Is Germany currently being left behind in artificial intelligence? No, says Wolfgang Wahlster, head of the largest AI research center in Germany, in a FAZ.NET interview – and explains what we do differently from America and China.
Professor Wahlster, are we in Germany and Europe already left behind when it comes to artificial intelligence?
Not at all. We can face the competition head-on. DFKI in Germany, with its 900 employees, is still the world’s largest research center in this field as a PPP with over 80 spin-off companies. Let me give you two simple examples: The American company Nvidia is currently building the fastest GPU computers, with which neural networks for deep learning can be calculated 100 times faster than with conventional computers. Nvidia has made this computer available free of charge to selected institutions, including top American universities such as Stanford in Germany and the German Research Center for Artificial Intelligence. The head of Nvidia handed over the computer along with annual research funds with the conviction that the best AI research institutions in the world should have the best tools.
And the second example?
Besides many Dax companies, we also have American companies such as Intel, Google and Microsoft as shareholders at DFKI and also a branch office in Beijing. We had to carry out another capital increase because six large global companies also want to become shareholders in our company this year. The great interest in working with German AI researchers clearly shows that we even have a lead in some areas of AI. DFKI is a career springboard and more than 20 former employees are now with Google, for example, but we have also been able to recruit Google employees for DFKI.
Can you honestly compete with the salaries and opportunities offered by a billion dollar company like Google?
Yes, it’s not about the absolute pay, it’s about the standard of living that you can afford with the salary in the area. For example, an excellent Google Deep Learning employee moved from Silicon Valley to DFKI, even though he earned more than three times what he now receives from us. In the United States, however, he would have had to invest a million dollars to educate his children, whereas in Germany this is free of charge. President Trump also irritates many scientists, so I also receive applications from Silicon Valley every week.
When it comes to machine learning based on the processing of huge amounts of data, however, we in Europe and Germany cannot currently keep up with America or even China. After all, we don’t have any companies like Google or Baidu or Tencent.
This is true in the area of evaluating consumer data for advertising purposes. We in Europe are rightly rather reserved in this area. I myself am not registered on Facebook and am not prepared to provide my personal data for advertising purposes. In Germany, however, we pursue a different approach.
We rely on AI applications for the refinement of our products manufactured in Germany and for disruptive business models through smart services on these product platforms. We are upgrading Germany’s export hits with artificial intelligence, in our machine tools as the world’s factory supplier, in medical technology, in our cars, agricultural machinery and high-quality household appliances. Siemens, for example, is the world market leader for high-performance scanners in medicine, and image interpretation using AI systems is a key success factor in clinical practice. In AI procedures for autonomous driving, German companies have considerably more patents than the Americans and Chinese.
And we rely on Industry 4.0, which you have coined as a term.
Yes, we are developing artificial intelligence algorithms for the next stage of Industry 4.0. There are thousands of sensors that will be evaluated in a factory in the future. There is enormous potential for higher product quality, versatile production and higher productivity if we use collaborative AI robots, machine learning, augmented reality and AI-based production planning systems as assistance systems for our skilled workers. And a huge opportunity: Here, by the way, we have a lead of two to three years over China and the United States.
This is all very much geared to concrete applications. What about basic AI research and especially machine learning in Europe?
I think that the distinction between basic research and application-oriented research is an absurd division in such a rapidly developing field. We must think of basic research and its transfer into concrete applications together. We need the timely transfer of knowledge into practice. At DFKI, our researchers rotate between basic research, application-oriented development and then transfer with our more than 100 industrial partners. Decades of basic research, as in high-energy physics, is not possible in computer science. It has to be done quickly, three years maximum from idea to product, not ten.
Bernhard Schölkopf, who researches machine learning at the Max Planck Institute, makes this distinction. Together with other top European researchers, he is concerned that we will no longer be competitive in basic research if we do not do more.
In recent years, many new funding opportunities for basic researchers have been created in Europe and especially in Germany, such as the clusters of excellence or the ERC grants. The BMBF has just selected additional centres for machine learning in Germany for funding. I cannot see that good top researchers do not have enough opportunities to get their research funded. There has never been so much additional funding for research as in recent years. Nor should you forget that machine learning is only one aspect of artificial intelligence and that computer-adequate knowledge representation, machine inference and automatic planning are absolutely necessary as methodological foundations for developing complete cognitive systems. Many of the foundations in machine learning were laid by European researchers and then transferred into products by American companies. We must ensure that even more transfer of practical experience takes place in Europe – also to small and medium-sized enterprises.
However, machine learning is currently seen as a very important component.
Yes, but it is the only way to develop artificial intelligence. All the knowledge that physicists, chemists and biologists have created over the past centuries does not have to be learnt from empirical data. We can directly contribute this knowledge through computer-adequate knowledge representation and the software can derive new conclusions and problem solutions from it.
The coalition agreement includes a Franco-German research centre for artificial intelligence. Will you tell us if that will dock you at your institute?
That is of course decided by politics as a donor, certainly together with industry. But we are preparing for it. I assume that we will play an appropriate role in this cooperation. We already employ numerous French researchers and have for many years carried out more than 20 collaborations with our French partner institute Inria, among others within the framework of the EIT. President Macron has entrusted Inria with AI coordination in France. Like us at DFKI, Inria combines cutting-edge research with application in transfer.