Initiation to Science alongside generative AI: preserving epistemic integrity

A recently proposed perspective on physics education, 'Initiation to Science' (ITS), views learning physics as an initiation into the ways scientific knowledge is constructed, tested, justified and legitimized. Prompted by this perspective, this paper offers a brief reflection on a new and pressing challenge: the widespread availability and increasing use of generative AI (specifically large language models, LLMs). While ITS demands students' active engagement in conceptual struggle and reconstruction, learners’ welldocumented tendency to regulate effort in relation to perceived costs and benefits may draw them toward frictionless routes to task completion. The central concern is that LLMs can provide fluent, apparently authoritative outputs that may be factually and procedurally adequate, allowing students to complete tasks without engaging in the epistemic work those tasks are intended to elicit. Rather than prescribing definitive solutions, the manuscript seeks to stimulate discussion among physics teachers and the wider physics education community. It outlines three broad dimensions for teacher professional practice—epistemological competence, epistemic-preserving task design, and aligned assessment. The claim is not that these dimensions are entirely new, but that their effective implementation is now critical for the practical viability of ITS in the age of LLMs.