Research
While practical applications of machine learning are ubiquitous, the theoretical and philosophical foundations of machine learning lag behind. Core concepts such as interpretability, contestability, generalization, or benchmarks are often only vaguely defined. Without precise definitions and conceptual foundations, machine learning risks remaining mere engineering rather than developing into a mature science or a reliable tool for scientific and industrial applications.
The goal of my research is to address these fundamental challenges. My approach combines philosophical conceptual analysis with rigorous mathematical modeling. I consider myself a mediator between disciplines: I am a philosopher close to machine learning practice, and a philosophically inclined AI researcher; but faculty lines are so 20th century anyway. You can find my complete publication record on my Google Scholar profile.
Philosophy of Machine Learning
My work in the philosophy of machine learning focuses on robustness, reliability, and the role of machine learning models in scientific inference.
-
The Benchmarking Epistemology.
Together with Sebastian Zezulka, I use insights on construct validity from psychology to show how benchmark results can support a wide range of scientific inferences. -
Scientific Inference with Interpretable Machine Learning.
Together with my colleagues, I show how interpretable machine learning can support scientific inference via property descriptions. -
Supervised Machine Learning for Science.
Together with Christoph Molnar, I analyze how learning is transforming scientific inquiry from an explanation to a prediction-focused enterprise and show how recent technical developments in machine learning still allow for insight beyond mere predictions. -
A Theory of Robustness in Machine Learning.
Together with Thomas Grote, I develop a unified account of robustness in machine learning as relative stability of a target under specific interventions on a modifier. -
Artificial Neural Nets and the Representation of Human Concepts.
I investigate what kinds of concepts are represented by neural networks trained via supervised learning. -
Foundation Models in Healthcare Require Rethinking Reliability.
Together with Thomas Grote and Philipp Berens, I argue that standard out-of-sample testing is insufficient for assessing the reliability of foundation models. -
The Intriguing Relation Between Counterfactual Explanations and Adversarial Examples.
I examine the conceptual relationship between counterfactual explanations and adversarial examples.
Ethics of AI
My work in AI ethics focuses on actionable explanations, recourse, and the normative implications of deploying machine learning systems in social contexts.
-
Algorithmic Contestability.
Work in progress. -
Performative Validity of Recourse Explanations.
Together with Gunnar König and other colleagues, I analyze the performative effects recourse explanations have on their own validity. -
Improvement-Focused Causal Recourse.
Together with Gunnar König and Moritz Grosse-Wentrup, I argue that many counterfactual explanations enable users to game systems rather than improve their qualifications, and show how causal knowledge can support meaningful recourse.
Explainable AI
My work on explainable AI examines the goals, limitations, and conceptual foundations of interpretability methods and explanation techniques.
-
Dear XAI Community, We Need to Talk!
Together with Gunnar König, I analyze widespread misconceptions in XAI research, particularly the lack of clarity about the purposes of explanation methods. -
Relating Partial Dependence and Permutation Feature Importance.
Together with my colleagues, I analyze what interpretability methods can tell us about the data-generating process. -
General Pitfalls of Model-Agnostic Interpretation Methods.
Together with my colleagues, we discuss common mistakes in model-agnostic interpretation. -
CountARFactuals.
Together with my colleagues, we provide an efficient algorithm for generating plausible model-agnostic counterfactual explanations.