Abstract We use LLMs through explicit and thematic prompts, yet we rely on them to inform our tacit engagements in practical contexts. This gives rise to a paradox: how do our promptings and the outputs they generate relate to our practical behavior? This paper thematizes this relation by first exploring the nature of human tacit engagements, engaging with the debate between Dreyfus and McDowell, and identifying strengths and weaknesses in both positions. Since neither account fully captures the continuity between tacit and explicit knowledge, the paper turns to Heidegger’s phenomenology as a third position. Drawing on a Heideggerian account of human practical understanding, this paper shows how tacit knowledge underpins skilled action and how it can emerge in relation to AI-generated outputs. LLMs, while explicitly thematic and propositional, can direct attention toward functional relations among entities, support practical circumspection, and facilitate the discovery of previously unnoticed possibilities. Crucially, however, this occurs only within the context of the user’s pre-existing understanding, practical skills, and embodied engagement. Unlike human instructors, LLMs are not yet able to fully observe, contextualize, or respond to the learner’s situated actions. Users must therefore rely on iterative interpretation, self-monitoring, and practical triangulation to transform explicit AI-generated knowledge into behavior that, over time, can give rise to tacit knowledge.
Rasmus Gahrn-Andersen (Mon,) studied this question.