Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam, quis nostrud exercitation ullamco laboris nisi ut aliquip ex ea commodo consequat. Duis aute irure dolor in reprehenderit in voluptate velit esse cillum dolore eu fugiat nulla pariatur. Excepteur sint occaecat cupidatat non proident, sunt in culpa qui officia deserunt mollit anim id est laborum.

The famous COMPAS case has demonstrated the difficulties in identifying and combatting bias and discrimination in AI-based penal decision-making. In this presentation, I will distinguish two kinds of discrimination that need to be addressed in this context. The first is related to the well-known problem of trade-offs between mutually incompatible accounts of statistical fairness, while the second concerns the specific demands of discursive fairness that apply when basing human decisions on empirical evidence. I will sketch the essential requirements for non-discrimination in each case. In the former, we must consider the relevant causes of observed correlations between race and recidivism, in order to judge the moral adequacy of alternative standards for statistical fairness. In the latter, we should analyse the specific reasons that are admissible in penal trials, in order to establish what types of information must be provided when justifying court decisions through AI evidence.

Deep neural networks have shown excellent performances in many real-world applications such as plant phenotyping. Unfortunately, they may show "Clever Hans"-like behaviour, making use of confounding factors within datasets, to achieve high prediction rates. Rather than discarding the trained models or the dataset, we show that interactions between the learning system and the human user can correct the model. Specifically, we revise the models decision process by adding annotated masks during the learning loop and penalize decisions made for wrong reasons. In this way the decision strategies of the machine can be improved, focusing on relevant features, without considerably dropping predictive performance.

Based on joint work with Patrick Schramowski, Wolfgang Stammer, Stefano Teso, Anna Brugger, Franziska Herbert, Xiaoting Shao, Hans-Georg Luigs, and Anne-Katrin Mahlein.

Evidence that discrimination is an issue in computer vision systems has been reported recently in face recognition, emotion analysis, pedestrian detection to name a few. These are examples of human-centric computer vision, i.e. AI systems where automatic decisions are made about humans based on their appearance and behaviour from visual data such as images or videos. My talk will address the key question: how to enable human-centric computer vision to be fair (non-discriminative w.r.t. protected characteristics) and explainable (interpreting how the design intent of the AI application has been met). I will highlight some of my recent works on 1) how to achieve fairness by learning representations that remove the semantics of protected characteristics as a data-to-data translation, and 2) how to mitigate the problem called tyranny of the majority (when the algorithms favour groups of individuals that are better represented in the training data) via GAN-generated contrastive examples. I will conclude with promising future directions.