Interpretability biases in machine and human learning

Description

Bias is a central concept in machine learning. It describes anything that is
relevant for preferring one model over another, beyond the mere correctness on the
training data. With the advent of powerful but intransparent machine learning models,
the need for interpretable models has gained in importance. Yet, the question of
interpretability of machine learning models is often reduced to a mere syntactic
interpretability, i.e., to whether the model can be read and understood by a human or
not. In this talk, we will argue that research in explainable AI should develop finer
grained distinctions between degrees of interpretability, and that human cognitive
biases may be helpful to develop better XAI techniques. To understand interpretability,
we must relate machine learning biases to cognitive biases, which let humans prefer
certain explanations over others, even in cases when such a preference cannot be
rationally justified. Only with such a collaborative effort can we develop suitable
interpretability biases for machine learning.

Period	01 Sept 2025
Event title	KogWis 2025: Symposium Cognitive aspects of trust in human-AI teams
Event type	Conference
Location	Bochum, GermanyShow on map
Degree of Recognition	International