Classification of artificial intelligence tools for educational research by the criterion of research autonomy

Existing frameworks classify AI tools for academic research by data type or functional role, leaving unanswered the question that most directly concerns research integrity: how much of the cognitive labour constitutive of scientific inquiry has been transferred to an algorithm? This paper proposes a classification built on a single criterion - research autonomy - defined as the degree to which a researcher retains control over the cognitive operations of scientific knowledge production. Five functional clusters form a spectrum from maximum to minimum research autonomy: (I) computational data analysis, where the algorithm performs only mathematically specified procedures; (II) content and discourse analysis, where it applies pre-validated category systems; (III) search and navigation, where it independently determines relevance; (IV) multimodal analysis, where it performs primary categorisation of pedagogical events; and (V) content generation and synthesis, where it generates text and proposes conceptual connections. For each cluster, the paper specifies educational research applications, characteristic methodological constraints, and ethical requirements. The framework supports three practical ends: methods reporting standards, cluster-differentiated institutional AI governance, and AI literacy curricula grounded in epistemic consequences.