Gerard de Melo Hasso Plattner Institute, University of Potsdam Confidently Wrong, Yet Effective: Techniques for Deploying Imperfect AI