No. 114/2022

You do basic research, but your results are also used in business. How does that col- laboration work? We develop basic methods. And, in AI, they are often tested using very realistic data, for example from business. The resulting software is usually published open source, not least for the sake of scientific reproducibility but also to make it easier for others to build on it. So, industry can apply our results really quickly. Do you get research funding from industry? Sometimes, yes. But for us, cooperation is less about money and more about observing how our methods stand up when faced with solving real-world application problems. Apart from which, I have the generous endowment that comes with the Alexander von Humboldt Professorship. The problem at the moment is not that we don’t have enough money, but that we have to recruit people with the necessary talent for research. And the market is already pretty tight. Because you compete with industry? It does happen that doctoral students from my group go on to work in industry. There is simply a lot of interest in our topics and in well-trained experts. Of course, I’m “WE NEED A CERN FOR AI IN EUROPE” Human-centred AI could constitute a huge locational advantage for Germany and Europe, says Alexander von Humboldt Professor for AI, Holger Hoos. A conversation about battling industry for talents, sustainable AI and what Europe can learn from Canada. KOSMOS: Professor Hoos, you campaign for human-centred AI. What does that mean exactly? HOLGER HOOS: AI should help us humans solve problems that we can’t manage to solve without AI. In medicine, there are many applications where this is the case. And when it comes to climate change, we are also dependent on AI to understand the atmosphere and the oceans better. But the ethical foundations of AI are also very important. We must ensure that using AI does not lead to a loss of privacy or the construction of lethal autonomous weapon systems. In short, I’m interested in AI for the benefit of people, which also means making AI itself more sustainable. How do you want to go about it? Data centres already account for a good deal of CO₂ emis- sions. And the trend is upwards ... Exactly. My research also looks at how we can make machine learning more efficient: achieve similarly good results using less data and fewer calculations to lower CO₂ emissions. Our goal is to make savings of between 50 and 90 percent. This would also cut down on the hardware required and thus reduce the number of resources consumed in manufacturing semiconductors. PROFESSOR DR HOLGER HOOS is a computer scientist and, as of 2022, Humboldt Professor for AI at RWTH Aachen University. He previously conducted research in Canada and the Netherlands. Photo: Elodie Burrillon › 21 HUMBOLDT KOSMOS 114/2022