Darstellung von Vernetzung in Deutschland in blauer Farbe
Ruslam / stock.adobe.com
2022-07-01 publication

"We could achieve much more!”

It’s said time and time again that Germany is a world leader in research, but bad at “cashing in” on its innovations. Klaus-Robert Müller from the Berlin AI research center BIFOLD explains the truth behind this myth.

VDE dialog - the technology magazine
Prof. Klaus-Robert Müller, Berlin Institute for the Foundations of Learning and Data

Prof. Klaus-Robert Müller is a professor of machine learning at TU Berlin and co-director of the Berlin Institute for the Foundations of Learning and Data (BIFOLD), one of Germany’s six national competence centers for research on artificial intelligence.


How do you think German AI research compares internationally?
We can certainly hold our own. So far, many of the important ideas in AI have come from Germany. 

It’s often said that we’re world leaders in research, but not good at putting research into practice and making money from it.
Well, the first part of that sentence is incorrect. We can only be called world leaders in research in proportion to the resources available to us. We may be pleased that the previous government took up this issue in its AI strategy, but that shouldn’t obscure the fact that the funding amounts in other countries are quite different. I’m not even talking about the Chinese, who of course invest huge sums of money, or about the US, which also deals with the issue in a very different way. My colleagues in Montreal, for example, receive at least ten times as much funding as we do here in Berlin. That needs to be clear to everyone in Germany. If we were to receive an internationally competitive level of AI research funding, we could achieve much more! There are simply far too few core professorships in AI.

And the second part of the sentence?
That’s where I see the real problem. There’s naturally a huge demand for well-trained AI specialists in the industrial sector. But since there are hardly any people who teach the subject here in Germany, we also have far too few people completing their studies here. A top university in the United States might have a hundred AI professors, but here in Berlin, we have only a handful in this field. And that makes a difference.

So we end up with not enough people who can turn AI into lucrative innovations?
Exactly. Of course, it’s not like nothing is happening here. Berlin is famous for its many startups, and students from my chair have already founded companies that now have over 500 employees. We just need to train many more people. Then we’d see more promising results in this area, as well. However, I also see another problem of a very different sort here in Germany.

And that would be?
The risk tolerance of German investors. Most of them are about as bold and daring as a small-town bank and obviously haven’t understood the “venture” part of venture capital. There’s also the fact that the investment volumes are much lower than in the US, for example, and we don’t have the professional management structures needed for investors to support a startup in all the stages of the founding process. We therefore shouldn’t be surprised if people from Germany prefer to go to the US when they want to start a business. They’d like to establish unicorns here, but it just doesn’t work so well with the structures in place. BioNTech proved to be a laudable exception, but it worked precisely because the Strüngmann brothers took risks as investors.  

But you apparently don’t consider it to be a fundamental problem in Germany that there’s too strong a focus on science and research and not enough on transferring ideas into industry and business. Right?
Yes. For me, the whole theory-or-practice discussion that keeps surfacing is utter nonsense. Especially in AI, the difference between the two realms is a very small one. If Daimler introduces an innovation, it takes five – or maybe more like ten – years for it to be implemented in a new generation of vehicles. That’s how mechanical engineering works. But if there’s a new AI innovation at Amazon – meaning a new mathematical, algorithmic feature – it can be introduced within two weeks. This is precisely the difference between AI and conventional value creation processes! And precisely why the main thing we need at our universities is a solid theoretical basis. I’m of the same mind as the social psychologist Kurt Lewin, who purportedly said: “There is nothing more practical than a good theory."

The previous government made it a priority to create a coordination unit for the centers of the German AI network. Was there concern that the centers would otherwise work redundantly in parallel?
First of all, cooperating with the other centers isn’t something the politicians have to tell us to do. We do it already, simply because good people work at these institutions and it makes sense in terms of the nature of what we do. That said, we obviously also want to cooperate just as closely with good people in Tokyo, Seoul, Paris, Stanford or Los Angeles. When solving a scientific problem, it doesn’t matter where your partners are.

The government was more concerned with the coordination within Germany, though. Or do you not see the danger of investing a lot of money and brainpower in a project in Berlin, for example, and then learning that a team in Munich is working on exactly the same topics?
For one thing, we’re well aware of what our colleagues are doing because we talk to each other. For another, though, it’s perfectly acceptable for research to be conducted on the same topics at different locations. The scientific competition that comes with wanting to be the first to make a breakthrough is a positive thing! Competition keeps us energized at BIFOLD, as well, and that’s exactly what leads to further cooperation; people realize that, together, we might be able to achieve the next breakthrough sooner. Sometimes everyone joins forces, and sometimes it happens in groups. But the process is never a waste of brainpower or investment. On the contrary: it’s the only way to create new knowledge.

Interview: Martin Schmitz-Kuhl