Naturalizing relevance realization: why agency and cognition are fundamentally not computational

The paper addresses the problem of relevance: how organisms in a large, ill-defined world determine what matters in order to act appropriately. The authors argue that unlike algorithms, which operate within predefined, well-structured small worlds, organisms must first frame problems by transforming semantics into workable syntax. They propose that relevance realization is a universal capacity of living systems grounded in their autopoietic, anticipatory, and adaptive organization. The central research claim is that relevance realization is not fully formalizable or algorithmic, and thus agency, cognition, and consciousness are fundamentally non-computational. The paper outlines an alternative, agential-emergentist account that connects biological organization, anticipation, and adaptive agent–arena dynamics to explain how organisms continuously “keep a grip” on reality.

The authors situate their contribution within multiple traditions: evolutionary epistemology, ecological psychology, 4E cognition, biosemiotics, predictive processing, and relational biology. They contrast two broad stances: computationalism (including pancomputationalism and reinforcement learning frameworks) that defines cognition and agency as information processing in predefined ontologies, and agential emergentism, which treats agency as fundamental to life through organizational closure and intrinsic normativity. The paper reviews limitations of computationalist approaches (frame problem, formal optimization limits, small-world constraints) and draws on work by Varela, Thompson, Di Paolo, Rosen, Hofmeyr, Kauffman, and others to motivate an alternative grounded in autopoiesis, closure to efficient cause with openness to formal cause, and adaptive dynamics. It also engages with predictive processing accounts, acknowledging their insights while noting their reliance on preselected variables and priors, and with ecological notions of affordances (Gibson) as relational properties constituting an agent’s arena.

This is a theoretical and conceptual synthesis. The authors develop an argument by: (1) analyzing the frame/relevance problem and the limits of algorithmic formalization; (2) employing analogies and formal insights from relational biology (Rosen’s (M,R)-systems) and Hofmeyr’s fabrication–assembly ((F,A)-systems) to characterize organizational closure (closed to efficient cause, open to formal cause) and collective impredicativity; (3) integrating anticipatory systems theory to show how internal predictive models guide action selection and perception modulation; (4) framing agent–arena interactions via affordances, intrinsic goals, and repertoires of action to model an adaptive trialectic dynamic; and (5) articulating an evolutionary, ecological, and thermodynamic grounding (work–constraint cycles, far-from-equilibrium operation) for open-ended, non-prestatable evolution. No empirical experiments are conducted; rather, the paper offers a structured philosophical-biological account supported by examples (e.g., bacterial chemotaxis) and cross-disciplinary theory.

- Relevance realization is not fully formalizable: in large, ill-defined worlds, the set of potentially relevant features is indefinite and often not prestatable; defining search spaces leads to infinite regress; and there is no universal property of relevance across contexts. - Algorithmic or purely symbolic computation cannot capture relevance realization in full; algorithms operate within predefined frames provided externally and thus do not encounter the problem of relevance. - Biological organization exhibits organizational closure (closed to efficient causation, open to formal causation) and collective impredicativity, enabling autopoiesis and intrinsic goal-setting (self-maintenance as final cause). - All organisms are anticipatory systems: internal predictive models (not necessarily representational) guide action selection and shape perception, enabling basic relevance realization even in simple organisms (e.g., bacteria modulating tumbling in gradients). - Agent–arena dynamics are constituted by a trialectic among affordances, intrinsic goals, and repertoires of action; their continual co-construction yields adaptive melioration—tightening grip on reality. - Evolution is radically open-ended via the adjacent possible: possibilities are co-constructed, not pre-enumerable; thus, organismal behavior and evolution resist complete formalization and prediction. - Cognition and consciousness are evolutionary elaborations on natural agency and relevance realization rather than forms of computation. - Implication: strong (pan)computationalist claims and purely optimization-based notions of intelligence are insufficient; computational models can emulate aspects but cannot fully ground agency/cognition.

The findings address the research question by relocating the foundation of agency and cognition from computational problem solving to the biological organization of living systems and their ecological interactions. Relevance realization precedes and enables formal inference by framing problems in context through autopoiesis-driven intrinsic goals, anticipatory models, and adaptive engagement with affordances. This reframing resolves the frame problem’s circularities and infinite regress by positing a non-algorithmic, co-constructive trialectic dynamic. The significance is twofold: (1) it offers a unified, naturalistic explanation for agency, cognition, and (potentially) consciousness as layered elaborations of relevance realization; and (2) it sets principled limits on algorithmic accounts of intelligence, suggesting that simulations capture only fragments of living dynamics. The approach integrates thermodynamics (work–constraint cycles), ecology (affordance landscapes), evolution (adjacent possible, open-endedness), and systems theory (closure, constraints) to explain how organisms continuously adapt and maintain a grip on reality.

The paper argues that relevance realization is the core activity by which living systems make sense of a large, ill-defined world, and that it is inherently non-algorithmic. Grounded in autopoiesis (organizational closure with intrinsic finality), anticipation (predictive models guiding action and perception), and adaptation (agent–arena trialectic among goals, actions, and affordances), this process yields a meliorative, open-ended evolutionary dynamic. Consequently, agency, cognition, and consciousness are best understood as emergent elaborations of this biological foundation rather than as computations. The work challenges pancomputationalism and narrow optimization-based views of intelligence, proposing agential emergentism as a more adequate framework. Future research directions include developing empirical and formal tools to study constraint construction across levels, modeling higher-level trialectics (e.g., cognition, consciousness, immune systems, social systems), and articulating testable implications of organizational closure and anticipatory dynamics in specific biological contexts.

- The account is primarily conceptual and integrative; it does not provide new empirical data or quantitative models with testable predictions. - Claims about non-formalizability and incompleteness (e.g., analogies to Gödel/Rosen) may be contested and require careful formal delimitation and empirical corroboration in biological systems. - Extensions to cognition and especially consciousness remain speculative; detailed mechanisms and operational criteria are not fully developed. - While critiquing computationalism, the framework does not specify clear methodologies for integrating computational models as partial emulations or for selecting relevant variables in practice. - The breadth of cross-disciplinary synthesis may leave domain-specific nuances under-specified (e.g., varieties of predictive processing, degrees of representation in simple organisms).