Reliable AI

A great deal of research has been carried out in the field of artificial intelligence (AI) in recent years. Promising applications have shown great potential in production, medicine and autonomous driving. However, there is still a gap between the laboratory and industry: current AI models are not reliable enough to be used in industry. Despite great successes, AI methods also have weaknesses: AI models are mostly black box methods whose inner workings and decision-making are no longer comprehensible due to their complexity. The predictions made are based on existing data, which is why unseen data or application situations often lead to unexpected predictions and results. For this reason, such AI processes cannot currently be used in safety-critical applications.

In our study, we research, test and develop current methods and algorithms for the design and verification of reliable AI models in order to close the gap between research and industry.