Practical security of evidence for regulated artificial intelligence modules

Artificial intelligence has recently led to numerous new applications in various industry sectors.Whenever artificial intelligence modules are used in a black-box setting, quality monitoring of such modules remains an open challenge.This implies that users of such modules cannot predict the modules' performance following software updates or retraining.Specifically for regulated devices, keeping track of an artificial intelligence module's behavior and compliance with requirements is crucial.To this end, existing methods for monitoring the functional behavior of software are investigated and evaluated regarding their practical usability in this paper.Based on the results of the investigation, a proposal for a new adaptive quality monitoring scheme for artificial intelligence modules is made.