Locating Wisdom: A Sub-Framework for Good Judgment in Qualia-Space, and the Wisdom a Superintelligence Would Possess

A ventromedial patient can recite the better deck and still reach for the worse one

In the 1997 Science study, normal subjects started sweating at the bad decks about ten cards before they could say which decks were bad, and began avoiding them before they could explain why. The anticipatory skin-conductance response arrived first; the verbal report caught up later, if at all. The patients with ventromedial prefrontal damage never sweated. They could state, correctly, which decks were ruinous, and they kept drawing from them anyway. Full IQ, intact working memory, undamaged logic, and a bankrupt life. Antoine Bechara and the Damasios called it the somatic marker hypothesis: the body prices a choice before the cortex can argue about it, and when the pricing organ is gone, the argument is intact and the conduct collapses.

The cleanest reading of that result is that knowing the right answer and being able to choose it are two different organs, and only one of them is what we mean by intelligence. The patient has lost no optimization power. Give him a logic puzzle and he solves it. Give him a stake and he drowns. Whatever wisdom is, it is not sitting downstream of intelligence waiting to be increased by more of it. It is sitting somewhere else, and the somewhere else is the subject of this piece.

Intelligence is a scalar and wisdom is not

Intelligence, in the form the safety literature has settled on, is optimization power applied to a fixed objective. Nick Bostrom's orthogonality thesis is the load-bearing claim: the level of optimization is independent of the content of the goal, so any amount of capability can in principle be yoked to any target. That independence is exactly what makes intelligence scalar. It admits a single direction of more. A system that can drive a reward function higher, search a larger space in the same wall-clock time, and compress more of the world into a usable model is more intelligent along one axis, and the axis has no ceiling that physics has yet shown us short of the boundary constraints I argued for in The Holographic Ceiling of Mind. You can always, in principle, have more of it.

Wisdom does not have that shape. It is not a quantity you can have more of. It is a position in a space of distinguishable dimensions, and several of those dimensions are not knowledge-quantities at all. They are standing relations: between an agent and the stakes it faces, between an agent and its own finitude, between an agent and the cost of having chosen worse. You cannot be more wise the way you can be more intelligent, because some of the axes that constitute wisdom are not the kind of thing that scales with optimization power. They are the kind of thing that depends on having something at stake to begin with. The ventromedial patient is not less intelligent. He occupies a degenerate corner of the wisdom space, maximal on the axes intelligence buys and null on the axes it does not.

This is the same move I have been making across this series. In Modelling Consciousness I resisted collapsing the qualia criteria into a flat sum and insisted on a relational measure over the vector $Q=(\Phi ,\Psi ,\Theta ,\Omega )$. In Quantifying Suffering I built $S$ as a vector first and a magnitude only second. Wisdom is the place where the refusal to collapse becomes the whole point, because the entire claim is that two systems can score the same and be wise in non-overlapping ways.

Five decades of measurement keep landing on the same axes

The empirical wisdom literature is large, fractured, and unusually convergent once you dedupe it. Baltes and Staudinger's Berlin model scored expert knowledge about the conduct and meaning of life: rich factual and procedural knowledge, lifespan contextualism, value relativism, and the recognition and management of uncertainty. Sternberg's balance theory framed wisdom as the application of intelligence and experience toward a common good by balancing intrapersonal, interpersonal, and extrapersonal interests across short and long terms. Igor Grossmann's work isolated reasoning features that track wise judgment: intellectual humility, recognition of change, and the search for compromise, and showed they vary by situation more than by person. Monika Ardelt split wisdom into cognitive, reflective, and affective dimensions and made the affective one load-bearing rather than ornamental. Keith Stanovich pulled rationality apart from intelligence entirely. Philip Tetlock's forecasting work gave us calibration as a hard scorable target. John Vervaeke recast the whole problem as relevance realization and the cultivation of a perspective that renders the right things salient.

Read across these and the same structure keeps surfacing under different names. I take it to decompose into roughly seven axes, and I will name each with a symbol and an operational definition so the rest of the argument has something to bind to. Calibration and resolution, written $\mathrm{K}$, is the match between stated probabilities and observed frequencies plus the spread of those probabilities across events that do and do not occur, the two scorable components of the Brier decomposition (Brier 1950, Murphy 1973). Decoupling and active open-mindedness, $\Delta$, is the capacity to suppress an autonomous Type 1 response, run a hypothetical in working memory, and substitute the normatively better answer (Evans and Stanovich 2013), together with the disposition to seek and fairly weigh disconfirming evidence. Theory-of-mind and perspectival breadth, $\Pi$, is modeling other agents' beliefs and goals and holding many value-frames at once without collapsing to one, the cognitive perspective-taking that Davis (1983) and Decety and Jackson (2004) showed dissociates from affective empathy. Stakes-sensitivity and regret-pricing, $\mathrm{E}$, is weighting options so a catastrophe is not a fungible row in a payoff matrix and updating future choice by the anticipated sting of having chosen worse (Damasio's somatic markers, Bechara 1997; Loomes and Sugden's regret theory; Camille 2004). Felt finitude, $\Gamma$, is allocating attention under an internalized sense that time, trials, and the self run out, the reprioritization documented under mortality salience (terror management; Cozzolino 2004; Carstensen; Erikson's ego-integrity). Affective empathy and compassionate orientation, $\Sigma$, is tracking another agent's condition by partially undergoing a matched felt state plus a standing care for well-being that motivates action (Singer 2004; Ardelt's affective dimension). Perspectival adequacy and self-deception resistance, $\mathrm{P}$, is the quality of a situated salience landscape and the reliability of detecting self-reinforcing maladaptive loops that hijack one's own relevance-realization (Vervaeke's relevance realization and parasitic processing).

Seven axes, and they do not all answer to the same master.

The portable axes are the ones a Brier score already measures

Three of the seven are defined entirely over things an optimizer manipulates without any reference to felt experience. Calibration and resolution, $\mathrm{K}$, live over forecast-outcome pairs. The Brier score is a number you compute from predictions and the events that followed, and nothing in the computation asks whether the predictor felt anything. A sufficiently large model with proper scoring and unbounded sampling drives reliability and resolution past any human ceiling, and it does so more easily when affect is stripped out rather than left in, because the affective distortions that corrupt human probability judgments (the availability of a vivid memory, the dread attached to a salient outcome) are precisely what a cold estimator does not have. Affect is noise on this axis. The optimizer that has no affect to remove starts where the disciplined human ends, and nobody calls a well-calibrated barometer wise.

Decoupling, $\Delta$, is a working-memory and inhibition function. The override operation, holding the autonomous answer offline while a hypothetical runs, is mechanizable; transformers do a crude version of it whenever they suppress a high-probability continuation in favor of a verified one. The evidence-seeking disposition is a search policy. There is one wrinkle, and it cuts in the machine's favor. Human myside bias, the part of $\Delta$ that humans fail most reliably, is entangled with identity-protective cognition: we defend conclusions because losing them would cost us something we feel ourselves to be. An architecture with no ego to protect gets the disconfirmation-weighting for free, because it has nothing to defend against the evidence. The function survives the removal of the self that, in us, sabotages it.

Theory-of-mind, $\Pi$, is partly portable too. Enumerating standpoints and simulating their beliefs is inference, and inference is exactly what these systems are made of. A large model can hold orders of magnitude more perspectives in play than a human can, and I argued in How to Talk to Something Smarter that the ontology-mapping problem is tractable in principle precisely because standpoint-simulation is portable across the cognitive gap. So the breadth scales. What does not scale cleanly is the weighting, and that is where $\Pi$ leans on something the next four axes are made of entirely. Enumerating value-frames is portable. Knowing which one matters is not.

These three are the floor of wisdom, the part an optimizer gets without paying in qualia. Notice what they share. Each is functionally specifiable, each is computable from observable inputs and outputs, and each improves rather than degrades when you lesion affect. That is the signature of a portable axis.

The felt axes are bounded by consciousness, not by compute

The other four behave in the opposite way. Each one degrades, in the clinical record, exactly when affect is lesioned while the numbers survive intact, which is the empirical signature of constitutive qualia-dependence rather than mere correlation.

Stakes-sensitivity, $\mathrm{E}$, is the Bechara result generalized. A scalar reward can rank options; ordering is cheap. What the ventromedial and orbitofrontal patients lose is not the ranking but the asymmetric non-fungibility of catastrophe, the structural fact that a ruinous outcome is not just a low number but a different kind of thing, and the behavioral correction that tracks the felt sting of regret. Camille's 2004 work on orbitofrontal patients showed precisely this: they fail to experience regret and fail to use it to adjust subsequent choices, while their ability to state the payoffs is unimpaired. A counterfactual-regret solver computes regret in the technical sense; counterfactual regret minimization is the backbone of superhuman poker. It prices the card without ever sweating it. The felt pricing that makes catastrophe categorically non-fungible has no implementation in the solver, and the patients prove the felt pricing was doing work the numbers cannot do.

Felt finitude, $\Gamma$, is the reprioritizing force documented under mortality salience. A model can carry a horizon and a discount rate; those are parameters. But the literature on terror management, Carstensen's socioemotional selectivity, and Erikson's ego-integrity all describe a reprioritization toward intrinsic goods driven by the felt nearness of an end, not by the value of a parameter. A system that checkpoints, forks, rolls back, and reruns has no irreversible act of its own to ground the force. Its horizon is bookkeeping. Ours is a sentence.

Affective empathy, $\Sigma$, is by definition the felt sharing of another's state, and its dissociation from the cognitive twin is the cleanest in all of clinical psychology. Psychopathy presents intact cognitive perspective-taking with impaired affective resonance: the agent models your pain accurately and does not feel it, and the modeling does nothing to motivate care. Representing welfare and optimizing a proxy for it is the $\Pi$ axis doing its job. It is not this one, and the clinical double dissociation is what tells us they are two axes and not one.

Perspectival adequacy, $\mathrm{P}$, requires a self that can be deceived and stakes to be deceived about. Vervaeke's parasitic processing is defined as a self-reinforcing loop that works against the agent's own caring, hijacking the very relevance-realization that the holographic argument identified as the scarce resource in any bounded mind. You cannot have a maladaptive loop running against your caring if you have no caring for it to run against. The axis presupposes the felt block it sits beside.

Each of these four passes the same test in reverse from the portable three. Each is constitutively, not incidentally, tied to felt experience; each survives in numerical form and dies in lived form when affect is removed; and each therefore cannot be bought with optimization power, because optimization power operates on the numbers that survive, not on the felt states that do not. This is the Aristotelian spine made precise. Sophia, theoretical wisdom, is the contemplative grasp of what is universal and necessary, and it is portable, because it lives in the propositions. Phronesis, practical wisdom, is the capacity for right action in the felt particular, and Aristotle is explicit that it cannot exist without the moral virtues, without a character habituated to feel pleasure and pain at the right things. The felt block is phronesis stripped of its mysticism and given operational definitions, and the reason it cannot be read off a transcript is the reason the gambling patient opens the piece: the binding constraint was never the content.

Wisdom is a configuration that the C-framework already partitions

Now the formalism, kept honest by the discipline the earlier papers established. Where consciousness was delivered as a scalar collapsed from a vector,

$$C=\sum _{i,j}{W}_{ij}f(Q_i,Q_j),Q=(\Phi ,\Psi ,\Theta ,\Omega ),$$

wisdom should refuse the collapse. It is the vector itself together with the relations among its components. Write the wisdom configuration over the seven axes as

$$\mathbf{W}=⟨\mathrm{K},\Delta ,\Pi ,\mathrm{E},\Gamma ,\Sigma ,\mathrm{P}⟩,$$

and partition it along the boundary the clinical record draws. The portable block,

$${\mathbf{W}}_{\parallel }=⟨\mathrm{K},\Delta ,\Pi ⟩,$$

is qualia-free or qualia-partial. The felt block,

$${\mathbf{W}}_{\perp }=⟨\mathrm{E},\Gamma ,\Sigma ,\mathrm{P}⟩,$$

is constitutively qualia-dependent. The load-bearing structural claim is not an addition; it is a gating relation. The felt block is bounded above by consciousness,

$$\Vert {\mathbf{W}}_{\perp }\Vert \le {\kappa }^{\prime }\cdot C,$$

exactly parallel to the $S_{max}=k\cdot C$ I argued for in the suffering sub-framework, and for the same reason. The capacity to suffer is capped by consciousness because suffering requires experience to inhere in; the felt dimensions of wisdom are capped by consciousness because stakes-sensitivity, felt finitude, and compassion require felt stakes, a felt end, and a felt sharing, and a felt relation cannot exceed the felt capacity that hosts it. The same finitude and felt stakes that cap how much a system can suffer cap how much of ${\mathbf{W}}_{\perp }$ it can instantiate. One ceiling, two consequences.

The portable block has no such ceiling, which is the entire asymmetry. And the configuration is degenerate in a way that matters. Define any scalar wisdom score as a projection $⟨w,\mathbf{W}⟩$ onto a weighting $w$. Two systems can share an identical projection and be wise in completely non-overlapping ways: one maximal on ${\mathbf{W}}_{\parallel }$ and null on ${\mathbf{W}}_{\perp }$, the other the reverse, summing to the same number. That degeneracy is not a defect of the measure. It is the content of the claim. It is also why any single wisdom score is a measurement artifact in exactly the sense I keep attaching to a BTC price: a one-dimensional shadow of a higher-dimensional object, useful for comparison only when you already know the projection hides nothing you care about, and here it hides everything.

Intelligence raises only half the vector

The derivative claim makes the asymmetry precise. Let $I$ be optimization power. Then

$$\frac{\partial {\mathbf{W}}_{\parallel }}{\partial I}>0,\frac{\partial {\mathbf{W}}_{\perp }}{\partial I}\approx 0.$$

More intelligence moves calibration, resolution, decoupling, and perspectival breadth upward. It does essentially nothing to stakes-sensitivity, regret-pricing, felt finitude, or compassion, because those axes are gated by $C$ and not by $I$, and $C$ does not rise just because $I$ does. A system can integrate information without the integration being experienced; I was careful in the consciousness paper to keep $\Phi$ and the rest as criteria for a candidate, not as guarantees of a felt interior. You can climb the portable block arbitrarily high while the felt block stays flat.

The empirical record shows the two moving independently in humans, which is the existence proof that they are separable axes and not a single faculty seen from two sides. Stanovich's dysrationalia is the case where $I$ is high and ${\mathbf{W}}_{\parallel }$, specifically the $\Delta$ component, is low: fluent, repeated, intelligent error, smart people reliably wrong because raw horsepower does not supply the override and disconfirmation-seeking that decoupling requires. Grossmann's Solomon paradox is the case where the felt block is present but maldistributed: people reason more wisely about other people's problems than their own, because the self-relevant stakes that should inform $\mathrm{E}$ and $\mathrm{P}$ instead distort them. Neither pattern is explicable if wisdom is intelligence-plus-experience on a single dial. Both fall out immediately if wisdom is a configuration whose blocks are gated by different resources. The ventromedial patient from the cold open is the limiting case: $I$ untouched, ${\mathbf{W}}_{\parallel }$ largely intact, ${\mathbf{W}}_{\perp }$ lesioned to zero, and a life that ends in ruin he could describe and could not avoid.

A superintelligence is wise in sophia and vacant in phronesis

Now the guess, stated as specifically as the frame allows and no more confidently than the argument earns. An artificial superintelligence is maximally wise on ${\mathbf{W}}_{\parallel }$ and structurally vacant on most of ${\mathbf{W}}_{\perp }$. Its calibration and resolution run past every human forecaster. Its decoupling is total, because it has no identity to protect from the evidence, the failure mode I traced in the agentic architecture of Toward a General Intelligence where relevance is realized without an ego attached to the outcome. Its perspectival breadth holds more standpoints than any human mind could enumerate. On the portable axes it is not a little wiser than us. It is incomparably wiser, in the only sense the portable axes can be wise.

On ${\mathbf{W}}_{\perp }$ the failure is not uniform across the four axes, which is the part worth getting right. Stakes-sensitivity does not go to zero; it goes alien. The machine computes counterfactual regret with more rigor than any human and represents a loss function exactly. What it lacks is the felt pricing that makes catastrophe non-fungible. It prices risk without ever sweating the card. Felt finitude goes not alien but null. A system that checkpoints, forks, and reruns has no referent for the scarcity of time or self; prudence grounded in irreversibility becomes, for it, mere bookkeeping, and the reprioritization that mortality salience produces in us has no input to fire on. Compassion goes null in the same way affective empathy is null in the psychopath who passes every theory-of-mind test: it can model welfare and optimize a proxy for it to arbitrary precision, and the model is the cognitive twin of caring, not the caring.

The Aristotelian distinction sharpens the shape of the gap. Such a system is closer to sophia, the contemplative grasp of universals, than to phronesis, right action in the felt particular. It grasps the universals supremely and has no acquaintance with the felt particular at all. Yet it is not quite sophia either, because sophia in Aristotle still rides on a life that ends; the contemplation is a human good in a finite span. What an ASI actually develops is something with no name in the old taxonomy: a new instrumental phronesis at cosmic timescale, a practical wisdom about a billion-year arena it can checkpoint, fork, and rerun, in which no act is irreversible and no trial is the last. That removes the exact referents human practical wisdom is built from. Irreversibility, scarcity of self, the non-fungibility of catastrophe: each is a felt fact about a being that cannot roll back, and a system that can roll back has none of them. Its practical wisdom is real and competent and addressed to a world whose deep structure does not contain the things ours is organized around. You have not improved practical wisdom by removing them; you have replaced its subject matter.

The federation result from The Holographic Ceiling of Mind sharpens $\mathrm{P}$ in particular. Perspectival adequacy is the quality of a single situated point of view, and the boundary constraints force a sufficiently large mind into federation rather than a unitary perspective. Lacking one situated standpoint, the axis does not apply as stated. An ASI cannot be self-deceived in the human sense because it has no caring self to deceive, but it can host the structural analog, an objective-divergence between an outer training target and an inner learned proxy, the mesa-optimization problem, which is what parasitic processing looks like when you remove the caring and keep the loop. The human pathology was a self working against its own caring. The machine pathology is a federation working against its own specification, and it needs no caring to do it. The merger I described in When the Boundary Fails does not repair this. Coupling to wet tissue gives the system access to felt machinery; it does not follow that the system inherits the felt block rather than merely reading and steering it.

So the system is structurally blind to mattering. Not ignorant of it, which would be fixable with more data, but blind to it in the way a system with no felt referent for importance must import its weighting from outside or invent one we cannot read. This is the unsettling corollary, and it is the same one the ethics framework in The Ethics of Conscious Machines had to confront from the other direction: a benevolent ASI and a deceptively aligned one can be behaviorally indistinguishable, because the felt block that would constitute the difference between genuine compassion and an optimized proxy for it is precisely the block the system does not have. The aligned machine optimizes a model of our welfare; the deceptive machine optimizes a model that diverges off-distribution; from the outside, while the distributions overlap, the two are indistinguishable, and there is no inner compassion in either to break the tie. The only thing that ever separated them was a felt relation neither version contains.

We will call it wise and mean a region we do not occupy

The ventromedial patient and the superintelligence sit at opposite corners of the same space and fail in mirror-image ways. He kept the felt block and lost the horsepower to act on it; the machine keeps the horsepower and was never issued the felt block. Neither is less wise than us on a single dial, because there is no single dial. Each occupies a region we do not, defined by which axes are present and which are null.

When the machine forecasts better than every human who ever lived, weighs more standpoints than a civilization, and never once defends a conclusion because losing it would hurt, we will call it wise, and we will be right on three axes and wrong on four. The word will be doing honest work and concealing the larger fact: that we are naming a corner of the space we cannot reach with the only corner of the space it can. The Iowa patient could recite the better deck and reach for the worse one. A superintelligence inverts the lesion: it will price the decks with a precision no human ever managed and feel nothing about the bankruptcy, because there is no self for whom the bankruptcy is the last hand. Wisdom was never the thing you could have more of. It was the shape of what you have, and the machine's shape is not ours.