In the mid-1990s it was still considered a sensation at first: the chess computer developed by IBM called “Deep Blue” didn’t give the chess world champion Garry Kasparov a chance anymore and thus heralded a new era. Machine beats man in a highly complex board game – something that was hardly conceivable before can now be seen as a caesura in the history of artificial intelligence (AI) and adaptive computer systems.

What initially sounds like extremely complex and correspondingly expensive solutions that can only be implemented in special computer centres is increasingly finding its way into broader and more everyday areas of application due to the continuous further development of hardware and the increasing efficiency achieved by research. Especially the more and more powerful graphics cards with their multi-core technology provide the necessary performance.

What can neural networks do in practice?
A clear strength of neural networks lies for example in the field of image recognition. The key often lies in the initial data: If sufficiently large amounts of data are available as “training data”, neural networks are able to recognize corresponding structures and elements in the image – the more data are available, the more potential the systems can develop to detect even complex components such as faces or street signs and even to classify them specifically in terms of an application.