This is the conclusion of scientists at TU Berlin (Berlin Technical University), the Fraunhofer Heinrich-Hertz Institute and the University of Singapore following their analysis of existing AI systems. For this they used technology they developed themselves that can identify the criteria on the basis of which systems make decisions. It emerged that many AI systems simply use so-called “clever Hans strategies”. These are named after a horse that was supposed to be able to do mathematical sums, but was in fact giving the right answers based on the reactions of the people asking the questions. Like “clever Hans”, some AI systems are able to carry out tasks such as allocating objects to categories very well, without actually recognizing the objects for what they are. According to the scientists, systems of this sort are “completely unusable for practical applications” – especially in safety-critical areas. The researchers are therefore calling for a systematic review of all the AI systems in use.
AI – provided that it is safe – can help manufacturing companies in many ways. VDI-Nachrichten presents an Israeli startup as an example. Using AI methods, it analyzes the data traffic in production networks in order to identify any cyber attacks at an early stage.