ivory depths

Introduction to Transcendental Logic

I. Of Logic in General

Our knowledge springs from two main sources in the mind: the capacity for receiving representations through being affected by the world, and the power of cognizing by means of these representations. Through the first, an object is given to us; through the second, it is thought. Intuition and conception constitute the elements of all knowledge, so that neither conceptions without an intuition corresponding to them, nor intuition without conceptions, can afford us knowledge.

Kant articulates this as two irreducible faculties: sensibility (receptivity) and understanding (spontaneity). He insists they cannot exchange their proper function: “Understanding cannot intuite, and the sensuous faculty cannot think.” And he delivers the famous dictum: “Thoughts without content are void; intuitions without conceptions, blind.”

The reconstruction preserves the functional insight while dissolving the faculty separation.

The Prediction Machine and the Markov Blanket

The brain is a prediction machine. It continuously generates expectations about incoming sensory signals and updates itself in light of prediction errors. This process is not divided into a passive reception phase and an active thinking phase. There is no fetch/execute cycle. Neural computation is continuous, bidirectional, and self-modifying through Hebbian learning. The very act of processing IS learning. There is never an absence of input.

What Kant calls “sensibility” and “understanding” are not two separate faculties but aspects of one continuous predictive process, describable through the Markov blanket formalism:

  • Sensibility corresponds to the Markov blanket: the statistical boundary where the organism meets the environment. Sensory transducers (retina, cochlea, mechanoreceptors) convert physical energy into neural signals. This is the interface, where predictions meet evidence.

  • Understanding corresponds to the generative model: the internal states that infer the external causes of blanket signals. Hierarchical cortical processing, from sensory-bound representations (V1, A1) to abstract pattern completion (prefrontal cortex, temporal pole).

But these are not two systems. The blanket is itself part of the model V1 already has evolutionary priors built into its architecture, as established in the Aesthetic. And the model’s high-level predictions flow back down to shape what the blanket “receives.” There is a genuine functional gradient from “closer to the blanket” (more sensory-bound, more concrete) to “deeper in the model” (more abstract, more amodal). But there is no clean boundary, and there is no passivity anywhere.

All Use Cases Supported

This framework explains naturally what Kant’s two-faculty model cannot:

  • Visual illusions: the generative model’s prediction overrides the sensory signal. Not a malfunction, the system generates the best posterior given its priors. Working as designed.
  • The McGurk effect: multimodal predictions are integrated. The system finds the best overall posterior across auditory and visual evidence. Normal operation.
  • Phantom limb pain: the generative model still represents the limb and generates predictions for it. No evidence arrives to disconfirm. The prediction persists.
  • Dreams: the generative model runs forward without sensory constraint. Predictions without likelihood. The same process that produces waking experience, unmoored from the blanket.
  • Blindsight: cortical prediction pathway is damaged, but subcortical paths still carry blanket signals to the model. Partial function of the same system, not a separate faculty at work.

Kant’s dictum reformulated: predictions without evidence are unconstrained (dreams, hallucination, metaphysical speculation); evidence without a model is noise (no such thing in a functioning brain, because the model IS the architecture through which evidence flows). Both contributions are necessary for knowledge, but they are contributions of one system, not products of two faculties.

The Functional Asymmetry That Survives

Though the faculty separation dissolves, a genuine functional asymmetry remains:

  • Sensory transduction really is different from cortical model-running. The retina converts photons to neural signals. This is a physical interface with the world, distinct in kind from abstract inference.
  • The cortical hierarchy really does gradient from sensory-bound to abstract. Lower layers handle concrete features; higher layers handle categories, relations, plans.
  • You DO need both boundary states (evidence) and internal states (model) for knowledge. The system can run without evidence (dreams) but produces unconstrained output. Evidence without model is impossible in a functioning brain.

What is lost from Kant: the clean binary, the passivity of sensibility, the claim of two irreducible faculties.

What is kept: the insight that knowledge requires both evidence and model, that they make genuinely different contributions, and that the understanding (model) must not be applied beyond the domain where evidence can constrain it.

Pure and Applied Logic

Kant distinguishes pure general logic (the formal laws of thought, abstracted from all content) from applied logic (how thinking actually occurs under empirical psychological conditions, attention, memory, habit, error, inclination).

Under the reconstruction:

  • Pure general logic corresponds to the formal properties of inference itself: the probability axioms, Bayes’ theorem, the rules of consistent updating. These hold regardless of content, regardless of what the model is about.

  • Applied logic corresponds to how inference actually occurs in biological systems: subject to attentional bottlenecks, memory limitations, emotional biases, fatigue, habit. The brain approximates Bayesian inference but does so imperfectly, under biological constraints. The approximation errors are the subject of applied logic.

Kant insists that pure logic “draws nothing from psychology.” The reconstruction agrees in a specific sense: the formal structure of optimal inference (Bayes’ theorem) does not depend on how any particular brain implements it. But the reconstruction adds: the formal structure is never perfectly realized in biological systems. The gap between optimal inference and actual neural computation is where applied logic lives.

II. Of Transcendental Logic

General logic abstracts from all content and deals only with the form of thought. Transcendental logic, by contrast, concerns itself with the a priori conditions under which objects can be thought, not the form of thought in general, but the structures the cognitive system brings to its encounter with experience.

Under the reconstruction: transcendental logic is the investigation of the generative model’s STRUCTURE, the architectural priors, the hierarchical organization, the specific computational principles that determine how sensory evidence is organized into coherent experience. It is not about the formal rules of inference (that is pure logic / the probability axioms) but about the specific model that the brain implements, the categories, the principles, the schematism by which abstract model structure meets concrete sensory prediction.

Kant distinguishes “transcendental” from merely “a priori”: a transcendental cognition is one that concerns the a priori POSSIBILITY of cognition, not just any a priori content. Space is a priori but not itself transcendental; the KNOWLEDGE that spatial representation is a priori and makes experience possible is transcendental.

The reconstruction preserves this meta-level distinction. The priors embedded in neural architecture are a priori (pre-experiential). The INVESTIGATION of those priors, what they are, where they come from, what they make possible, where they fail, is transcendental. This investigation is what we are conducting.

III. Of the Division of General Logic into Analytic and Dialectic

Kant asks: what is truth? The correspondence of cognition with its object. But he immediately shows that a universal, content-independent criterion of truth is impossible, because truth is precisely about content, and a formal criterion abstracts from content.

Formal logic (non-contradiction) is a necessary but not sufficient condition of truth. A cognition can be logically impeccable and still false to its object. Formal validity is the negative test only.

Under the reconstruction: a model can be internally consistent (satisfying the probability axioms, free of contradiction) and still fail to predict the world. Consistency is necessary but not sufficient. Calibration, the model’s predictions matching observed frequencies is the positive test, and calibration requires evidence. There is no content-independent criterion of calibration, because calibration is precisely about the fit between model and world.

Kant then introduces the analytic/dialectic division:

  • Analytic: logic as a CANON, a negative test, a tool for evaluating cognitions that have independent empirical grounding. Legitimate use.

  • Dialectic: logic as an ORGANON, an attempt to use formal principles to PRODUCE knowledge about objects, without independent empirical grounding. Illegitimate use. The “logic of illusion.”

This maps directly to the convergence continuum:

The prediction machine operates identically regardless of domain. It generates predictions, compares them with evidence, updates. In falsifiable domains (where evidence constrains), this produces reliable models, analytic use. In unfalsifiable domains (where no evidence can arbitrate), the same process produces unconstrained speculation dialectical use.

The “seductive charm” Kant describes, the temptation to use formal principles to generate knowledge without empirical grounding, is a structural feature of the prediction machine. The generative model has no built-in stop mechanism at the boundary of the falsifiable. It generates predictions beyond the evidence with the same confidence it generates predictions within it. The system cannot feel the moment it leaves the domain of the analytic and enters the dialectic. This is why the critique is necessary: the system cannot diagnose its own overreach from the inside.

IV. Of the Division of Transcendental Logic

Transcendental analytic will examine the elements of pure cognition the categories and principles by which the understanding organizes experience. This is the logic of truth: no cognition can contradict these principles without losing all reference to objects.

Transcendental dialectic will examine what happens when these principles are extended beyond experience. The understanding, seduced by the universality of its own principles, attempts to judge objects in general objects not given in experience, perhaps incapable of being given. In this territory, the understanding becomes dialectical.

Under the reconstruction:

  • Transcendental analytic investigates the generative model’s structure insofar as it WORKS, the architectural features (priors, hierarchical organization, prediction error minimization) that constitute experience and produce reliable models within the domain of falsifiable evidence.

  • Transcendental dialectic investigates what happens when this same architecture operates beyond the evidence, generating confident predictions about totalities, absolutes, and unconditioned entities that no observation could confirm or disconfirm. Here the generative model encounters its own limits: the architectural limit (not enough compute), the formal limit (Godelian self-reference), and the thermodynamic limit (entropy). The dialectic maps the terrain where these limits produce characteristic failures, the antinomies, paralogisms, and ideals of pure reason.

We proceed now to the analytic: the investigation of the model’s categories and principles.