Since 2001, the Berlin-based Fraunhofer Institute for Open Communication Systems (FOKUS) has been researching and developing the networked world. How AI is used and what strategic potential lies behind machine learning was discussed by #KI-Berlin and Prof. Dr. Manfred Hauswirth, Managing Director of the Institute.
What role does artificial intelligence play at Fraunhofer FOKUS?
We are not a classical AI institute, but - and this is a very big but - we use it out of the domains. These include networked driving, eGovernment and also Industry 4.0. In these areas, questions repeatedly arise in which we have to use and apply AI. AI is not a panacea, this must be said out loud. In some areas, however, there are good applications in which it is an extremely useful tool.
Können Sie da Beispiele nennen?
At FOKUS, we have been working with the „Daimler Center for Automotive Information Technology Innovations (DCAITI)“ for just over ten years. In close cooperation, we use state-of-the-art AI methods to develop autonomous driving, which we are now even testing in various Berlin test beds on the road. In our test vehicles, for example, which are equipped with cameras and laser scanners, we use AI technologies to mechanically perceive the environment. Among other things, this enables us to "experience" accurate, self-updating map material. We also wish to design networked driving in such a way that not only do several vehicles "talk" to each other, but they can also cooperate with non-vehicles such as pedestrians, cyclists or others. This necessitates foresighted driving. This is still a challenge for AI!
In which other areas are AI solutions used?
A Norwegian company for nautical measuring instruments approached us to support its customers with measuring systems for salmon farms. Up to one million fish swim in a net there. Presumably it works like the merchandise exchange in the USA. We then receive an inquiry: we need one million salmon with such and such a weight. But how do you find the net in which most fish with the right weight swim? Until now, the salmon farm has been faced with the problem of not being able to make precise statements about the weight of the fish in the nets. We have used our many years of expertise in image recognition to estimate how large and heavy the salmon are in the nets with extremely poor visibility. This allowed us to optimise the selection of salmon with the desired weight. This "optical weighing machine for fish" is now being sold as a product.
From networked vehicles to salmon detection – the range seems wide ...
Yes, not to mention medical applications. We use machine learning algorithms for the treatment of cancer cells. This involves focusing ultrasound on the tumour cells in an organ. A similar approach is used for retinal damage caused by circulatory disorders. This is extremely difficult to detect. AI is needed because physicians can be successful only with a great amount of experience. Here we see many cases in which AI is used: it is intended to reduce the number of routine activities without completely replacing humans. This is also possible in the day-to-day work of authorities. There are numerous routine tasks here from which administrative staff can be relieved with AI. For example, an intelligent input management system can automatically assign incoming documents to the correct files and processes. Learning virtual assistants can help citizens in their dealings with the authorities. We have recently developed a prototype language assistant which allows citizens to apply for parental benefits. The situation is more difficult with discretionary decisions in which the administrative employee has his or her own room for manoeuvre. At present, AI can only have a supportive effect here, and even then the employee must understand what the machine is actually doing and how it reaches certain conclusions. AI still has its problems today with this transparency. From my point of view, AI is just another tool. Here, as with Big Data or 5G, you have to lower the expectation and the hype a bit. And with what is left over, Europe will not fall by the wayside either.
What do you mean by that?
We in Europe must move away from the "mouse at the mercy of the snake" attitude and change from AI users to AI developers. It is not true to say that Amazon and Co. will leave us behind. The US corporations don't have the domain knowledge that we have. It gives us an incredible strategic advantage. We have world market leaders here in Europe and should not let ourselves be discouraged. It hurts my heart when VW enters into a partnership with Amazon. We have the same competence in Europe. I don't see China as a competitor either. The only advantage there is access to data. Data are extremely important, without data you cannot operate AI. But they should not be centralized as much as in large American corporations. In Europe, we must consider how we can use our potential strategically - with distributed systems which comply with data protection requirements and are sustainable.
You talk about a different approach in Europe. How can this be achieved?
Firstly, as I said, AI is not a panacea. The benefits must be carefully analysed before use. In many areas you don't need AI. When it comes to process control in production systems, for example, our colleagues have been doing this excellently with hardware for 40 years. The guiding principle of computer science is "Never change a running system". Secondly, you can also make a system worse. The more complex a system becomes - and through the use of AI it becomes more complex - the more possibilities for errors occur. That's why it's important to weigh things up carefully beforehand. In addition, there are many open questions in AI research which have a high economic potential.
What are these?
One possible question is: How do we get away from centralization? In the human body, a comparatively large amount of information about the nervous system is filtered out before it penetrates the brain. There is an eye nerve disease in which everything reaches the brain unfiltered - people go insane. The same applies to AI. The question of distribution would also ensure response times in real time. To do this, the AI system must be no more than 50 km away from the data source. So you need a distributed infrastructure. You have to build this. However, there are at present no edge clouds (note: clouds which process data streams on the spot in a resource-saving manner) that can be touched. To change this, we are cooperating with German Edge Cloud (GEC), a start-up of the Friedhelm Loh Group. At this year's Hannover Fair, we showed together with GEC a robot controller that runs on an Edge-based 5G network. These are the exciting topics with economic potential. Why do language assistants have to transfer all human biodata into the cloud? Only because the operators benefit from it. Here alternative concepts are necessary.
What do we need to make these alternative concepts bear fruit?
We must give massive support to research and the transfer of know-how. This does not mean that we should expand research everywhere, but rather in a very targeted way. We need fast funding. Speed is the trick. If you have to wait three years for EU funds, the horse has bolted. When I hear that the German Federal Government is investing 500 million euro in AI, that is far too little. The MIT alone, for example, is investing one billion dollars. And that is "only" one university. At the same time, however, this shows our efficiency: we achieve a lot in science and research with little funding. However, we now have to transfer this to industry. Nevertheless, more funding is needed, which must flow quickly into the science system and into technology transfer.
Wie gelingt es mehr Fördergelder für den Technologiebereich zu akquirieren?
How can more funding be acquired for the technology sector?
Firstly, as scientists we must approach industry and explain what added value they can have through technologies - beyond the promises of a cure. At Fraunhofer, which is a non-profit organization financed one-third by taxpayers' money, this running is also our task. We have now celebrated our 70th anniversary, so we have been around for quite some time. With 26,600 employees and an annual budget of 2.6 billion euro, we are the largest research organization for applied research in Europe. We have know-how in many areas and combine the expertise of several institutes in a variety of issues. We do the same in Berlin, for example with the "Digital Networking" performance centre. In the automotive and steel industries, it has become clear that Fraunhofer can be the outsourced research institution for companies. But we also work closely with SMEs and start-ups. We try to overcome initial fears of contact by attending trade fairs or cooperating with the Chamber of Commerce or with "Next Big Thing AG", which offers blockchain and IoT solutions for start-ups. But there is very much more potential here.
What is your second recommendation?
We need to train more specialists. Computer scientists, electrical engineers, whoever is involved, there are not enough of them. I hear this again and again in Berlin and Germany. But we also have to offer well-paid positions. When Americans offer twice as much, the skilled workers are gone. In computer science in German-speaking countries we have the problem that many do not complete their studies. Even before their bachelor's degree, they are so much in demand that their training falls by the wayside. AI in self-study does not work. The graduate rates at the universities are so bad because people see no need to obtain a formal degree. Although they are obstructing further development opportunities and ultimately better pay, this is not perceived as such. I give the introductory lecture on programming at the TU Berlin. I have 1,300 first-year students sitting in the lecture at the beginning - in my opinion this is too few. So many companies come to Berlin because they can't find people in their traditional environment, even if they attract people with starting salaries of 100,000 euro and company cars. There is simply nobody de facto.
Finally, let's take a look into the crystal ball: How do you see AI in the future?
Basically, AI must become a commodity. Today people say: "Oh God, AI! I would like to have a modular system with which you can build industry-specific module solutions that can be used in the infrastructure. For me, this infrastructure means Edge Cloud. Machine learning in the cloud - the race for this is over, the winner is in the USA. With Edge Cloud, we still have possibilities from the point of view of information technology. We must not miss them.
Many thanks for the interview.