What Predictive Processing and Kashmir Shaivism Agree About

Two descriptions of a world not directly given

Two very different projects, separated by roughly a thousand years and by every institutional marker that usually keeps disciplines apart, have converged on a shared starting observation. The world you seem to simply see is not simply seen. It is built, moment by moment, by a process that ordinary experience does not disclose to itself.

The first project is computational neuroscience's predictive processing framework, associated with Karl Friston's free energy principle, Andy Clark's account of the "predicting mind", and Jakob Hohwy's work on prediction error minimisation. It treats the brain as a hierarchical inference engine, one that generates a model of the world and continuously revises that model against incoming sensory signal. The second is the Pratyabhijna school of Kashmir Shaivism, the "recognition" tradition developed with great philosophical precision by Utpaladeva in the tenth century and elaborated shortly after by Abhinavagupta.

This article takes both seriously on their own terms, which means refusing two lazy moves that are equally common and equally wrong. The first is to dismiss the comparison outright, on the grounds that one is science and the other is religion. That dismissal is too quick: Pratyabhijna is not devotional literature asking for belief, it is a rigorously argued phenomenology with its own internal logic.

The second, far more common in popular writing, is to announce that neuroscience or physics has "proven" non-dualism, or that the free energy principle has dissolved the hard problem. That move is worse than the first, because it borrows the authority of one domain to settle a question the other domain was never built to settle. Both errors are named specifically later in this piece. First, the structural parallels deserve to be laid out carefully, before the divergences are traced with equal care.

The predictive brain: a working account

Predictive processing proposes that the brain does not build perception from the bottom up, passively assembling sensations into objects. It works top-down first: higher levels of a cortical hierarchy generate predictions about what lower levels should be receiving, and only the mismatch between prediction and actual signal, the prediction error, is passed upward. Perception, on this account, is closer to a controlled and constrained guess than to a photograph.

Andy Clark and Jakob Hohwy both describe the resulting picture as one in which the brain is, in a specific technical sense, an organ for minimising surprise, either updating its model to fit the world or acting on the world to make it match the model, with both read as instances of the same underlying inferential currency.

Karl Friston generalised this into the free energy principle, a mathematical formalism claiming that any system that persists over time against entropic dissolution must be describable as minimising a quantity called variational free energy, which functions as an upper bound on surprise. This is an ambitious claim, meant to apply not just to perception but to homeostasis, action selection, and, in Friston's more expansive writing, to the organisation of life itself.

The neuroscientist Anil Seth has popularised the resulting picture with the phrase "controlled hallucination": what you perceive is your brain's best current guess about the causes of its sensory input, held in check. But not replaced, by the error signal arriving from the world.

None of this is a claim about phenomenal consciousness as such. Predictive processing is a functional and computational theory. It specifies an architecture (hierarchical generative models), a currency (prediction error). A control parameter, precision weighting, the confidence assigned to a given prediction error, which functions similarly to attentional gain and determines how much a mismatch is allowed to revise the model. It explains why you see a stable world despite constantly shifting retinal input, why illusions arise, and why attention sharpens some signals and mutes others.

What it does not explain, and does not claim to explain, is why any of this inferential activity is accompanied by experience at all, why there is something it is like to be a system running this kind of inference rather than the same computations occurring with the lights off. This is David Chalmers's hard problem, and predictive processing, including Friston's more totalising formulations, addresses what Chalmers called the easy problems: the mechanisms of discrimination, integration, reportability, and behavioural control.

Some proponents write as though solving the easy problems at sufficient scale dissolves the hard one. That is a philosophical wager, not a result the mathematics itself delivers, and it will matter later.

The pratyabhijna vision: recognising what is already the case

Kashmir Shaivism's non-dual branch, and the Pratyabhijna school within it specifically, makes a claim that sits at a different level entirely. Its subject is not information processing but the nature of awareness itself, and its central technical vocabulary is worth defining carefully, since these terms carry precise meanings that get flattened in loose popular use.

Shakti is the creative power of consciousness, not a separate force acting on consciousness from outside but consciousness's own capacity to appear as multiplicity while remaining, in its own nature, undivided. Svatantrya is the absolute freedom or autonomy of awareness: the claim that consciousness manifests the world of appearances not because it is compelled by some prior cause but as an expression of its own unconditioned creativity, the way a performer manifests a role without ceasing to be themselves.

Spanda names the intrinsic vibration or pulsation of awareness, the subtle dynamism by which consciousness is never static but continuously active, closer to a standing wave than to a frozen substance. Abhasa refers to appearance in the sense of a shining-forth: each object, thought, and perception is an abhasa, a manifestation within consciousness rather than a mind-independent thing consciousness subsequently represents. Sankoca is contraction, the process by which unlimited awareness takes on the appearance of limitation, producing the sense of being a bounded body-mind confronting an external world.

Anu names the resulting limited experiencer, the individual as it takes itself to be under conditions of contraction: finite, embedded in time, cut off from its own source. Pratyabhijna, which gives the school its name, means recognition: not the acquisition of a new fact or faculty. But the direct re-cognition of what awareness already and always is, obscured only by habitual identification with the contracted state.

Utpaladeva's argument, refined by Abhinavagupta, is that this recognition is available because nothing was ever actually lost. The bound individual, the anu, is not a different substance from the unconditioned awareness the tradition calls Shiva, a technical term here for pure reflexive consciousness, prakasha combined with vimarsha, luminosity together with self-awareness of that luminosity, not a deity to be petitioned. Bondage is a matter of self-forgetting, a contraction of the field of recognition, not a fall into a genuinely separate condition.

Liberation is not the construction of a new state but the removal of a superimposed limitation, rather like recognising a person you already knew well but had, for a moment, failed to place. This recognition claim is explicitly first-person, meant to be validated, if at all, from within sustained contemplative attention, though the tradition also offers extensive argument for it as a coherent account of experience.

Where the two structures genuinely rhyme

With both frameworks stated in their own terms, three structural parallels stand out as worth taking seriously, points where independently developed models of experience reach similar architectural conclusions.

The given world as construction

Naive realism, the default assumption of ordinary waking life, holds that the world simply is as it appears, delivered whole to a passive observer. Both frameworks reject this starting point, for structurally similar reasons. Predictive processing argues that what is presented in perception is already the product of an inferential process running on prior expectations; there is no raw, unprocessed given beneath it, only prediction constrained by error.

Pratyabhijna argues that what appears is abhasa, a manifestation within awareness, never a mind-independent object encountered from outside. There is no world "out there" that consciousness subsequently copies, only appearance generated by its own creative capacity. In both accounts the ordinary conviction of direct, unmediated access to a ready-made external world is the thing to be explained away, not the thing that requires no explanation.

The self as appearance, not a given primitive

Both frameworks treat the felt sense of being a separate self as constructed rather than as a metaphysical starting point built into experience. In predictive processing, several researchers, notably Anil Seth and, in a related vein, Thomas Metzinger, treat the sense of selfhood, including the sense of owning a body and occupying a location, as itself a predictive model, generated and maintained the same way object perception is.

It can be perturbed, as in depersonalisation, out-of-body experiences, or certain meditative states, precisely because it is a model and not a metaphysical bedrock. Pratyabhijna makes an even stronger claim: the anu, the individual bound experiencer, is the product of sankoca, a contraction of awareness producing the appearance of a bounded self set over against a world. In neither framework is "I am a separate located self, encountering a separate external world" treated as a brute, unanalysable fact. Both treat it as an output.

Generative priority over passive copy

The third parallel follows from the first two. Both frameworks give explanatory priority to an active, generative process over a passive, receptive one. In predictive processing, the direction of primary causal traffic in perception runs top-down: the generative model leads. Bottom-up signal serves mainly to correct it, rather than build it up piece by piece as classical bottom-up theories of vision once assumed.

In Pratyabhijna, appearance is an expression of Shakti operating through svatantrya, consciousness's own freedom to manifest as many while never actually dividing into many, rather than an object impinging upon a passive witness from outside. In both cases the picture of a mind quietly recording an already-finished world is replaced with one in which the world experienced is continuously generated by the system doing the experiencing.

These are structural parallels, meaning parallels of pattern and relationship, not identity claims. The next section makes that structure visible as a diagram, before a fourth, more delicate parallel, involving precision weighting and contraction, is handled on its own with the caution it requires.

A concept map of the comparison

The diagram below lays out the two vocabularies side by side, showing where their structures run parallel and where a dotted line marks a cautious cross-framework comparison, discussed in the next section.

Read the left branch as the predictive processing account: a constructed appearance built from a generative model, refined through prediction error, gated by precision weighting, including a self-model among its constructions. Read the right branch as the Pratyabhijna account: appearance as abhasa, understood as the pulsation of Spanda, subject to contraction that produces the limited experiencer, with recognition as the resolution of that contraction through disclosure of underlying freedom. The dotted lines mark the two places this article draws a cautious structural comparison; everywhere else the branches are kept deliberately separate, because collapsing them would flatten two different kinds of claim into one.

Precision weighting and contraction: a careful analogy

The most tempting point of contact between the two frameworks, and the one that most needs careful handling, concerns precision weighting on one side and sankoca, contraction, together with its release, on the other.

In predictive processing, precision weighting is the mechanism by which the brain decides how much a given prediction error should be allowed to update the model. High precision assigned to an error signal means it is treated as reliable and given strong weight in revising the model; low precision means it is discounted, treated as noise. This functions somewhat like attentional gain control: raising precision on a channel is close to what it means, in ordinary language, to attend to it more. Disorders of precision weighting have been proposed as partial explanations for conditions ranging from autism to schizophrenia.

In Pratyabhijna, contraction and its release describe something at a different level: the degree to which awareness recognises itself as the ground of appearance versus identifying exclusively with a bounded, located, separate self. When contraction is total, the sense of being a separate, embattled self dominates experience entirely, and awareness does not recognise its own wider nature within any appearance arising in it. When contraction loosens, without appearances necessarily changing in their sensory content at all, the felt sense of separateness softens; the same perceptions continue to arise but exclusive identification with a bounded viewpoint relaxes.

There is a real structural resemblance worth naming: both mechanisms describe a variable, gradable modulation that governs how strongly a given content grips or organises experience, rather than describing the content itself. Turning precision up or down changes how much a signal shapes the model without changing the raw sensory input. Loosening contraction changes how much separateness organises experience without necessarily changing the sensory content of the moment. Both are, in this narrow sense, gain controls on something other than raw content.

The resemblance stops there, and the difference matters more than the similarity. Precision weighting is a quantitative parameter inside a computational model, measurable indirectly through pupillometry, neural gain proxies. Behavioural sensitivity to statistical structure, and manipulable experimentally through the reliability of a sensory cue in a controlled task.

Contraction is not a parameter inside any information-processing model; it is a qualitative characterisation of a state of recognition, described from within first-person contemplative investigation, with no operational definition, no independent third-person measurement procedure. No claim to being decomposable into discrete computational operations. Saying contraction is "like" precision weighting turned down is a suggestive metaphor about gain and salience.

It is not a discovery that the two are the same mechanism described in two vocabularies. Treating it that way would smuggle a philosophical claim through the back door of an analogy that only holds at the level of abstract pattern.

Concept by concept: a comparison table

The table below sets the core vocabulary of both frameworks side by side, with an explicit testability column, since testability is the axis this article insists on tracking throughout.

Where they radically part

Having traced the parallels honestly, the divergence needs equal honesty, because it is not a minor caveat appended for balance. It is the main event.

Predictive processing, in every serious formulation, is a functional and computational theory. It specifies mechanisms: hierarchical generative models, prediction error, precision weighting, active inference, defined operationally and tested through neuroimaging, behavioural experiment, and computational modelling. The theory remains agnostic, or in most of its major proponents' hands quietly physicalist, about phenomenal consciousness itself, meaning about why information processing is accompanied by subjective experience rather than occurring, functionally identically, without anyone home.

Friston's free energy principle is the most ambitious version, extending the same mathematics to living systems generally and, in some of his writing, gesturing toward consciousness as an emergent property of sufficiently deep generative modelling. But gesturing toward an emergence claim is not the same as deriving one, and the mathematics contains no term that specifies why minimising free energy should feel like anything at all. This is Chalmers's hard problem, restated in a new technical vocabulary rather than solved by it.

Pratyabhijna's central claim sits on the other side of exactly that gap. Its subject is not the mechanism by which information is processed but the nature of awareness as such. Its claim is that awareness is not a byproduct of any process, computational or otherwise, but is ontologically primary: the ground within which all appearances, including bodies, brains, and computational processes themselves, show up.

This is an idealist claim in a specific, carefully argued non-dual sense, not the folk idealism of "the world is a dream" but a rigorous position holding that reflexive awareness, prakasha combined with vimarsha, is what all appearance is made of, rather than something appearing within a prior physical world. This is a metaphysical claim, sitting entirely outside the domain predictive processing is built to test.

A functional theory of information processing has no instrument capable of confirming or disconfirming a claim about the ultimate nature of awareness, because the theory is written in a vocabulary, prediction, error, precision, model, that presupposes a subject-object structure from the outset. Pratyabhijna's claim is precisely that this structure is itself an appearance within awareness, not a description of awareness's own nature. A theory operating entirely inside the presupposed structure cannot, by its own methods, adjudicate a claim about what lies prior to that structure.

Svatantrya makes this sharpest. The claim that awareness manifests appearance through its own unconditioned freedom is not offered as a hypothesis awaiting laboratory confirmation. It is offered as something recognised, if it is recognised at all, through sustained first-person contemplative investigation, verified by criteria internal to that investigation, a shift from felt bondage and contraction to felt freedom, not by third-person replication using instruments external to the investigating awareness itself. This does not make the claim meaningless.

Careful phenomenological traditions have internal standards of rigour and cross-checking across generations of practitioners, and Utpaladeva's and Abhinavagupta's texts are dense, technical, and argumentatively demanding rather than merely reportorial. But the claim belongs to a different epistemic category than one tested by neuroimaging, and honesty requires keeping that distinction visible rather than letting the shared vocabulary of "construction" and "appearance" blur it.

The category error: why physics does not prove non-dualism

This is the point at which a specific, common, and avoidable error needs to be named directly, because it appears constantly in popular writing that gestures toward both neuroscience and contemplative traditions. The claim that physics, or neuroscience, or the free energy principle specifically, "proves" non-dualism, or dissolves the hard problem, or vindicates Advaita Vedanta or Kashmir Shaivism as scientifically confirmed metaphysics.

The error is a category error in the precise philosophical sense. It treats a claim belonging to one category of question, functional mechanism, as though it settled a claim in an entirely different category, the ultimate nature of awareness. Predictive processing, however elaborated, is a theory about how systems that process information organise that processing: it describes architecture and dynamics. Even in its most ambitious form, the free energy principle is a claim about which mathematical description applies to self-organising physical systems generally.

Neither is, or could by its own construction be, a claim about whether awareness is ontologically fundamental. A functional isomorphism between two models, for instance the loose resemblance between precision weighting and contraction traced above, is not an ontological identity between what the two models are about. Showing that a computational system's internal gain control resembles, in abstract shape, a contemplative description of contraction and its release demonstrates a pattern-level parallel.

It demonstrates nothing about whether awareness itself is fundamental, because the demonstration never leaves the domain of pattern-matching between two models to make contact with the metaphysical question at all.

The deeper reason the two projects cannot settle each other's questions is that they are answering different questions from the start, not competing to answer the same one. Predictive processing asks: by what mechanism does a system generate perceptual content, select action, and maintain itself against disorder. That is a question about process, answerable in principle by specifying the process and testing its predictions. Pratyabhijna asks: what is awareness, as such, prior to and constitutive of any process whatsoever, including the process of asking the question.

That is not a question about mechanism at all; it is a question about the ground in which mechanism, along with everything else that appears, shows up. No amount of detail about mechanism answers a question about ground, for the same reason that no amount of detail about the grammar of a sentence answers what makes marks on a page mean anything in the first place. The two enquiries can inform each other and genuinely rhyme structurally, as earlier sections tried to show honestly.

But one cannot stand in for the other's verdict, and claiming that it does is not humility toward either tradition, it is a failure to read either one carefully enough to see what kind of claim it is actually making.

A note on method: third-person instruments and first-person recognition

The asymmetry traced above has a further consequence: the two frameworks are not just answering different questions, they are answerable by different methods. Conflating the methods causes as much confusion as conflating the questions.

Predictive processing generates falsifiable predictions. If precision weighting genuinely functions as proposed, specific perceptual illusions should behave in specific ways under specific manipulations of cue reliability, patient populations should show specific patterns of over- or under-weighting of sensory versus prior information. Neural signatures should track prediction error and precision separately. These predictions have been tested, refined, and in places revised, functioning imperfectly and unevenly across different sub-claims of the broader framework, as is normal for an active field.

Pratyabhijna's central claims are not tested this way, and the tradition itself does not claim they should be. Its method is sustained first-person investigation under the guidance of a specific practice lineage, with criteria for progress that are themselves first-person. Does felt contraction loosen, does identification with the bounded anu weaken, does recognition of one's nature as the ground of appearance stabilise across ordinary activity rather than remaining confined to isolated moments.

These are real criteria, in the sense that practitioners across a long tradition report convergence on them and disagree sharply when a claimed recognition does not hold up under further scrutiny. But they are not criteria a third party can check by attaching electrodes or running a controlled trial, because the object of investigation, awareness's own nature, is not the kind of thing that shows up as a variable in someone else's dataset. It is, on the tradition's own account, that within which any dataset whatsoever appears.

None of this means first-person claims are beyond scrutiny or that anything reported from contemplative practice should be accepted uncritically. Contemplative traditions are themselves demanding about internal consistency, about ruling out claims that amount to little more than a temporarily altered mood mistaken for recognition. About distinguishing genuine shifts in identification from mere conceptual assent to a doctrine. The point is narrower: first-person recognition claims and third-person mechanistic claims are validated by different kinds of evidence. Asking one method to confirm or refute the other's central claim is asking a ruler to weigh something.

A practice note: noticing the constructed character of perception

Since this article is comparative and theoretical rather than a manual, one modest protocol is enough to make the constructed character of perception available to direct notice, without asking that noticing to settle the metaphysical questions above.

Sit somewhere with a plain, evenly lit surface in view, a blank wall or ceiling is sufficient. Stare steadily at a single fixed point for perhaps thirty seconds without letting your eyes wander, then shift your gaze to the plain surface. An afterimage will appear, a shape with definite colour and edge, clearly present in your visual field, and just as clearly not caused by anything currently in front of you.

Hold attention on it as it fades and notice two things at once. First, that the visual system is producing a definite, structured, located appearance from an internal process alone, with no external object corresponding to it in that moment. Second, that ordinary perception, the moment before, did not announce itself as running the same kind of generative process, it simply presented itself as "the wall", with no felt seam between construction and given fact.

A second, quieter version of the same noticing: at some point during an ordinary activity, pause and ask, not analytically but as a direct check, where exactly the sense of "being here, behind the eyes, looking out" is located. Whether that location is given in the same way a colour or a sound is given, or whether it is more like an assumption running continuously underneath perception, never quite examined because never quite needed to be.

Most people, doing this carefully for the first time, find the located, centred sense of self remarkably difficult to pin down as a distinct perceptual content, even though it organises the interpretation of every other content constantly.

From the predictive processing side, both exercises can be described as loosening a high-confidence prior and noticing the model operating in the momentary absence of strong corrective error. From the Pratyabhijna side, both can be described as noticing abhasa as abhasa, appearance manifesting as appearance, and, in the second exercise, directly investigating the anu's assumed location as itself a contracted appearance rather than an established fact. Both descriptions are compatible with what gets noticed. Neither is proved by the noticing.

What the exercise offers is a controlled, low-stakes way to feel, directly and briefly, the gap between how perception presents itself and what both frameworks, from their very different starting points, argue is actually going on.

For practitioners

Treat the parallels in this article as an invitation to look more carefully, not as a settled conclusion to adopt on authority, whether that authority is a neuroscience paper or a centuries-old commentary.

When you read predictive processing literature, ask of any given claim whether it concerns mechanism, testable and revisable by experiment, or smuggles in a claim about the nature of experience itself, not testable by the methods being used. Friston's writing in particular slides between these registers, and it repays reading slowly rather than quoting from summaries.

When you read Pratyabhijna texts, or work with a teacher in that lineage, keep the reverse habit: notice which claims are offered as descriptions to be recognised through your own sustained attention. Resist converting them prematurely into either literal cosmological assertions to be believed on authority or loose metaphors enjoyed without rigour. Utpaladeva and Abhinavagupta wrote careful, technical argument; read it that way.

If you compare the two frameworks in your own thinking, use the table above as a check on yourself. For any parallel you find compelling, ask which column it belongs in, alignment or divergence, and what its testability status actually is. If you catch yourself thinking this proves the other one true, stop and locate where the category error crept in. It is almost always where a structural resemblance between two models gets quietly promoted to an identity between what the two models are claiming.

In your own first-person investigation, whichever tradition or none you draw on, the afterimage exercise and the located-self enquiry above are safe, minimal starting points for noticing the constructed character of ordinary perception directly, rather than only accepting it as a theoretical claim. Do this noticing for its own sake, and let it inform how carefully you read both kinds of literature afterward, rather than treating either as having already told you what you will find.

Hold the distinction this article insists on as your working discipline: mechanism and ground are different questions, answered by different methods. The respect due to each tradition is best paid by refusing to let one quietly do the other's job.