E029. Fabricata VII: The Firmware of the Flesh
Ratio sub Ratione — Predictive Processing, the Justificatory Function of Reason, and the Neural Substrate of the Unexaminable Prior
Abstract
The three prior monographs established a layered architecture of human cognition — an application layer of explicit doctrine, cleared by secularization, resting on a firmware layer of unexamined axiom that no argument installs and none removes. That architecture was, to that point, a structural conjecture supported by cultural and intellectual-historical evidence. This monograph descends one level further and asks whether the firmware has a physical substrate — whether the layered model is merely a useful metaphor for the history of ideas or a description of how the nervous system is in fact built. The contemporary computational neuroscience of perception and cognition, organized around the predictive-processing framework and its formalization in the free-energy principle, is shown to supply exactly such a substrate: the brain is not a passive receiver of an external world but a hierarchical generative model that predicts its sensory input and propagates only the residual error, so that what is consciously experienced is the model's "best guess" — a controlled hallucination — rather than the world itself. Within this architecture, the deepest priors function precisely as firmware: they are the bedrock on which all shallower predictions are conditioned, their revision would propagate catastrophic error through the entire hierarchy, and they are therefore held at a precision so high as to be effectively unrevisable — and, crucially, never experienced as priors. The paper then advances the harder thesis the prior monographs gestured toward: that what is called reason is not an independent truth-tracking faculty layered above this machinery but a specialized output of it, performing two functions — prediction (minimizing error within the generative model) and justification (the production, documented as the evolved primary function of explicit reasoning, of arguments defending conclusions the deeper layers have already reached). Confirmation bias, on this account, is not a malfunction of reason but its design signature. Two firmware priors named by the prior monographs — the linear arrow of time and the non-representability of one's own death — are shown to have specific, documented neural correlates, including a measured mechanism by which the brain suppresses its own predictive machinery when self and death are conjoined. The argument closes by deriving, from the neuroscience itself, both the impossibility of direct firmware-inspection and the partial possibility of its triangulation through comparison — grounding the methodological externality the prior monographs claimed in the architecture of the organ that claims it. Counter-evidence and live disputes (the empirical status of the free-energy principle, challenges to the argumentative theory, anatomical objections to the hallucination model) are stated at full strength and the thesis bounded accordingly.
0. Orientation: from metaphor to mechanism
The prior monographs deployed the language of computing — application layer, firmware, BIOS, compilation, sealing — and a fair critic could have dismissed all of it as metaphor: a vivid way of restating the ancient observation that people are dogmatic and do not know their own minds. This monograph takes the wager that the metaphor is load-bearing, that the layered architecture is not borrowed from computer science to decorate the history of ideas but is, approximately, how the brain is constructed. The wager is testable, because the relevant neuroscience exists and is specific. If the brain were a passive sensor delivering the world to a separate rational faculty for judgment, the firmware thesis would be metaphor and nothing more. If, instead, the brain turns out to be a hierarchical prediction engine in which the deepest priors are physically privileged and functionally unrevisable, and in which explicit reasoning is a late, specialized, justification-producing output rather than the seat of judgment, then the firmware was never a metaphor — it was an anatomical description awaiting its anatomy.
A discipline boundary must be marked at the outset and held throughout, in keeping with the provisional-floor commitment of the series. The neuroscience invoked here is current and in several respects contested; the predictive-processing framework is powerful and widely adopted but not settled, the free-energy principle is regarded by some as bordering on the unfalsifiable, and the argumentative theory of reason has able critics. This paper does not claim these frameworks as established fact. It claims that they are the best current science of the relevant functions, that they converge with striking precision on the architecture the prior monographs derived independently from cultural evidence, and that this convergence — science descending from below, intellectual history ascending from above, meeting at the same layered structure — is itself the argument. Convergence from independent directions is evidence; it is not proof, and the difference is maintained.
1. The predictive brain: perception as controlled hallucination
The folk model of perception, and the model implicitly assumed by classical rationalism, is that the senses deliver a representation of the external world to the mind, which then reasons about it. The contemporary computational neuroscience of perception inverts this almost completely. On the predictive-processing account, the brain is a hierarchical generative model of the causes of its sensory input; it continuously generates top-down predictions of the signals it expects to receive, compares them against the actual bottom-up signal, and propagates upward only the prediction error — the residual discrepancy — which is then used to update the model (Clark 181–207; Hohwy 1–46). Perception is not the reception of data but the brain's "best guess" about the hidden causes of its sensations, a guess held in place where it matches input and corrected where it does not.
Anil Seth's formulation makes the consequence vivid and exact: perceptual experience is "a neuronal fantasy that remains yoked to the world through a continuous making and remaking of perceptual best guesses, of controlled hallucinations" (Seth 92). The world we see is generated from within and merely constrained from without; hallucination, on this view, is not the opposite of perception but uncontrolled perception — the same generative machinery running with the sensory leash slackened (Seth 89–94). The claim extends inward: emotion and mood are the brain's best guesses about the causes of interoceptive signals — blood pressure, cardiac rhythm, gastric tension — to which the brain has no more direct access than it has to the external world, so that affective experience too is "a distinctive kind of controlled hallucination," a process Seth terms interoceptive inference (Seth 130–55). The self that does the perceiving is not exempt: it is itself "another perception, another controlled hallucination," an assemblage of self-related prior beliefs, bodily-regulatory predictions, and best guesses bundled together and stabilized because a stable self-model serves the organism's survival (Seth 147, 167, 220).
The formal foundation beneath this picture is Karl Friston's free-energy principle, which casts the brain — indeed any self-organizing system that resists dissipation — as minimizing variational free energy, a quantity that under simplifying assumptions reduces to long-term average prediction error (Friston, "Free-Energy Principle" 127–38). Perception minimizes error by updating the model to fit the input; action minimizes it by changing the input to fit the model (this is active inference: the organism samples the world so as to confirm its predictions); and crucially, the relative influence of any prediction error is gated by its estimated precision — its reliability — implemented neurobiologically as a gain control on the cells encoding the error, with neuromodulators such as dopamine reporting that precision (Friston, "Rough Guide" 1230–33; Feldman and Friston 1–23). Attention, in this framework, simply is the up-weighting of the precision of selected prediction errors; this is why one can stare directly at a thing and not see it, when precision is allocated elsewhere (Seth 75–77). Two features of this architecture are about to do all the work of the paper: that the system is hierarchical, with deeper levels conditioning shallower ones, and that the influence of any signal depends on its assigned precision, which can in principle be set so high that no contrary evidence can move the belief it protects.
2. Why the deepest priors are firmware: the neural reading of the unexaminable
The prior monographs defined firmware as the layer of axiom that no argument installs and none removes, because it functions as the medium of thought rather than its object. The predictive-processing architecture supplies a precise neural rendering of why such a layer must exist and why it is unrevisable.
Consider the hierarchy. Shallow priors (this surface is red; that shape is a face) sit near the sensory edge and are cheap to revise: a single strong prediction error can overturn them, because little else is conditioned on them. The deepest priors (objects persist when unobserved; causes precede effects; space has three dimensions; time runs one way; I am a bounded continuant who will go on existing) sit at the apex of the generative hierarchy, and every shallower prediction is conditioned on them. To revise one of these is not to update a belief but to invalidate the entire stack of predictions that presuppose it — to propagate catastrophic error through the whole model at once. A system minimizing long-term prediction error will therefore protect its deepest priors with the highest possible precision, treating contrary evidence as noise to be explained away rather than signal to be learned from, because the cost of revision is the momentary collapse of coherent world-modeling itself. The deepest priors are, in the exact functional sense, unrevisable by evidence — not because they are certainly true but because the architecture cannot afford to doubt them.
This is the firmware, rendered in neural terms, and it has three properties the prior monographs asserted and the neuroscience now explains. First, it is not represented as a belief: Seth's pivotal observation is that "we do not experience the models as models" — the generative model's outputs are experienced as the world, transparently, and the deepest priors least of all appear as anything one holds, because they are the frame within which holding-anything occurs (Seth, in del Rio; cf. Metzinger's "transparency"). One does not experience the prior "time runs forward" as an opinion about time; one experiences time running forward, fact and frame fused. Second, it is unexaminable from within: to inspect a prior one would need to represent it as an object, but the deepest priors are the conditions under which objects are represented at all, so the inspecting act is itself constituted by the very priors it would inspect — the eye cannot turn to see itself seeing. Third, its revision is experienced as threat or unreality rather than as learning: when deep priors are forced toward update, the phenomenology is not "I have changed my mind" but derealization, depersonalization, dread — which is precisely what the clinical literature on predictive-processing accounts of psychosis and dissociation reports, the deep model losing its grip experienced as the world losing its solidity (Sterzer et al. 1–14). The firmware announces its disturbance not as counter-argument but as the floor giving way.
The seal of the first monograph — the system's concealment of its own constructedness — is here given its deepest grounding. Sealing was described as a cultural-institutional operation; it is also a neural default. The brain seals everyperceptual construction by not presenting it as a construction; "we do not experience the models as models" is the seal operating at the level of the visual field. The cultural seal (this doctrine was always thus) and the political seal (this authority is natural) are higher-order exploitations of a sealing that the nervous system performs constitutively, on every percept, below the reach of any awareness. Constructed reality presented as given reality is not first a trick of priests. It is first a property of cortex.
3. Reason as a justification-prediction system
The strongest and most contestable claim of this monograph is that what is called reason — the explicit, effortful, verbalizable faculty classically opposed to mere intuition — is not an independent truth-tracking organ seated above the predictive machinery, but a specialized output of it, discharging two functions: prediction and justification. The first follows directly from §§1–2: reasoning, like perception, operates within the generative model, conditioned on its priors, seeking the conclusion that minimizes error against what the model already holds. The second is the more startling, and it is the central finding of a convergent body of work in the cognitive science of reasoning.
The argumentative theory of reasoning, developed by Hugo Mercier and Dan Sperber, holds that human reasoning did not evolve to track truth for the solitary thinker but to produce and evaluate arguments in a social setting — to justify oneself and to persuade others. Their statement is unambiguous: "skilled arguers... are not after the truth but after arguments supporting their views," and this, they argue, "explains the notorious confirmation bias" (Mercier and Sperber, "Why Do Humans Reason?" 57). The reframing is total. On the rationalist model, confirmation bias — the systematic search for evidence favoring what one already believes, and the systematic discounting of evidence against — is a defect, a failure of reason to do its truth-tracking job. On the argumentative model, confirmation bias is reasoning functioning exactly as designed: an argument-production device should, of course, produce arguments for the position it is tasked with defending, and it is precisely because the function is justification rather than truth-discovery that the bias is, in Mercier and Sperber's words, the most robust and prevalent disposition in all of reasoning (Mercier and Sperber, Enigma of Reason 207–48). What looks like the central bug is the central feature; the device is not broken, it is a lawyer, and a lawyer who argued both sides with equal vigor would be the broken one.
The convergent finding from moral psychology is Jonathan Haidt's social intuitionist model, which holds that moral judgment is not generally caused by moral reasoning; rather, "moral reasoning is usually a post hoc construction, generated after a judgment has been reached" by rapid automatic intuition (Haidt 814). The order of operations is the firmware thesis stated in the psychology of morals: the deep layer (intuition) delivers the verdict, and the explicit layer (reasoning) is summoned afterward to build the case — the rational tail wagged by, or chasing, the emotional dog (Haidt 814–34). Haidt's experimental signature is moral dumbfounding: subjects deliver a confident moral verdict, are stripped by questioning of every reason they offer for it, and yet do not revise the verdict — they retain the conclusion while conceding they cannot justify it, which is exactly what one predicts if the verdict issued from a deep prior and the reasons were post hoc advocacy never load-bearing in the first place (Haidt and Björklund). The intuition was the judge; the reasons were the lawyer's brief; losing the brief does not unseat the judge.
Place these alongside the predictive architecture and the synthesis is exact. The deep generative model — neural firmware, including the affective and moral priors — delivers a verdict as a high-precision prediction; explicit reasoning is then recruited to justify that verdict to others and to oneself, producing precisely the confirmation-biased argument-search the argumentative theory predicts. Reason does not consult the firmware and report neutrally on it; reason is the firmware's advocate, generating the justifications that protect the deep priors from the social demand for accounts. This is the neural-cognitive form of the thesis the third monograph reached historically: the economist does not examine his confessional firmware and then reason from neutral ground; he reasons from it, and experiences the reasoning as neutral, because the justificatory faculty does not have access to — and is not in the business of auditing — the priors it defends.
4. Two firmware priors with neural addresses
The thesis gains its sharpest empirical purchase where a named firmware prior can be tied to a specific neural mechanism. Two priors that the prior monographs treated as paradigmatic — the linear arrow of time and the non-representability of one's own death — admit exactly this.
4.1 The arrow of time
That time is experienced as directional, linear, and irreversible is, the first and third monographs argued, a firmware prior — the one whose secularized form (history as a line bending toward fulfilment) underwrites the entire modern faith in progress, against the cyclical time of antiquity. The neuroscience locates this prior below culture, in the architecture of episodic memory and prediction. The hippocampus encodes sequence: "time cells" fire at successive moments to stamp events with temporal order, and episodic memory is structured as ordered, non-commutative sequence (Eichenbaum 732–44). More fundamentally, the predictive architecture is itself temporally asymmetric by construction: it predicts forward, from past and present toward future, and it learns from prediction error generated when the arrived future violates the predicted one — a mechanism demonstrated directly in the finding that mnemonic prediction errors, elicited by interrupting familiar narratives before their expected endings, drive hippocampal updating of episodic memory (Sinclair et al.). The brain's basic operation — predict the next state, update on the error — is an arrow, a one-way flow from expectation to correction that cannot be run backward. The cultural prior of linear time rests on a neural prior of asymmetric prediction; a civilization that held time to be cyclical (and several did, at the level of cosmology and doctrine) nonetheless ran, in every individual hippocampus, the same forward-only sequence-encoder. The firmware here is doubled: a near-universal neural arrow beneath a culturally variable cosmology, the deeper layer constraining what the shallower can durably be.
4.2 The non-representability of death
The most striking case is the prior the prior monographs named as the limit of the model — one's own non-existence, the slot the generative model cannot fill because the absence of experience is not a state the experience-generating system can represent. Here the neuroscience supplies not merely a correlate but a measured mechanism of active suppression. Dor-Ziderman and colleagues, recording magnetoencephalographic responses, used the brain's automatic prediction-and-surprise signal — the response by which the predictive system registers a violated expectation — and showed that when participants' own faces were paired with death-related words, the surprise response vanished: the brain failed to generate the prediction that would have linked self to death, a result the authors describe as a "failed update of a prediction," an unconscious shielding mechanism by which the brain treats the proposition I will die as unreliable information not to be learned (Dor-Ziderman et al. 12–22). The deep prior — I am a continuant; my non-existence is not a representable state— is not merely a passive gap but is actively defended by the suppression of the very predictive machinery that would otherwise update it.
The downstream consequence connects this monograph to the entire series. Terror Management Theory documents that mortality salience — the activation, even subliminal and in milliseconds, of awareness of death — drives intensified worldview defense: heightened adherence to one's cultural and ideological frame, increased in-group attachment, increased hostility to and demonization of out-groups (Dor-Ziderman et al. 1–4; Greenberg et al.). The mechanism the first monograph called legitimacy-armor and the third called firmware here acquires its motive force: the unrepresentable prior of death, when it presses against the model, is bought off by a tightening grip on the firmware that promises continuance — the religious become more religious, the patriotic more patriotic, the believer in any frame more fiercely a believer. The full-version meaning engine of the prior discussions is, at the neural level, in part a death-suppression technology: it furnishes the continuance the generative model demands and cannot otherwise represent. The seal, the worldview, the sealed legitimacy that resists stress — these are, mechanistically, what the brain reaches for when the one prediction it cannot make (its own cessation) threatens to arrive.
5. The architecture of partial externality
If reason is the firmware's advocate and the deepest priors are neurally unexaminable, the reflexive problem the prior monographs repeatedly flagged becomes acute: how can this analysis claim to see firmware, when it too is produced by a brain running on firmware and reasoning justificatorily from it? The honest answer is derived not by exempting the present analysis but by reading the escape route off the neuroscience itself.
Direct firmware-inspection is impossible, and the neuroscience says why: the deepest priors are the conditions of representation, so they cannot be represented as objects by the system whose representing they condition (§2). There is no view from nowhere, because the viewer is constituted by the priors the view would have to transcend; the locality principle of the first monograph is not a philosophical posit but a corollary of generative-model architecture. But partialexternality is available, and the same architecture explains it. A single generative model cannot objectify its own deep priors; but when two models with different deep priors are placed in contact — through comparison, through the encounter with a genuinely foreign frame — each model's predictions diverge from the other's precisely at the locations of their differing priors, and that divergence is a signal. One cannot see one's own firmware directly, but one can triangulate it from the systematic pattern of one's disagreements with a differently-configured system: the prior shows itself as the invariant that generates the divergence. This is the neural rendering of the third monograph's controlled experiment (the same doctrine compiling differently on different confessional firmware) and of its closing methodological claim — not a view from nowhere but "a view from the difference between somewheres."
The argumentative theory supplies the complementary mechanism, and redeems reason partially in the process. Mercier and Sperber's evidence shows that while confirmation bias dominates the production of arguments by an individual, the evaluation of others' arguments is far less biased, so that reasoning performs well not in the solitary thinker but in the group of disagreeing interlocutors who share an interest in the truth: the confirmation bias of each, aimed at defending different conclusions, collectively functions as a division of cognitive labour that filters arguments better than any individual could alone (Mercier and Sperber, "Why Do Humans Reason?" 65; Enigma 247–331). Reason is a poor solitary judge and a competent social filter. The externality available to the analyst is therefore the same in kind as the externality available to a deliberating group: not the transcendence of firmware, which is impossible, but the mutual exposure of differing firmwares to one another, in which each serves as the instrument that renders the other visible. The method of the entire series — comparison across confessions, across civilizations, across the FOX-and-Atlantic sofa — is not an arbitrary stylistic choice. It is the only epistemically available route to partial firmware-visibility, and the neuroscience explains why it is the only one.
The reflexive verdict, then, is neither the false confidence that this analysis stands outside firmware nor the paralysing concession that it is therefore worthless. It is the bounded claim the locality principle always implied: the present analysis runs on firmware it cannot fully see, achieves only the partial externality that contrast with other firmwares affords, and is most trustworthy exactly where it is doing what reason does well — evaluating, comparatively, the arguments and structures of differently-configured systems — and least trustworthy exactly where it is doing what reason does badly — defending, from the inside, conclusions its own deep priors have already delivered. Including this one.
6. Limits, counter-evidence, and the bounded claim
The provisional-floor discipline requires the strongest objections in full, and the science here is live enough to supply several.
The first concerns the empirical status of the free-energy principle. The FEP is criticized as so general as to border on the unfalsifiable — a framework that can be fitted to almost any system and therefore predicts little in particular — and the relationship between the grand principle and the specific, testable claims of predictive coding remains disputed (Colombo and Wright; Williams). The reply is that the paper's load is carried not by the FEP's most ambitious form but by the specific, independently-evidenced claims of predictive processing — hierarchical prediction, precision-weighting, prediction-error-driven updating, the controlled-hallucination account of perception — each of which has experimental support that does not depend on the FEP's universality. The firmware thesis needs the architecture, not the cosmology; if the FEP were withdrawn as overreach, the hierarchical-predictive substrate on which the argument actually rests would stand.
The second concerns the argumentative theory. Critics have argued that confirmation bias appears outside the reasoning domain — in perception, memory, and learning — which, if reasoning's confirmation bias is supposed to be the unique signature of an argumentative function, undermines the inference from the one to the other (Mercier's own framing is challenged on exactly this point; see the Synthese critique by Peters). Others question whether it is in fact evolutionarily advantageous to persuade others of whatever one antecedently believes, the premise the theory rests on (Dogramaci). These are serious and unresolved. The paper's reply is that its thesis survives a weakened argumentative theory: even if the precise evolutionary etiology is contested, the descriptive finding — that explicit reasoning systematically produces confirmation-biased justification and that moral and many other judgments are post hoc rationalized (Haidt's dumbfounding, robust and independently replicated) — is what the firmware thesis requires, and that descriptive finding is not in serious doubt even where its evolutionary explanation is. The justificatory function can be challenged; the justificatory behavior is documented.
The third concerns the controlled-hallucination model of perception and its neuroanatomical assumptions. The pediatric neurosurgeon Michael Egnor has objected that the model predicts impairments that clinical reality contradicts: split-brain patients, whose massive disconnection should fragment the integrated prediction the theory requires, show "divided perception but undivided consciousness"; and children with hydranencephaly, lacking the cortical hemispheres on which the predictive machinery is held to depend, are nonetheless demonstrably conscious and emotionally responsive (Egnor). These observations are real and are a genuine problem for strong cortical-integration versions of the theory; they bear on consciousness more than on the narrower claim this paper needs. The paper requires that perception and cognition be substantially predictive and prior-laden, and that the deepest priors be privileged and hard to revise — claims supported by the perceptual and reasoning evidence independently of whether the controlled-hallucination model correctly localizes or explains consciousness as such. Where Egnor's objections wound the consciousness theory, they leave the firmware architecture standing; the paper claims only the latter.
The fourth is the reflexivity already addressed in §5 and conceded without reservation: the analysis cannot exempt itself, claims only partial comparative externality, and is bounded accordingly.
The fifth bounds the scope of the neural claim. To show that a cultural prior (linear time, the meaning-engine, the confessional firmware) has a neural correlate or substrate is not to show that the neural level determines the cultural content. The arrow-of-time case explicitly displays a culturally variable layer (cyclical versus linear cosmologies) atop a near-invariant neural layer (forward-only sequence prediction); the relation between levels is constraint, not determination, and the paper claims the weaker relation. Neuroscience explains why firmware exists and why it is unrevisable; it does not, and the paper does not, claim that neuroscience explains why a given culture's firmware has the specific content it has. That content is set at the cultural-historical layer the third monograph addressed; this monograph supplies the substrate, not the substance.
7. Compiled summary
The layered architecture the prior monographs derived from cultural and intellectual-historical evidence — an application layer of explicit, revisable doctrine over a firmware layer of unexamined, unrevisable axiom — is not a metaphor borrowed from computing but an approximate description of the nervous system, supplied by the contemporary computational neuroscience of perception and cognition. The brain is a hierarchical generative model that predicts its sensory input and experiences its own best guess as the world — a controlled hallucination yoked to reality only by the correction of prediction error, and never experienced as a model, which is the seal of the first monograph operating constitutively at the level of cortex. Within this hierarchy the deepest priors function exactly as firmware: every shallower prediction is conditioned on them, their revision would propagate catastrophic error through the whole model, and they are therefore held at maximal precision, unrevisable by evidence, unexaminable from within because they are the conditions of examination, and defended — when pressed toward update — by derealization and dread rather than by counter-argument.
Reason, on this account, is not an independent truth-tracking faculty above the machinery but a specialized output of it, performing prediction (error-minimization within the model) and justification (the production of confirmation-biased arguments defending conclusions the deep priors have already delivered). The argumentative theory of reason and the social-intuitionist model of moral judgment converge on this: confirmation bias is not reason's malfunction but its design signature, and explicit reasoning is the post hoc advocate of a verdict issued elsewhere — the rational tail of the emotional dog, the lawyer briefed by an intuition it cannot overrule, as the persistence of conviction through total loss of stated reasons (moral dumbfounding) demonstrates. Two firmware priors named by the prior monographs are shown to have neural addresses: the arrow of time in the forward-only sequence-prediction of hippocampal memory, and the non-representability of one's own death in a measured mechanism by which the brain suppresses its own surprise response when self and death are conjoined — actively shielding the prior of continuance, and, under mortality salience, tightening the grip on the very worldview-firmware the series has tracked throughout, which is thereby revealed as in part a death-suppression technology.
Finally, the neuroscience that forecloses direct firmware-inspection also licenses partial externality: a single generative model cannot objectify its own deepest priors, but two differently-configured models placed in contact render each other's priors visible at the points of divergence, and the evaluation (as against the production) of arguments is the one reasoning operation that performs well across differing frames. The comparative method of the entire series — the difference between somewheres — is therefore not stylistic but the sole epistemically available route to firmware-visibility, grounded in the architecture of the organ that needs it. The thesis is bounded throughout: it rests on the contested-but-best-available science and claims convergence rather than proof; it survives the weakening of the free-energy principle, the argumentative theory, and the consciousness model to their defensible cores; it claims that neuroscience explains the existence and unrevisability of firmware but not the specific content of any culture's firmware, which is set at the historical layer; and it exempts neither its method nor its author from the justificatory architecture it describes, claiming only the partial, comparative externality that contrast affords. Ratio sub ratione latet — beneath reason lies a deeper reason, which reason serves and cannot see, and which it was built not to examine but to defend.
Works Cited
Clark, Andy. Surfing Uncertainty: Prediction, Action, and the Embodied Mind. Oxford UP, 2016.
Colombo, Matteo, and Cory Wright. "First Principles in the Life Sciences: The Free-Energy Principle, Organicism, and Mechanism." Synthese, vol. 198, supp. 14, 2021, pp. 3463–88.
Dogramaci, Sinan. "What Is the Function of Reasoning? On Mercier and Sperber's Argumentative and Justificatory Theories." Episteme, vol. 17, no. 3, 2020, pp. 316–30.
Dor-Ziderman, Yair, et al. "Prediction-Based Neural Mechanisms for Shielding the Self from Existential Threat." NeuroImage, vol. 202, 2019, 116080.
Egnor, Michael. "Is Consciousness a 'Controlled Brain Hallucination'? No." Mind Matters, 2022.
Eichenbaum, Howard. "Time Cells in the Hippocampus: A New Dimension for Mapping Memories." Nature Reviews Neuroscience, vol. 15, no. 11, 2014, pp. 732–44.
Feldman, Harriet, and Karl Friston. "Attention, Uncertainty, and Free-Energy." Frontiers in Human Neuroscience, vol. 4, 2010, art. 215.
Friston, Karl. "The Free-Energy Principle: A Unified Brain Theory?" Nature Reviews Neuroscience, vol. 11, no. 2, 2010, pp. 127–38.
Friston, Karl. "The Free-Energy Principle: A Rough Guide to the Brain?" Trends in Cognitive Sciences, vol. 13, no. 7, 2009, pp. 293–301.
Greenberg, Jeff, et al. "The Causes and Consequences of a Need for Self-Esteem: A Terror Management Theory." Public Self and Private Self, edited by Roy F. Baumeister, Springer, 1986, pp. 189–212.
Haidt, Jonathan. "The Emotional Dog and Its Rational Tail: A Social Intuitionist Approach to Moral Judgment." Psychological Review, vol. 108, no. 4, 2001, pp. 814–34.
Haidt, Jonathan, and Fredrik Björklund. "Social Intuitionists Answer Six Questions about Moral Psychology." Moral Psychology, vol. 2, edited by Walter Sinnott-Armstrong, MIT Press, 2008, pp. 181–217.
Hohwy, Jakob. The Predictive Mind. Oxford UP, 2013.
Mercier, Hugo, and Dan Sperber. "Why Do Humans Reason? Arguments for an Argumentative Theory." Behavioral and Brain Sciences, vol. 34, no. 2, 2011, pp. 57–74.
Mercier, Hugo, and Dan Sperber. The Enigma of Reason. Harvard UP, 2017.
Peters, Uwe. "Confirmation Bias without Rhyme or Reason." Synthese, vol. 198, no. 10, 2021, pp. 9657–82.
del Rio, Manuel. "Review of Being You by Anil Seth." 2025.
Seth, Anil. Being You: A New Science of Consciousness. Dutton, 2021.
Sinclair, Alyssa H., et al. "Prediction Errors Disrupt Hippocampal Representations and Update Episodic Memories." Proceedings of the National Academy of Sciences, vol. 118, no. 51, 2021, e2117625118.
Sterzer, Philipp, et al. "The Predictive Coding Account of Psychosis." Biological Psychiatry, vol. 84, no. 9, 2018, pp. 634–43.
Williams, Daniel. "Predictive Processing and the Representation Wars." Minds and Machines, vol. 28, 2018, pp. 141–72.
