Algorithmic Detection of “Pseudoscience” and the Demarcation Problem: Educational Lessons from Impey et al. (2025)

Abstract

This essay critically examines Impey et al.’s (\citeyear{Impey2025}) application of machine-learning to detect ostensible scientific misinformation, arguing that their operational definition of pseudoscience is philosophically inadequate. While the authors achieve impressive accuracy rates (> 95\%) in flagging content across domains including astrology and UFO/UAPs, their reliance on a binary “real versus fake” classification ignores decades of philosophy and sociology of science literature demonstrating that demarcation between science and pseudoscience is contested, context-dependent, and socially negotiated rather than algorithmically determinable. Drawing on foundational work by Laudan, Hansson, Pigliucci and Boudry, and Gieryn’s boundary-work theory, we show that encoding oversimplified demarcation criteria into neural networks risks creating what might be called “algorithmic pseudo-demarcation”—technologically sophisticated tools lacking epistemological grounding. The essay examines methodological limitations including the black-box problem, synthetic training data concerns, and educational implications. It is argued that effective misinformation detection requires integrating philosophical rigor with technological innovation, acknowledging epistemic ambiguity, and fostering critical thinking rather than algorithmic deference.