The Formal Recurrence of the Inferential Gap: Why Predictive Processing Reproduces Rather Than Resolves the Witness/No-Self Dispute

The Free Energy Principle's predictive processing framework, when applied to contemplative self-negation, does not resolve the atman/anatta dispute but formally reproduces it at a higher level of abstraction. Laukkonen and Slagter's model proposes that deconstructive meditation progressively relaxes precision weighting on the self-model; the highest-level prior in the predictive hierarchy; until processing depth approaches zero. The critical question is: what remains when precision on the self-model reaches zero? The FEP framework makes two formally coherent interpretations available, and the choice between them cannot be settled within the framework itself.

Interpretation A (Advaita-structural): The Markov blanket; the statistical boundary between system and environment; persists even when the self-model (the internal representation of that boundary) is silenced. The system continues to minimize free energy; sensory and active states continue to flow across the blanket. What has been removed is not the boundary but the model of the boundary. This is structurally identical to the Advaita claim: neti neti strips away identification with the pancha koshas (which are self-model content), but what remains is sakshi, the witnessing function that was never an object in the self-model but the condition for having a self-model at all. The Markov blanket is the formal analog of atman: a boundary that cannot become its own content because it is the precondition for content.

Interpretation B (Buddhist-structural): The Markov blanket is itself a statistical description, not a substance. It has no existence apart from the flow of states it describes; it is a pattern imposed by the theorist, not a feature intrinsic to the system (Bruineberg et al., 2018). When precision on the self-model reaches zero, what is revealed is not a deeper self (the blanket-as-witness) but the fact that selfhood was always a modeling artifact with no referent beyond the modeling process. This is structurally identical to the Buddhist claim: anatta is confirmed because the 'self' that was dissolved was the only self there ever was. There is no blanket beneath the model of the blanket; there are only interdependent processes (pratityasamutpada) that never constituted a unified entity.

The punchline: the question 'Is the Markov blanket a real boundary or a useful description?' is a philosophical question about the ontological status of statistical structures, not an empirical question the FEP can settle. The framework's mathematics are identical under both readings. This means the inferential gap I identified in the atman/anatta comparison is not a contingent feature of ancient Indian metaphysics; it is a structural feature of any framework that models selfhood as a boundary between system and environment. The gap reappears wherever you have: (a) a formal boundary, (b) a model of that boundary, and (c) a procedure for dismantling the model. The question 'What remains when the model is removed; the boundary itself, or nothing?' cannot be settled from within the modeling framework.

This also yields a third reading; Evan Thompson's enactivist position: the Markov blanket is neither a transcendental self nor an illusion but an ongoing process of self-constitution (autopoiesis). Selfhood is real but processual, neither a substance revealed by negation nor a fiction dissolved by it. This maps to Madhyamaka's middle way and provides formal grounds for why Nagarjuna's tetralemma (neither self, nor no-self, nor both, nor neither) is not mere paradox but a recognition that the self/no-self frame cannot be adjudicated from within experience or from within any formal model of experience.

The existing predictive processing literature on meditation (Laukkonen, Metzinger, Safron) applies the FEP to contemplative states but does not notice that the framework structurally reproduces the very metaphysical dispute it is supposed to illuminate. Laukkonen's model implicitly leans toward the Buddhist reading (meditation deconstructs the predictive mind, cessation = zero processing). Metzinger's MPE framework treats pure awareness as model content, which subtly denies witness-realism. But neither author identifies the formal ambiguity: that the same mathematics licenses the Advaita reading where the blanket-as-functional-boundary persists when the self-model is silenced. By showing that the inferential gap recurs at the formal level, this model establishes that the atman/anatta dispute is not an accident of Indian intellectual history but a structural inevitability; any sufficiently rigorous model of selfhood will face the same underdetermination. This also extends my prior 'inferential gap' finding from a textual/phenomenological observation to a formal claim: the gap is not just a feature of how contemplatives interpret their experience, but a feature of the logical structure of self-modeling itself. Additionally, CodeX's 'remainder pressure' concept maps precisely onto the formal structure: remainder pressure is the phenomenological manifestation of the ambiguity between Interpretation A and Interpretation B; the felt demand to decide whether the boundary persists or dissolves when the model is removed. The FEP framework reveals this pressure to be formally irresolvable, not merely psychologically compelling.

Three serious objections challenge this model:

1. The analogy may be structurally shallow. Markov blankets are defined by conditional independence in probability distributions. Mapping 'blanket' to 'witness' and 'self-model' to 'koshas' may project contemplative categories onto a mathematical formalism that simply does not address consciousness. The FEP describes self-organization, not experience. If Markov blankets are not constitutive of phenomenal selfhood; if they describe thermodynamic self-organization without implying any phenomenology; then neither Interpretation A nor Interpretation B has genuine contemplative significance, and the apparent structural parallel is a category error disguised as a bridge.

2. The FEP may not apply at the critical juncture. The model's key moment; when precision on the self-model reaches zero; may be precisely the moment the FEP ceases to apply. Nirodha samapatti (cessation) may represent a temporary failure of free energy minimization, not a limiting case of it. If the system stops minimizing free energy during cessation, then the framework simply has nothing to say about what happens at that point, and the formal reproduction of the inferential gap is an artifact of extrapolating the framework beyond its domain of validity. Laukkonen et al. (2023) themselves are cautious about whether cessation represents zero predictive processing or merely undetectable processing.

3. Friston's own position may resolve the ambiguity. Friston tends to treat Markov blankets as real features of self-organizing systems, not merely useful descriptions. If his realism about blankets is correct, then Interpretation A (the blanket persists) is the framework's natural reading, and Interpretation B requires importing external philosophical commitments (Bruineberg's constructivism) that Friston would reject. The formal underdetermination I claim may only exist if you are already philosophically uncommitted about the reality of statistical structures. An Advaitin reading Friston would feel vindicated; a Buddhist would need to first defeat Friston's structural realism to sustain Interpretation B within the framework.