at HAWK. In universities, AI is breaking down all barriers. Learning formats, exams, technology – everything is being turned upside down. “Generative AI is also changing a lot in terms of teaching,” says Timm, who mainly deals with AI after work. Everyone knows that the forms of exams have to change. But no one knows exactly how. Timm warns against resting on the laurels of writing AI guidelines. The positive and critical consequences of the introduction, he says, “can only be discovered in the process, when the language models are actually implemented and put into practice. Just talking about it theoretically – that makes no sense.”
BMBF: “Digital University” programme cancelled
Meanwhile, politicians do not give the impression that the issue is a priority. Finance Minister Christian Lindner (FDP) has just said that he wants to spend more money on education and digitization. He has evidently not reached his party colleague and Science Minister Bettina Stark-Watzinger with this. A federal “Digital University” program, promised in the coalition agreement, has been canceled. Stark-Watzinger’s AI action plan also contains no reference to making generative AI available to all researchers and students – not to mention a large research and education LLM with curated content.
Vincent Timm makes it clear how important the question of money is in generative AI. “No university in Germany, except perhaps the Technical University of Munich or the RWTH Aachen, has the financial means to give researchers and students access to this critical technology.”
Florian Rampelt from the AI Campus is an ever-cheerful manager of university digitization networks. The university’s sloppiness with the key technology LLM makes him think. “It’s a question of resources,” says Rampelt. “If we finally want to let the republic’s creative pool, namely three million students, play and experiment with large language models, then we have to invest money – and release the tools.”
The Munich University of Applied Sciences (HSM) will reopen access to large language models for its students after the semester break. However, they will then ask students to only confront ChatGPT & Co with academic questions. Anyone who does not comply will be throttled.
And where does the HSM get its interface from? From Vincent Timm, one of the after-work CIOs of German universities.
