Was müssen wir jetzt tun? Wrap-Up von Tanja Rückert und Wolfgang Wahlster

So the task you now have, I don't envy you, because you have the task of drawing a conclusion from all that we've heard today, and that's a lot. Wolfgang Wahlster, may I ask you to come up, and Tanja Rückert, welcome. Please, come up.

Yes, I need to structure this a bit. Let me start: the focus was very much on bringing science and business closer together. That was clearly stated at the beginning, and I really took away that we are on a good track—I knew that before, of course, because I've always practiced it myself. What became even clearer to me is the strong consensus. I found it great that this event didn't deal with conflicts of the past but really looked to the future and new things. For instance, machine learning absolutely convinced me that all speakers, starting with Mr. Kersting, Mr. Grübel—actually all of them—pulled in the same direction, saying that machine learning is our flagship, but we must connect it. They talked about distributed learning and the next stage of machine learning where you integrate it into conversation with a human—learning by being told. And Ms. Sczech said at the beginning, to my surprise, that everyone always talks about learning from mass data, but that's a wave; if you can already access that as a commodity from Google, it won't go well for long. We have to continue our own creation, and now often try to do it with very small data sets. What I also liked very much is that there is general consensus: we invented Industry 4.0 here, we had a platform, the federal government set it up very quickly back in 2011 when we started, and it's a worldwide success. Now connecting this topic of AI and learning systems with Industry 4.0—that's outstanding. Soon we'll have a congress in China on the topic; Mr. Darlegen from acatech is leading it. We already did that in New York, so there's enormous demand, and that was shown again today—we really have a lead of two to three years in linking Industry 4.0. What I miss—I want to mention something negative—we are very strong in Germany in the application area of retail, as Mr. Riet said, in English and logistics. You mustn't forget that the Schwarz Group, companies like Lidl, Kaufland, or Aldi—people laugh about it, but these are companies that are truly world leaders in retail and mail order, like Otto. So I think we should also look at what's there; you could talk about it all day. And logistics is very strongly tied to it; intralogistics is becoming stronger. That was missing a bit, but I think this focus is good. We are not lagging behind Google in online retail or advertising; that's not our thing. Our thing is to create physical products, and with Ms. Rückert we have someone from the right company for that. Ms. Rückert heads the Bosch Building Technologies business unit; she was at SAP before, so she knows both the user and development sides—the physical and the IoT. So that's exactly the point.

What do you take away from this event? What are your learnings today?

So we've heard the views of research, science, and the economy. But how should we conduct the societal debate on AI? What is important?

That was actually already pointed out in the opening statements—I think Ms. Carlitscheck and Mr. Wahlster emphasized it again: we need to take away this fear of loss of control. That means we must have explainability components in the systems to create transparency. That is still partly a research topic; especially in deep learning, there's still a lot lacking. There are initial approaches coming from Germany, but it's not perfect. In model-based systems, it works quite well, so we need to do something there. Especially, explanations must give the end user the certainty that what the system outputs can be passed on with good conscience to their clients or patients. I believe that's a common thread running through all the presentations—there is absolute consensus that transparency is absolutely necessary. The second thing is certifiability. As a colleague from Adam said, that's partly extremely difficult because the very claim of artificial intelligence also applies to situations that are dominated by incomplete information. A truly verifiable correctness doesn't exist there; a residue of probability will always remain. Otherwise, we'd be on the other side. As was nicely said: where it is possible, for example with hardware systems, we want the proof—like the airbag from Bosch must work, the computer from Infineon must compute correctly, the sensor must function. So you need to differentiate, and the FKG (Fraunhofer?) is making great contributions there. That was great. What now needs to be done, and this was also brought up at the end, is ethics by design. We have already prepared that term in the data ethics commission for the final report—similar to security by design. It's completely wrong to first build the system and then think about ethics; it should be truly integrated. That also came out nicely; several people demanded it, but no one knows exactly how to do it. There are too few people who can do it professionally. But I think the movement is clearly visible here. And I think this collaboration—that we are all of the same opinion, one might say maybe that's bad, but it's also a consequence of the work in the platform. Because we said this morning: the platform is not a debating club that meets once a year; it's hard work. acatech is a working academy in the literal sense; people work hard on things. The boards you've seen don't get papers automatically from heaven; they are written by people. So I think we are on a good path. And I must say, despite all the pessimism sometimes spread, especially by American consulting firms—if you listen to them, you'd think Germany is in a very bad spot. But I see that we cannot rest; we must continue to work hard, but it's clear that the intertwining with industry here is extremely good. I think there is no country in Europe that is so admired for the way research and industry work together, not least thanks to the Fraunhofer Society and its centers. If you had said that 40 years ago, such close cooperation would have been impossible. And I think that's our asset: real trust, not like in the US where there's only sponsorship—scientists are sponsored by large companies, get gifts and can do as they please. In Germany, industry takes it seriously and says, 'I rely on this; I want to develop something with the professor and the institute, and it should end up in my products.' That's our great chance. Ms. Carlitscheck said: when we win people from abroad, we don't have as much to offer in terms of salaries, but we have these physical objects. That makes young people more excited—not optimizing some ad, but actually building a car or an autonomous ship. The second thing is this close connection; we don't research for the filing cabinet but create real engineering outcomes, which attracts people. Also with the Alexander von Humboldt Foundation's idea—that's excellent. I've been a reviewer there for years; we can bring back top talent to Germany. So I think the program is excellent. To supplement Professor Wahlster: it's always a challenge, but what I want to say is that the Platform Learning Systems has an ethics working group. That's a sign. I used to be a coach at the Industrial Internet Consortium, which was more US-driven; no one there thought about an ethics working group. So that's a strength of ours. What got me thinking is that now companies in China are bringing this up. I'll probably mull over that again tonight, because it's a sign that they may have seen something we haven't. I also liked that in the last panel they pointed out the GDPR from Ms. Hessen—such things are easily forgotten; we are already protected. That categorization, as Mr. Druck said, helps us think in boxes; it helps us set guardrails. So the wish is: help yourselves with that and try it out in your company, because it's not just a budget exercise—we think it will hopefully advance everyone. So I believe we heard a lot of good things today. China concerns me; the focus on ethics in the Platform Learning Systems is good. I think we need to implement what we already have—the categories and guidelines—in every company as much as possible.

With this constructive appeal, I think we should end here, or do you have any additions?

Everything is good. Perhaps one more thing: what is missing—few people addressed it—is the infrastructure, the technical infrastructure. In Germany, we have long relied on pure supercomputing, classical numerical supercomputing, which is good for weather forecasting, climate, physics, but we need a different computing structure. Mr. Streibich mentioned it: we need GPU-based computing, sensor-based computing, neuromorphic computing. That's a weak point. We absolutely need to offer that to our researchers, especially to attract top talent from abroad. They sit in the lab with a beautiful application but don't have the right equipment. We need to make a shift. Those systems are expensive and mostly come from the US, but we have to invest. And we should also try to develop something ourselves; we have good companies. The second thing: we need to combine our excellent sensor technology—that was also missing today. Apart from large companies like Siemens and Bosch, there are many small and medium-sized enterprises in Germany that are excellent, like SICK in laser sensors. We are leaders there. We need to bring them into AI—we are doing too little of that currently. If we connect that with our managed computing, then we are on the winning track. That's the era of distributed AI, which we absolutely need for logistics and Industry 4.0. For a big advertising campaign, maybe not, but that's not our thing. So I think that's still needed, but I'm optimistic. I just want to set a positive concluding note: I think the motto at the beginning from Heinz Streibich was also 'don't despair'—though he didn't say it exactly—but we have the things, we have real products. We can strengthen them: reliable, robust AI. So I'd like to end with the little motto: don't despair, and let's all move forward together as a platform as quickly as possible.

Thank you very much. I give my heartfelt thanks to both of you. Thank you. Thank you, Ms. Rückert. Thank you.