A Tale of How a Drop-Out Algorithm Lost Its Powers: Predicting Student Dropout in Vocational Education

MT/DigiTrans higher seminar with Helene Friis Ratner, Professor in Organization Studies and Technology, Technical University of Denmark (DTU).

Predictive algorithms, developed through machine learning to identify patterns in large datasets, promise to anticipate and pre-empt future events. In education, algorithmic prediction is increasingly used to profile and target students, flagging individuals at risk of disengagement, poor performance, or drop-out. Much critical scholarship has examined the harms of predictive algorithms, raising concerns about behaviourist pedagogies, reductive datafication, bias, proliferating inequalities, and surveillance.
While important, such a critical departure also risks overlooking accounts that do not align with the ‘algorithmic drama’ (Ziewitz, 2016) of powerful and harmful algorithms so often depicted in critical algorithm and education studies. An alternative form of critique could be to examine the practical figurations and situated practices that (fail to) sustain algorithmic power.

To do so, this presentation tells the story of how a drop-out algorithm in Denmark gradually lost its agency, utilizing sensitivities from post-actor-network theory. Analysing the algorithm as a ‘partially existing object’ (Jensen, 2010), this analysis elicits the distributed socio-technical elements that failed to sustain the drop-out algorithm and eventually stripped it from its powers. The presentation concludes with reflections on the critical edge of a rather mundane story of a non-powerful algorithm. While it is crucial to critique the harmful deployment of algorithmic prediction in education, this type of critique also risks reifying algorithms as stable, coherent and agential objects. Examining ‘failed’ algorithms as the one studied here, allows us to understand how algorithms gain their powers and, in turn, helps unpack the “thingness” of AI (Suchman, 2023). On a more normative scale, insights into rather useless algorithms allows us to question the hype surrounding digitalization in education and challenge dominant imaginaries about an inevitable future of data-driven education.

Jensen, C. B. (2010). Ontologies for Developing Things: Making Health Care Futures Through Technology. Sense Publishers.
Suchman, Lucy (2023) The uncontroversial ‘thingness’ of AI. Big Data & Society, July–December: 1–5.
Ziewitz, M. (2016). Governing Algorithms: Myth, Mess, and Methods. Science, Technology, & Human Values, 41(1), 3–16.

Helene Friis Ratner is Professor in Organization Studies and Technology and head of section of Organization Science and Technology. With a background in anthropology and science and technology studies, her research examines how data-driven technologies transform the relationship between State and citizens. She is PI of ‘AI in the Welfare State: Accountability and Responsibility at stake’ (IRFD, 2026-29), chief scientist in the National Center for AI in Society (CAISA), and co-PI in the research project ‘Algorithms, Data and Democracy’ (VELUX Foundations, 2021-31).