DAILOGUES: What makes a successful AI location?

Philipp Gerbert: A successful location produces organizations that help determine the direction of AI and shape the transfer of technology into the real world. This requires successful interaction between science, business, and politics. If we look at Boston or Silicon Valley in the US, we can observe this kind of cooperation in action.

DAILOGUES: How can knowledge transfer from science to business and government succeed?

Philipp Gerbert: There are two approaches here. There is the possibility of solving existing problems in industry or government with new technologies. To put it simply, the Fraunhofer Institute in Germany, for instance, is geared towards this as it employs researchers who are closely connected to science and who address contracts from industry or government with new research. It is another matter to bring ideas for new companies to the world as this requires the commitment of scientists, especially doctoral students, postdocs, or professors (the latter often still want to advance first-class research). I believe that in Germany, the company BioNTech has shown that cutting-edge research is also possible outside of universities. I would even go as far as to say that BioNTech's research would hardly have been possible at an academic institute. In AI, we have now reached this point almost everywhere. We need more such role models who inspire the next generation of scientists to follow BioNTech's example (or even DeepMind or OpenAI).

DAILOGUES: It sounds a bit like fresh wind only comes from new companies with innovative people and ideas.

Philipp Gerbert: In times of radical change like the present, that's probably correct. If we look at the most successful companies in the US, each of them is significantly younger than the top 10 companies in Germany.

DAILOGUES: Geoffrey Hinton said in an interview that during his time at Google, strong language models already existed. This was before OpenAI went public with ChatGPT. However, according to Hinton, Google held back on its models, fearing that errors could damage the company's reputation. He said OpenAI had nothing to lose and therefore dared to release ChatGPT.

Philipp Gerbert: It's always difficult to introduce something new that doesn't fit into the current business model – large language models, for example, challenge traditional search engines. The telecommunications industry could have introduced a service similar to WhatsApp in the 2000s, but in doing so would have wiped out billions in revenue from text messaging. Whether a service like WhatsApp would have been profitable for them remains unclear to this day.

DAILOGUES: The Technical University of Munich (TUM), with its numerous spin-offs, demonstrates how successful knowledge transfer can be achieved. The university has significantly improved its position in international rankings (e.g., QS World) and is now considered one of the best universities in the world. What is Munich's recipe for success?

Philipp Gerbert: The Technical University of Munich had a strong drive to build something. The driving force behind this was its former president, Wolfgang Herrmann, who, in close collaboration with the state of Bavaria, focused entirely on technology and management. To this end, major new institutes were created at the university. TUM also did something unusual: It placed entrepreneurship as a third pillar alongside research and teaching. Even though this pillar is not explicitly reflected in international rankings, it contributes significantly to the university's success by attracting talent. One expression of the entrepreneurial spirit at TUM is the numerous student initiatives in the fields of aerospace, automotive, or AI. Every competition Elon Musk has organized to date has been won by students at TUM. Many of these students are also motivated to found a startup after completing their studies.

DAILOGUES: How can we position the TUM Venture Labs in this context?

Philipp Gerbert: The TUM Venture Labs is a non-profit organization dedicated to scaling the transfer of science and research to the world. Their focus is on deep tech and life sciences and their goal is to foster the founding of new startups that can contribute to the world's major challenges, such as the climate crisis, with technologically driven developments. We want to foster the next generation of founders. Even though they are based at TUM, they don't necessarily have to study or work at the university – in many areas, we support founders from the entire DACH region. The goal of the TUM Venture Labs is also to contribute to European sovereignty.

DAILOGUES: Are Europeans still reluctant to make risky investments in start-ups? Or are we seeing a reversal of this trend?

Philipp Gerbert: I would say that the venture capital market in the deep tech sector for early-stage startups in Europe is healthy, even if there is about 50% less funding available for young companies at the start than, for example, in the US. The main problem in Europe is that there is no competitive, unified capital market. The lack of exit options puts pressure on late-stage investments, that is, financing rounds between €100 million and €1 billion. It certainly doesn't help that pension funds in Europe traditionally hardly ever invest in young companies. American pension funds hold approximately 15% of the shares in German startups, while German pension funds hold around 0.2%. Even large financial institutions in Europe still sometimes lack the right tools to inject capital into startups. In the deep tech sector, this is increasingly being offset by the involvement of international investors from Asia, the Middle East, or North America. I would therefore argue that the real bottleneck is the limited number of strong entrepreneurs.

DAILOGUES: We spoke with the founders of Lapis for www.dailogues.ai. Lapis' CEO, Pietro Arina, explained that there's a different culture of entrepreneurship in Europe. Startups in Europe tend to grow organically rather than rushing to the next stage. Furthermore, there are also fewer college dropouts among European founders. Do you share this assessment?

Philipp Gerbert: I would agree, but I think we are currently experiencing a cultural shift. While starting a tech company has long been an attractive option in the US, we are now also seeing in Europe that many young people are considering starting their own business. They are realizing that this allows them to generate impact that is not just about getting rich quickly. I would say 70-80% of the founders we support through the TUM Venture Labs have a purpose. They want to change the world with their companies and technologies. Of course, founding a company remains a difficult task and takes a lot of time and energy. A fertile ecosystem like Munich and institutions like the TUM Venture Labs can provide crucial support in this process.

DAILOGUES: What could successful cooperation between an existing corporation and a university in the field of AI look like?

Philipp Gerbert: For example, SAP and TUM have been successfully collaborating for several years. SAP has built a research laboratory directly on TUM's Garching campus. Researchers from TUM are located on the first floor, with SAP employees on the floor above. In addition, the company has more than a dozen AI projects integrated into TUM's chairs. The university has similar collaborations with Siemens and BMW.

DAILOGUES: What trends in AI do you predict?

Philipp Gerbert: I see the biggest AI trends in the coming years in three areas: First, generative AI is rapidly evolving toward so-called agentic systems – that is, AI that reflects, plans, uses tools, and acts in multi-agent environments. Second, AI will increasingly learn through direct interaction with the real world – no longer only via language, but through direct signals from the world, independent action, and feedback, similar to AlphaZero in chess. Third, cybersecurity is becoming increasingly important, as AI-controlled action is becoming a new point of attack and AI is moving extremely quickly. In the future, this security can only be guaranteed by AI itself.

We thank Philipp Gerbert for the DAILOGUE.