Why relevance is not computable
a kitchen story
Abstract
How are contexts structured to separate the relevant from the irrelevant? This question — the frame problem in its broadest form — has proved resistant to computational solutions for a principled reason: it arises from the representational architecture it is posed within. If cognition works by matching inputs against stored templates, relevance becomes a preliminary computation, and the computation is intractable.
Dreyfus (1992) argued that the frame problem reveals a structural limitation of representational approaches and that embodied coping avoids it. But his account leaves a gap: if relevance is not computed, what is the unit of embodied skill in which relevance is already operative? Without a positive answer, the Dreyfusian intuition remains a promissory note— a gap that remains even in accounts that reject content for basic minds (Hutto & Myin 2013)
I claim that the frame problem is generated by a specific conflation: treating “pattern” and “recognition” as two things — a specification and a subsequent verdict — rather than as one event. This serial architecture manufactures the relevance problem because the template does not contain the context-sensitivity of its own application. I propose Pattern–Recognition Unity (PRU): the basic cognitive unit is a Pattern-Constellation (noted ), a learned complex of sensory, motor, affective, and contextual sensitivities acquired and activated together. is the organism’s situated readiness — the way a situation shows up as calling for a continuation. A cat does not first perceive a mouse and then compute relevance; the crouch, the narrowed attention, the anticipatory arousal are the recognition. Relevance is constitutive of the constellation, not computed over it.
Three consequences follow. First, contexts do not need to be “structured to separate” relevant from irrelevant: the organism never faces a neutral totality requiring pre-sorting. Developmental history builds constellations in which what matters is already integrated into recognition. The frame problem dissolves — it asks how a system performs a task that situated cognition never undertakes. Second, a minimal notation — (experiential constellations), (linguistic constellations), (their coordination) — locates precisely where the frame problem re-enters: linguistic constellations can float free of experiential anchors, and when they do, relevance must be managed externally. The frame problem is real for systems operating in the linguistic register without experiential grounding — an artifact of a specific architecture, not a general condition of minded beings. Third, can be modeled as attractor basins in Hopfield-like networks shaped by Hebbian learning, where “recognition” is convergence to a basin rather than a separate step — giving PRU a bridge to dynamical-systems analyses of agent–environment coupling (cf. Kelso 1995) and a mathematical entry point for modeling situatedness.