Robust AI in hazard prevention

Kurzfassung

AI systems are more and more integrated into decision‑making processes. Yet, for a safe deployment in safety‑critical situations, numerous challenges regarding their trustworthiness and reliability must be addressed. For that suitable metrics to ensure high enough data quality as well as robust AI-algorithms are indispensable. We provide an overview on current problems of AI systems and how to mitigate them. Moreover, we define three types of effect chains in order to schematically represent and analyze arbitrary complex AI systems. In addition, three Use Cases with increasing complexity in correspondence to the effect chains are examined based on the scenario of using unmanned aircraft systems for support in forest fire fighting.