Magdalena Krysztoforska

Magdalena Krysztoforska is an interdisciplinary scholar interested in the philosophical and socio-political implications of data-driven technologies. Her doctoral research proposed a framework for addressing the blind spots of critical and computational approaches to high-stakes implementations of machine learning (such as predictive policing), drawing on perspectives from generic epistemology, critical AI studies, and the philosophy of science. More recently, her work has been exploring issues around data imaginaries, generative AI art paradigms, and the politics of measurement in machine learning.

Magdalena Krysztoforska received her PhD in Critical Theory and Cultural Studies from the University of Nottingham in 2024, funded by the UK Arts and Humanities Research Council (AHRC). She has held teaching posts at the University of Nottingham, the Université Catholique de Lille, and the New Centre for Research and Practice, and since 2024 she has been a Visiting Researcher and member of the Responsible Computing group at the Max Planck Institute for Security and Privacy (MPI-SP).

Scalar ambitions have been at the centre of recent progress in AI, with increasing amounts of data, labour, and planetary resources powering the development of ever-larger models with a continuously expanding range of applications. However, much of the discourse around AI engages in a form of scalar collapse: systems presented as general problem solvers are in fact typically developed based on standardized benchmark tasks, which are inadvertently reductive interpretations of real-world problems.

This project engages in a critical mapping of the canonical datasets and evaluation frameworks commonly used in machine learning research, and proposes ‘benchmark perspectivism’ as an analytical lens for examining how these metrics shape the scope of resulting models. The diagrammatic approach adopted in this project aims not only to explore the politics of scale at work within AI narratives, but also to sketch out alternative scalar dynamics by drawing on non-hegemonic AI epistemologies.