The new German government has resolved to strengthen 'investment in artificial intelligence (AI)' - this declaration of intent in the coalition agreement is being closely monitored in Kaiserslautern. With the German Research Center for Artificial Intelligence (DFKI) and other institutes, Kaiserslautern is a focal point of AI research. The Director of the Fraunhofer Institute for Experimental Software Engineering (IESE), Peter Liggesmeyer, would like to see new priorities set in research funding.

"Broad AI research is good," says Liggesmeyer. "But particularly important areas are not yet being funded in the concentrated way that they should be for our industry." In certain areas of AI research, German institutions are barely competitive, explains the scientist, citing speech recognition as an example.

The Director of the Fraunhofer Institute IESE in Kaiserslautern, Peter Liggesmeyer. The computer scientist argues that research into artificial intelligence should be geared more towards industry requirements than it has been to date.

© Fraunhofer IESE/dpa

"We should consider which AI applications are particularly well suited to our economic structure," recommends Liggesmeyer. "Automotive engineering, plant engineering, the chemical industry and the pharmaceutical sector would benefit greatly from well-functioning AI if it can be used well and without risk. This could be a focus of AI research in Germany."

Under the buzzword Industry 4.0, a lot is already being done to research how humans and machines could optimally share tasks in production in the future, the professor notes. However, the use of cooperative robots, for example, presupposes that the risks are manageable.

"Nobody expects a residual risk to be reduced to zero, but you have to be able to quantify it." Only then can decisions on residual risk acceptance be made. Procedures therefore need to be developed to measure the residual risks of AI systems. In building statics, this is done with the use of theoretical models - for example, to assess the loads a bridge can withstand. In other technical areas, test procedures are the appropriate method. "Neither of these methods are currently available for machine learning processes."

For AI geared towards industrial applications, expertise in safety technology would have to be combined with AI expertise. Apart from Fraunhofer IESE and the University of Oldenburg, there are currently only a few locations in Germany where both are equally available, says Liggesmeyer.

Much more AI could be used in the chemical industry, for example at BASF in Rhineland-Palatinate, if this could be done safely. Vaccine production is also still very manual today because there is no reliable information on the risks of AI systems. "If we could manage to make AI more suitable for such critical areas of application, that would be an important step forward."