Introduction

I. Of the Relationship Between Prior and Empirical Knowledge

That all our knowledge begins with experience there can be no doubt. The faculty of cognition is awakened into exercise by means of objects which affect our senses, producing representations which rouse our powers of understanding into activity, to compare, to connect, to separate, and so to convert the raw material of sensuous impressions into a knowledge of objects.

But, though all our knowledge begins with experience, it by no means follows that all arises out of it. Our empirical knowledge is a compound of that which we receive through impressions and that which the faculty of cognition supplies from itself. In the language of statistical inference: every act of knowing is a posterior, composed of a likelihood (what we receive) and a prior (what cognition supplies). This decomposition, posterior = prior x likelihood, is the formal articulation of a distinction that, in lived cognition, is invisible.

For the prior is not stored separately and then combined with incoming data. It is the architecture itself. Consider the primary visual cortex: its orientation-selective columns constitute ready-made knowledge of what a straight line is. This is not a parameter that gets multiplied by retinal input. It IS the synaptic wiring, the receptive field structure, the columnar organization through which visual data flows. We never observe these priors directly. We witness them only by their effects edges detected, objects segmented, motion tracked. The multiplication sign in Bayes’ theorem is a fiction of the formalism. In the actual tissue, prior and likelihood are fused.

This inseparability is not merely phenomenological but architectural. It is, as Kant rightly observed, a distinction that requires careful theoretical work to articulate, which is precisely the work the present investigation performs.

By the term “knowledge a priori” we shall understand knowledge that the cognitive system supplies from itself, prior to any individual experience. This a priori is not, however, independent of ALL experience in the absolute sense Kant intended. The priors embedded in neural architecture were shaped by hundreds of millions of years of sensory experience at the species level. They are the species’ accumulated posterior, crystallized into the individual’s wiring. What is a priori relative to a single life is deeply a posteriori relative to evolutionary history.

The a priori is therefore biological, not logical. And being biological, it is fragile. We can observe its degradation and absence in lesions, birth defects, and developmental disorders. When the prior-generating architecture is damaged, the capacity to model the environment collapses not into a different but equally valid form of experience, but into profound disruption. This confirms the functional role Kant identified: these structures ARE conditions of coherent experience. But the necessity is biological, the system fails without them, not logical, as though their absence were inconceivable.

The a priori operates at two levels:

Content priors. Specific expectations embedded in neural architecture by evolutionary and developmental processes: edge detection, face selectivity, object permanence, intuitive physics. These are prior to individual experience but could have been otherwise on a different evolutionary path. They are contingent in origin but functionally absolute within a human life.

The operating principle. The brain’s fundamental mode of operation Bayesian surprise minimization, is itself a prior in the philosophical sense. Not a prior about any particular feature of the world, but the prior that the world is the kind of thing that admits of prediction. The prior that there are regularities to exploit. This meta-prior is the condition under which experience becomes possible at all. Without it, there is no updating, no learning, no perception, no coherent experiencer. It is what makes there be a subject of experience, and it is the deepest structural claim this investigation will defend.

We are, at root, machines for entropy minimization. This is how we structure the universe internally, through models that compress and predict, and externally, through the technologies and practices by which we impose local order. But the universe we inhabit is governed by the second law of thermodynamics: entropy grows. We are therefore in a state of constitutive opposition to the fundamental tendency of our reality. Cognition is not a neutral mirror of the world. It is an anti-entropic activity embedded in an entropic cosmos, building order against the grain, and losing.

II. The Human Intellect Possesses Certain Cognitions A Priori

The question now arises: by what criterion may we distinguish what the cognitive architecture supplies from what experience teaches?

Kant proposed necessity and strict universality as infallible tests. Under the present reconstruction, these marks are not abolished but reconceived. They are not binary properties but the limiting behavior of a statistical process.

The key variable is contextual rigidity. In domains with maximally constrained contexts, formal mathematics, pure logic, variation between individual reasoners approaches zero. The logic appears necessary and universal because the constraints are tight enough to eliminate divergence. In moderately constrained domains, experimental science shared evidence forces convergence, though it requires work. In loosely constrained domains, colloquial language, everyday reasoning, priors diverge and people can come to reside in mutually exclusive worlds. And in domains with no constraining mechanism at all, traditional metaphysics, there is no convergence, only endless contest.

Necessity, then, is what convergence looks like when contextual constraints eliminate variation. It is not a different kind of knowledge but the same inferential process (surprise minimization) operating under maximally constraining conditions. The purer and more contextually rigid we get, the more robust and shared is the logic.

Kant asks us to consider causality: “Every change must have a cause.” He insists this cannot be derived from experience, since experience gives only regularity, not necessity. We agree that causality is a category of the mind, an inherent feature of the predictive architecture, not learned from any particular sequence of events. The brain is a prediction machine, constantly and multimodally connecting input and intermediary processing. Causal binding is what this architecture DOES.

But the content of causal judgement is statistical and individual. The form is universal: every functioning brain imposes causal structure. The application varies: a loud noise binds to fireworks for one person, to gunfire for another, to nothing coherent for someone with relevant processing damage. “Every change has a cause” is true at the formal level, the brain will seek causes. It is indeterminate at the content level, which cause it finds depends on the individual’s particular priors and evidence.

Kant was therefore right that causality is not derived from Humean association. But he was wrong that it delivers the same necessary judgements for all. The category is universal; its application is probabilistic and personal.

As for Kant’s demonstration that certain conceptions, space, substance cannot be thought away: this is correct, but the explanation is architectural, not transcendental. Space cannot be annihilated in thought because spatial representation is wired into the very structure of the cognitive system. It is not that the concept of space is logically necessary; it is that the neural architecture cannot operate without it. The inconceivability is a fact about the brain, not about logic.

III. Philosophy Stands in Need of a Science of Cognition

Certain of our cognitions appear to rise above the sphere of possible experience. By means of conceptions to which no object in experience corresponds, we seem to extend our judgements beyond their proper bounds. In this domain, where experience affords neither instruction nor guidance, lie the traditional problems of metaphysics: God, freedom, immortality.

Kant correctly identifies that metaphysics, as traditionally practiced, proceeds dogmatically, taking upon itself ambitious tasks without first investigating whether reason is capable of them. The present investigation shares Kant’s diagnostic aim: to determine what cognition can and cannot achieve, and why.

But we ground this investigation differently. Complete knowledge of reality is impossible, and the impossibility converges from three independent directions:

The architectural limit. Approximately 86 billion neurons cannot model a universe of approximately 10^80 particles. The gap between knower and known is, in the first instance, a capacity problem. Tools writing, mathematics, computing, are extensions of the cognitive processing faculty that increase capacity but cannot close the gap.

The formal limit. Any system powerful enough for self-reference contains truths it cannot prove about itself. The brain is part of the universe it models. A complete model of a system that includes the modeler would require the modeler to model itself modeling, and so on without end. The liar paradox can be coded into any system capable of self-reference. This is not a contingent limitation but a formal one, demonstrated by Godel: it holds regardless of scale.

The thermodynamic limit. We are entropy-minimizing structures in an entropy-increasing cosmos. Cognition imposes order; reality’s fundamental tendency dissolves it. We build models against the grain, and the grain always wins in the end.

These three limits are not independent problems but expressions of the same universe, one where finite, self-referential, anti-entropic systems attempt to model something that exceeds them quantitatively, outruns them formally, and opposes them thermodynamically.

The thing-in-itself, Kant’s unknowable noumenal reality, is therefore reinterpreted not as a metaphysical mystery but as the convergence of these three limits. It is what the model cannot hold (architectural), what the self-referential system cannot prove about itself (formal), and what the entropic process dissolves faster than cognition can reconstruct (thermodynamic). We can model the world better or worse, extend our toolsets, approach asymptotically, but completeness is ruled out on all three fronts simultaneously.

Kant’s dove metaphor, that reason, like a dove feeling the resistance of air, imagines it would fly better in a vacuum, translates directly. Without the resistance of evidence (empirical constraint, experimental falsifiability), reasoning does not become more free. It becomes unconstrained, and unconstrained inference is not flight but free fall. The dove needs the air. Reason needs evidence. Where evidence runs out, reason does not soar; it spins.

IV. Of the Difference Between Analytical and Synthetical Judgements

Kant distinguishes two kinds of judgement. In an analytical judgement, the predicate is contained in the subject: “All bodies are extended.” In a synthetical judgement, the predicate adds something new: “All bodies are heavy.”

This distinction exists but is contextual, not absolute.

In Bayesian terms, whether a predicate is “contained in” a concept or “adds something new” depends on how tightly those features are bound in the agent’s probabilistic model. Concepts are statistical clusters shaped by experience, evolution, and architecture, not fixed definitions with sharp boundaries. A judgement that is analytic for one person (whose model tightly binds those features) may be synthetic for another (whose model has not yet learned that binding).

Post-Kantian philosophy has already demonstrated this instability. Quine showed that no judgement is immune to revision; the analytic/synthetic boundary shifts with the total web of belief. The present reconstruction goes further: it grounds this instability in the statistical nature of concepts themselves.

What survives is not the specific distinction but the meta-process. The invariant across all judgement, analytic and synthetic, a priori and a posteriori, is surprise minimization. The brain generates predictions and updates them in light of evidence. Whether a given judgement is “merely unpacking what was already known” (analytic) or “genuinely adding something new” (synthetic) is a matter of degree relative to a particular model at a particular time, not a fixed property of the judgement itself.

For synthetical judgements a priori, Kant’s central concern, the question becomes: how can the cognitive system generate expectations that go beyond what any particular experience teaches, yet prove reliable in structuring experience? The answer is evolutionary and developmental shaping. The “unknown = X” upon which the understanding rests when it synthesizes cause and effect is the phylogenetically tuned architecture of the predictive brain, shaped by millions of years of organisms encountering a world in which events have causes.

V. In All Theoretical Sciences, Synthetical Principles Are Contained A Priori

Kant observes that mathematics, physics, and metaphysics all contain synthetic a priori principles. The proposition 7 + 5 = 12, the conservation of matter, the claim that the world must have a beginning all go beyond mere analysis of concepts, yet claim to hold independently of particular experience.

Under the present reconstruction, these examples occupy different positions on the convergence continuum:

Mathematics sits at the maximally constrained end. The formal rules are rigid enough that variation between reasoners approaches zero. “7 + 5 = 12” appears necessary because the context eliminates all ambiguity. This is the limiting case of statistical convergence, not a different kind of knowledge.

Physics occupies the moderately constrained middle. Conservation of matter is a synthetic principle that holds under experimental conditions. It is falsifiable, and its apparent necessity reflects the depth and consistency of the evidence supporting it. It is robust because the domain is tightly constrained by experimental methodology.

Metaphysics sits at the unconstrained end. “The world must have a beginning” is a proposition for which no experimental arbitration is possible. Different thinkers, with different priors, will arrive at different conclusions, and there is no shared evidence to force convergence. This is why metaphysics has remained, as Kant notes, in a state of “vacillating uncertainty and contradiction.”

VI. The Universal Problem of Pure Reason

Kant formulates the central question: “How are synthetical judgements a priori possible?”

Under the Bayesian reconstruction, this question transforms. It becomes: How does a finite, biological, entropy-minimizing system generate reliable expectations about a world that exceeds its modeling capacity?

The answer is: through the interplay of evolutionary shaping (which builds structural priors into the architecture), developmental tuning (which calibrates those priors to the local environment), and ongoing Bayesian updating (which refines the model in light of incoming evidence). The “synthetic a priori” is not a mysterious third category of knowledge. It is the prior distribution that the organism brings to every encounter with experience, shaped by evolutionary history, embodied in neural architecture, invisible from the inside, witnessed only by its effects.

Kant notes that metaphysics, as a natural disposition, will always exist that human reason cannot help but pose questions it cannot answer. This observation survives fully. The predictive brain, in its drive to minimize surprise, will always attempt to extend its models beyond the evidence. It will always generate hypotheses about the unconditioned, the total, the absolute, because the architecture that minimizes surprise locally has no built-in mechanism to halt at the boundary of the falsifiable. Transcendental illusion is not a correctable error but a structural feature of the prediction machine: the same process that makes science possible makes metaphysical overreach inevitable.

Against this overreach, Kant proposes the Critique as a tribunal. We propose something different: not a tribunal that delivers final verdicts, but a calibration, an investigation into the architecture’s capacities and limits, grounded in what we now know about how the system actually functions. The question is not “what can reason as such achieve?” but “what can THIS system, biological, finite, self-referential, entropic reliably model, and where does its reliability break down?”

The answer is domain-dependent: where hypotheses are falsifiable and evidence is shared, the system converges on reliable models. Where mechanisms are loose and evidence is absent, it diverges into mutually exclusive worlds. The critical investigation maps this terrain, not once and for all, but as an ongoing scientific enterprise, itself subject to revision.

VII. Idea and Division of This Investigation

The present investigation, like Kant’s, concerns itself not with objects but with the mode of our cognition of objects, so far as this mode can be understood prior to any particular experience.

We retain Kant’s division into a Doctrine of Elements and a Doctrine of Method. The Doctrine of Elements examines the components of the cognitive architecture: first, the forms of sensory processing (what Kant calls the Transcendental Aesthetic, space and time), and then the structures of understanding (what Kant calls the Transcendental Logic, the categories and principles by which experience is organized).

But where Kant’s investigation proceeds by pure philosophical argument, ours draws on the empirical sciences of cognition. This does not make it circular, using empirical science to investigate the conditions of empirical knowledge, because the investigation is itself situated on the convergence continuum. It makes falsifiable claims about the architecture of cognition, and it is answerable to evidence.

There remain two sources of human knowledge: sense and understanding. By the former, objects are given to us; by the latter, thought. These probably spring, as Kant suspects, from a common root, and the Bayesian reconstruction names that root: the unified predictive process of surprise minimization, which generates both sensory predictions (perception) and conceptual predictions (thought) as expressions of a single underlying architecture.

The transcendental doctrine of sense, the investigation of space and time as forms of intuition, must come first, because the conditions under which objects are given must precede those under which they are thought.