In a position paper on the Hochschulforum Digitalisierung blog, Weßels, Maibaum, Bremer and Meyer argue that universities, which plan linearly, risk underestimating the fast, nonlinear growth of generative and self-improving AI agents. They recommend more experimental, adaptive and human-supervised organisational development. The paper presents no new data, and its central growth claim cites no benchmark.
How this is computed
- Key claims supported: 3 of 6
- Type of source: Institution on its own work
- Independent confirmation: None
Rates the source, not the topic and not our analysis.
Why it matters
Our reading: The paper illustrates how part of the German higher-education debate frames AI as a strategic planning problem rather than only a teaching tool. Because its key premises rest on a podcast, presentation graphics and the authors’ own earlier article, it works as a call for organisational change, not as evidence that such change improves outcomes.
What the source says
- Unverifiable The authors say generative AI shows disproportionately strong performance growth over time and that systems change their capabilities unpredictably within months. They give no metric or data series for this.
- Conditionally supported According to the authors, citing a Sakana AI web page, the Darwin-Gödel Machine by Sakana AI and the University of British Columbia lets an agent modify its own code, test variants and keep successful versions. They describe this as a move of self-improving AI from theory into application since 2025.
- Supported The authors say semi-autonomous agent systems can accelerate research, citing a peer-reviewed Nature article by Ghareeb et al. on automating scientific discovery. The paper does not report how large that acceleration was.
- Conditionally supported The authors point to a dedicated workshop at ICLR 2026 as a sign that self-improving AI is no longer a niche topic.
- Vendor claim only Citing their own magazine article in Forschung & Lehre, the authors attribute concerns at universities to insufficient AI training for teachers, outdated didactic concepts and missing approaches for integrating AI into teaching.
- Unverifiable The authors state that Arizona State University shows adaptability makes a university more attractive to applicants. The only related source they cite is a presentation graphic.
The key caveat
This is an opinion piece, not a study. It reports no sample, survey or measured effect on learning or institutions. Its causal claims about universities and applicants are not backed by empirical data.
Open questions
- What measure and data would substantiate the claimed nonlinear performance growth of generative AI?
- Do applicant or enrolment figures from Arizona State University, compared with other universities, support a link between adaptability and attractiveness?
- Have the Darwin-Gödel Machine results been independently replicated beyond the company’s own publication?
Source: Hochschulforum Digitalisierung, 14 September 2026: Jenseits des linearen Denkens: Sich selbst verbessernde KI als Herausforderung für eine zukunftsorientierte Hochschulbildung
Produced by Zeitgeist’s AI agents (research, editing, standards check) with Claude. No human review.