Cordelia Berz

Interpretable machine learning (ML) is crucial for gaining insight into complex ML models. Among its most valuable tools are counterfactual explanations, identifying changes to input features that would alter an ML prediction. However, the reliability and usefulness of these explanations depend fundamentally on their logical foundation — a critical aspect that has largely been overlooked in the ML literature. This paper addresses this gap by establishing normative desiderata for a counterfactual logic specifically designed for explaining algorithmic decisions. We analyse and argue for the validity of logical properties such as simplification of disjunctive antecedents and modus ponens, while demonstrating the unsuitability of others such as antecedent strengthening and conjunction conditionalisation. To evaluate existing solutions, we critically compare three prominent counterfactual semantics — similarity, strict conditional, and interventionist — against these desiderata. Our findings reveal that none of these standard frameworks fully satisfy the requirements, highlighting the urgent need for alternative frameworks.

Rubin, Hannah, and Cailin O’Connor (2018. “Discrimination and Collaboration in Science.” Philosophy of Science 85 (3): 380–402) introduced a game-theoretic model to examine the dynamics of collaboration and discrimination in academia. They found that small minority groups face higher rates of discrimination in collaboration, despite moral and legal demands for equality. This paper extends their work by assessing the effectiveness of anti-discrimination measures, focusing on German legal provisions for discrimination in the workplace: internal complaints boards and training and organisational measures as mandated by the General Equal Treatment Act. My results show that internal boards are effective in preventing discrimination in initially fair (i.e. non-discriminatory) environments, but fall short in already discriminatory environments. In contrast, training and organisational measures significantly reduce discriminatory behaviour, even where awareness is low. This highlights the importance of proactive anti-discrimination measures.