Naturalizing Relevance Realization
Summary: Jaeger et al. (Frontiers in Psychology, 2024) argue from evolutionary biology and philosophy that agency and cognition are fundamentally not computational — relevance realization, the conversion of ill-defined to well-defined problems, is an impredicative, co-constructive process that no algorithm can fully capture.
Sources: Academia/Agency and cognition.pdf
Last updated: 2026-05-06
The core problem: the large world
Organisms live in a world that overflows with potential meaning. Environmental cues are scarce, ambiguous, and fragmentary. Problems are almost never well-defined in advance. Before any rule-governed inference can operate, an organism must first realize what is relevant — convert the large, open-ended world into a tractable task.
This is the relevance realization problem (Vervaeke et al. 2012). It is a necessary condition for any cognition at all. Without it, everything and nothing is relevant simultaneously.
Algorithms live in a small world
Algorithms (finite sets of symbols encoding operations executable on a Universal Turing machine) exist in a small world by definition. Their “digital environment” is a pre-defined, formalized ontology. They can only handle problems that are explicitly or implicitly pre-coded within their rules. They cannot identify or solve problems that are not pre-coded.
This is why relevance realization is not an algorithmic process. An algorithm never even encounters the problem.
The equivalence fallacy
Computationalists — particularly pancomputationalists who claim all physical processes are Turing-computable — commit what Jaeger et al. call the equivalence fallacy: treating the brain as a computer confuses the model with the thing being modeled.
The Church-Turing thesis was a description of a procedure performed by a human agent (the original “computer”). It was never intended as a claim about physical reality or biological brains. Assuming brains compute is a category mistake that “imputes symbolic content onto physical processes that is only really present in our simulations.”
Agential emergentism
Against computationalism, the authors propose agential emergentism: agency is fundamental and real, not a useful fiction or a higher-level emergent property of information processing.
The key distinction is between intrinsic goals (possessed by organisms, arising from precariousness and mortality — you are driven to continue living) and extrinsic purpose (possessed only by machines, imposed from outside). This is a discrete discontinuity between the living and the non-living. Machines can only have goals assigned to them; organisms define their own.
Contrast with Dennett’s intentional stance from dennett-intuition-pumps, which merely encourages treating living systems as if they had agency while retaining a reductionist worldview. Agential emergentism treats agency as genuinely present, not a useful interpretive posture.
Relevance realization as trialectic
Relevance is transjective — neither purely subjective nor purely objective. It arises from the organism-world interaction: the organism enacts its world by bringing forth the arena of task-relevant features through its actions and goals.
The process operates via opponent processing: organisms iteratively play competing adaptive strategies against each other, dynamically switching between them as needed. This is not optimization toward a fixed target; it is a Darwinian dialectic that continuously re-evaluates the problem space itself. The process is:
- Impredicative: the arena the organism constructs partly determines what counts as relevant for constructing the arena
- Co-constructive: organism and arena mutually constitute each other
- Open-ended: there is no fixed termination criterion
Because of its impredicative and co-constructive nature, relevance realization cannot be completely captured by any algorithm.
Connections
dual-process-cognition — the fast/slow framework (Kahneman, Sapolsky) describes biological architectures with amygdala and frontal cortex. Jaeger’s argument grounds these biological structures in something deeper: the evolutionary necessity of relevance realization. System 2 is not just “slow” — it is the apex of a biological capacity rooted in mortality.
free-will-and-moral-responsibility — if genuine agency requires intrinsic goals rooted in precariousness and mortality, this has direct consequences for what responsibility can mean. It also sharpens the question of whether LLMs can be held responsible for anything.
Empirically, illusion-of-thinking confirms the gap: LRMs fail at exactly the kinds of tasks that require genuine relevance realization — open-ended planning under novel constraints — while succeeding at tasks that can be pattern-matched from training data.