Abstract

Generative AI functions effectively at the periphery of academic research but is structurally absent from its cognitive core. Drawing on autoethnographic documentation of writing an MA dissertation with extensive GenAI use, this session introduces analytic rumination – the five-phase iterative process of assimilating, contemplating, conjugating, expressing, and reflecting – and demonstrates why GenAI cannot perform it. The session challenges the discourse of AI as ‘co-thinker’ and argues that higher education must name, teach, and protect analytic rumination as the irreducible precondition for intelligent engagement with GenAI.

Session and activities

This paper presentation offers a confessional account of academic research in the age of GenAI. Using my own MA dissertation (UAL, 2024–2025) as the case study, I will present findings from a detailed process log documenting what GenAI could and could not do across the full research cycle.

What GenAI did well: functioned as personal tutor, literature navigator (via tools like Anara), data processor, and research facilitator – genuine contributions that made aspects of the research more tractable.

What GenAI could not do: the hours of reading that produced nothing usable; the theoretical framework (Actor-Network Theory) abandoned after months because it did not work; the arguments that only became clear through writing; the feedback from supervisors and peers that dismantled and rebuilt thinking at the deepest level.

The session introduces analytic rumination – a five-phase cognitive model (assimilating, contemplating, conjugating, expressing, reflecting) anchored in Critical Realist theory and Archer’s reflexivity. I will argue that GenAI’s empirical fluency conceals its actual limitation: absence of the consciousness required for contemplation, conjugation, and reflection. This is not a temporary technological constraint. It is structural.

The presentation engages evidence that undeveloped thinkers are most harmed by AI dependency (Habib et al., 2024), challenges the assumption that students will instinctively interrogate AI outputs (Bardyn, 2025), and connects this to the documented phenomenon of model collapse (Shumailov et al., 2024; Wenger, 2024) – what happens when AI feeds on AI-generated content.

The discussion will focus on three institutional propositions: (1) name what must be protected; (2) teach analytic rumination explicitly as a learnable skill; (3) distinguish GenAI use at the periphery from GenAI use at the core. The session concludes with Gough’s (2004) provocation: what kind of cyborgs are we becoming?