stein | HAdW

WIN-Projekte

M.A.

Joachim Stein

University of Tübingen
Ethics and Philosophy Lab
Cluster of Excellence "Machine Learning: New Perspectives for Science"
Maria-von-Linden Straße 6 | 72076 Tübingen | Germany

joachim.stein@hadw-bw.de

I am a PhD student at the University of Tübingen specializing in the philosophy of AI. My research applies complexity theory to examine in what sense machine learning models are complex, and to identify how to mitigate this complexity to achieve a better understanding of these models. I conduct this research as a part of the KEI research group. I am also an associated member of the International Max Planck Research School for Intelligent Systems (IMPRS-IS) and the Philosophy of Machine Learning Group at the Munich Center for Mathematical Philosophy (PhilML@MCMP).

I studied philosophy, mathematics and physics in Heidelberg and completed a M.A. in Logic and Philosophy of Science at the Munich Center for Mathematical Philosophy (MCMP). In my master thesis (supervised by Dr. Tom Sterkenburg and Dr. Levin Hornischer), I investigated the connection between machine learning theory and the principle of Ockham's razor (prefer simple solutions).

Writings and Events

Writings

How Complexity Contributes to Learning Opacity in Machine Learning (joint work with Dr. Eric Raidl)
We examine how complexity limits researchers' understanding of the learning process in machine learning. For this, we draw insights from the complexity analysis in the sciences. We demonstrate that key properties of complexity are inherent in the learning process of current machine learning techniques, and illustrate how these properties contribute to a lack of understanding of it.
Preliminary version: https://arxiv.org/abs/2606.24953
Does Mechanistic Interpretability Research Provide Mechanistic Explanations?
In recent years, there has been a proliferation of efforts to understand NNs through their internal mechanisms, giving rise to the field of mechanistic interpretability (MI). NN behavior is explained by decomposing the system into its constituent parts, identifying the functions of these parts, and showing how their interactions causally produce the phenomenon under investigation, thus yielding mechanistic explanations. I argue that the optimistic view of MI as a means to understand NNs, prevalent in philosophy and machine learning, is misguided. I present case studies showing that central cases of MI research do not, in fact, yield mechanistic explanations. I then argue that there is empirical evidence against the claim that NNs are decomposable, which undermines the central assumption of MI. Work in progress.

Events

11/2026: 11th annual international Philosophy Students' Symposium, Maribor

10/2026: PhilML'26 - Philosophy of Machine Learning Conference, Munich

09/2026: SOPhiA 2026 - Salzburg Conference for Young Analytic Philosophy

02/2026 - 05/2026: Guest Researcher in the Vici-Project "Gold Rush in the Data Mine" hosted by Prof. Dr. Jan-Willem at the Department of Theoretical Philosophy, University of Groningen

07/2025: SAS25 Conference – Uncertainty, Stuttgart

04/2025: Tübingen and Friends Workshop for Philosophy of Machine Learning 2025

03/2025: GWP 2025 - Triennial Conference of the German Society for Philosophy of Science, Erlangen