Student epistemic agency in AI-mediated practices

Abstract Little is known about students’ epistemic agency in relation to AI-mediated practices. This study was conducted in a Swedish school that had started the pilot project “AI on the schedule” for compulsory school students, for whom the national curricula do not include AI training. A survey covering the aspects of source evaluation, data privacy, algorithmic manipulation, transparency, responsibility, and governance was distributed to these year 7 students ( n = 47). The data was analysed through cross-tabulations of question pairs. This revealed how epistemic agency varied across aspects within individual students. Results show that students demonstrated strong epistemic agency in certain areas but weak agency in others, for example, in data privacy and algorithmic manipulation, even when these had been part of earlier instruction. The contribution of the study is that epistemic agency in AI-mediated practices is not a unified construct: the data suggest distinct output-directed and system-directed dimensions, each with its own area of application. This has implications for instruction and guidance when integrating AI into K-12 teaching.