The topic of AI has already experienced peak phases in the 1970s and 1980s. What is new?
The current hype is mainly driven by a special area of AI: machine learning using deep neural networks. Here, remarkable breakthroughs have been achieved and published in recent years. An important factor is the computing power of modern graphics cards, which makes it possible to learn large networks. The required extremely large amounts of data are also available. In addition, there have been advances in the architecture of neural networks, such as convolutional neural networks and the Long Short-Term Memory approach, as well as new tricks to help networks learn. But there is actually much more to AI: logical reasoning, optimal decisions under uncertainty, searching for solutions, environmental perception, understanding and modeling, motion planning and so on.
What is the importance of AI at Fraunhofer IOSB?
We are broadly engaged in this topic: We successfully use deep neural learning, but also many other methods. I would say that we master a large part of the range of machine learning and other AI methods and are certainly a scientific and technical heavyweight in this field. So we can also solve problems where there is not enough data for deep neural learning - but for example additional expert knowledge. We use our AI expertise in numerous application domains: to intelligently control production processes, make mobile systems autonomous, optimize supply infrastructures, and for image and video analysis in a wide variety of areas, from material sorting to human-machine interaction and public safety.
How do you assess the new AI strategy of the German government?
It is right and welcome that the Federal Government wants to invest noticeably in this topic. This opens up new opportunities for us to contribute and develop our expertise - now we need to formulate good ideas and work with the right partners to score points in the upcoming calls for proposals. The hype that has developed around the topic of AI can be seen in a more differentiated way. Here, justified concerns get mixed up with exaggerated expectations and the danger of focusing too one-sidedly: Of course, AI is not the solution to all problems, but a tool, albeit one with enormous potential. Every tool should be used with sense and reason, preferably creatively, but only for tasks for which it is suitable.
Artificial intelligence also triggers fears. Will machines one day be superior to humans and thus become a danger?
Computers were developed to solve certain tasks more efficiently and better than humans. AI methods enlarge the spectrum of these tasks. But: Every software is designed for a specific purpose - and in this respect it fundamentally lags behind the universal, creative and highly adaptive problem-solving ability of humans. An AI that threatens humanity is therefore still a fiction far removed from reality. However, one has to keep the fundamental danger in mind. Presumably, the effect of AI on our society is overestimated in the short term, but underestimated in the long term. Certainly, we need an ethical discussion today: Who has what data, who is allowed to do what with it and who owns it? What is the potential for abuse? Is it transparent how algorithms make decisions and who is responsible for them? We must not lose sight of such questions. And, of course, the purposes for which AI is used must stand up to ethical scrutiny - as with any technological development.
Interview: Ulrich Pontes.