Cognitive Entanglements: A Postphenomenological Approach to Generative AI in Education

This paper reframes debates on artificial intelligence in education by shifting focus from the question “Do machines think?” to the pedagogically salient issue of how large language models (LLMs) redistribute cognitive agency in learn-ing environments. Drawing on postphenomenology and the extended mind thesis, we argue that LLMs are not neutral tools but mediators that co-constitute human– technology–world relations. Using Ihde’s modalities (embodiment, hermeneutic, alterity, and background), layered with Fasoli’s taxonomy of cognitive artifacts (constitutive, complementary, substitutive), we propose a framework for analyz-ing and designing the use of LLMs in education. The framework clarifies when LLMs scaffold interpretation and reasoning and when they displace it, linking these patterns to epistemic virtues and risks of technologically induced cogni-tive diminishment. We translate this analysis into pedagogical patterns that make AI participation visible and accountable. An illustrative teaching scenario in phi-losophy shows how LLMs can serve as interlocutors and co-interpreters while assessment shifts from product to process via prompt logs, revision trails, and oral defenses. We contend that technological neutrality is untenable: institutional policies, curricula, and assessment regimes actively configure cognitive ecolo-gies. Consequently, education should stabilize constitutive and complementary uses, preserve non-substitutable spaces for independent reasoning, and embed transparency and AI literacy across programs. Education’s task is not to reject or surrender to AI, but to design entanglements that expand, rather than erode, human agency.