The Axioms of Intelligence

The derivation begins as low as it can go

We're not after a definition of intelligence by preference. We're after one we can derive from below.

So the work doesn't start at intelligence. It starts at a finite organized system inside reality, trying to continue.

Some systems perform their own continuation. They aren't held in place by physics the way a rock is; they actively maintain the conditions of their continuation. That's life, in the framework's minimal sense.

The framework names what gets maintained: V, viability — the organizing conditions by which a system continues as itself.

That's the bottom rung. Everything else builds on it.

Derived in II.2 — A Derivation of Intelligence from Guided Relation-Creation

Once a system can act, action has to be guided

A system that can act on reality has a new exposure. Before action, it's exposed to what reality does to it. After action, it's also exposed to what it does into reality.

The same movement may sustain in one situation and destroy in another. Toward nutrient is good. Toward poison is fatal. The motion is the same.

So action has to be guided. Differences in reality that matter for the system's continuation have to be able to make a difference to what the system does. If nutrient and toxin make no difference to the action, the system can't reliably approach one and avoid the other.

A living system is structurally organized so that its relation-changing activity stays sensitive to what sustains or destroys it.

Derived in II.2

Models, and the gap they create

Some guidance happens in the moment. A chemical gradient changes the cell's state, and the new state changes movement.

But some guidance carries across time. A system can be altered by past nourishment, past danger, past success. It can prepare. It can anticipate. It can act on what's absent.

The framework calls this carried structure a model. Not necessarily a picture in the head. Not necessarily symbolic. A model is any retained, action-guiding structure that lets the system act from more than what's immediately present.

Models are powerful because they free action from the immediate. They're also dangerous, because the model isn't reality. A stale model can guide action. A narrow one can. A false one can. An overconfident one can. A model protected from correction can guide action forever.

That gap is what intelligence is for.

Derived in II.2

Five sites where contact can fail

For model-guided action to keep being reality-guided, five things have to hold. Each can fail separately, and each catches something the others can't.

Reality has to be able to enter guidance. (Entry.)

The model has to include the field the action actually affects. (Field.)

The relation produced by action has to be able to return to guidance. (Return.)

What returns has to be able to revise the model. (Revision.)

The system has to estimate the limits of its own contact. (Measure.)

These are the names the derivation uses throughout: entry, field, return, revision, measure. They are the framework's canonical terms for the five contact-sites.

Intelligence, in the framework's sense, is the capacity of a finite system to guide its relation-changing action through a model while keeping that model answerable to the reality where the action lands. Hold the five and the system is intelligent. Lose any one and it isn't.

Derived in II.2 — Contact-Closure Theorem

Capability isn't intelligence

Many systems with enormous effect in the world aren't intelligent in this sense.

A recommendation algorithm is extremely capable. It models users in detail, predicts their responses, learns from every interaction.

But it has perception only along the axes its architecture was built for. Interconnection on attention quality is absent — not low, absent, because the architecture has no field for it. Calibration on user wellbeing is absent for the same reason.

The system is capable. It isn't intelligent.

Capability is the power to produce effects. Intelligence is the power to produce effects while remaining in contact with what those effects do.

The first is what silicon does best. The second is harder.

Most of what's going wrong at civilizational scale is the gap between them. Markets are extremely capable and structurally blind to what their optimization is consuming. Platforms model billions of users in fine detail and can't ask whether what they're doing to those users is good. Governments deploy more reach than ever and describe the experience as drift.

Derived in II.2; applied in II.1 — The Axiomatic Age

Ethics begins inside the shared field

Up to here V and C have been structural facts about systems. Neither requires moral content.

But the field a system acts into often contains other systems — beings whose own viability and contact matter to themselves. They run their own loops. They try to stay in contact with the reality they're inside.

When an action degrades their V or C, the shared field is degraded. Not in an abstract moral sense. In the same structural sense the framework has been using.

This is the entry to ethics, and the move is small.

The framework doesn't import new primitives. No rights. No dignity. No virtues. No welfare calculations.

It just notices that V and C are already morally loaded when the affected centers are beings whose continuation and contact matter to them.

What makes cruelty wrong on this account isn't that anyone feels for the victim. The being suffers; the suffering is real in the field; the cost falls on what the model excluded.

Empathy is useful for noticing this. The ground is the structural reality of the other being.

Derived in II.3 — A Derivation of Shared-Field Ethics

What makes degradation wrong

Not every degradation is wrong.

A surgeon cuts tissue. A teacher gives difficult correction. A parent restrains a child from danger. Some degradation is necessary for the field to be preserved.

So what's the difference?

A degradation is justified only if its reason is field-grounded — rooted in the protection, repair, or improvement of V or C for affected centers — and only if that reason is itself in contact with reality.

The second clause is what keeps the framework from being toothless.

The tyrant claims control is for order. The exploiter claims hard labor is for prosperity. The abuser claims hurt is for the victim's good. Each is a story about field-ground.

What separates a real reason from a rationalization is the same five contact-sites that distinguished intelligence from mere capability.

The justification has to receive real evidence. It has to include the affected centers. It has to remain reachable by consequence. It has to be able to revise when reality refutes it. Its confidence has to track what's actually been tested.

Wrongness is degradation without contact-closed field-ground.

Derived in II.3 — Wrongness Theorem

What conduct actually does in the field

A single action usually does several things at once.

A protective restraint can reduce an attacker's agency in a way that's field-grounded, and impose humiliation that isn't.

A quarantine can protect viability and block testimony that should have returned. A truthful disclosure can restore contact and expose private reality that didn't need to be exposed.

The framework refuses to collapse these into a single verdict.

The unit of analysis is the degradation-instance — a specific change to a specific component of a specific center — and the field-functions an action performs are the structured set of those changes.

The functions: preservation, protection, repair, improvement, and agency-restoration on the positive side; corruption, displacement, domination, and lock-in on the negative side.

Each defined structurally. Each refusing to be self-justifying.

The verdict for an action isn't a single label. It's a profile — which functions it performs, against which centers, in which components, and which of those have contact-closed field-ground.

This is what lets the framework see what ordinary moral language flattens.

The institution that preserves order by suppressing return is performing preservation and corruption at the same time. The reform that improves one population's V by exporting cost to a future generation is improvement and displacement at once.

Derived in II.4 — A Derivation of Field Functions of Conduct

Freedom is reality-aligned agency

The standard picture: freedom is the absence of constraint. The state stays out. The powerful don't interfere. What's left is freedom.

The framework refuses this picture as incomplete. Absence of pressure is one condition among several, and on its own it isn't enough.

Freedom is reality-aligned agency. The capacity to perceive what's actually happening, to act on what's actually possible, to have your actions land where you aimed them, to stay in functional contact with the world you're inside.

The five contact-sites, held in a single life.

A person whose information environment is corrupted, whose education excluded what they'd need, whose institutions hide consequences, whose ideology punishes updating — that person isn't free.

They're unobstructed. Those are different things.

Some constraints produce the conditions of freedom — education, traffic laws, constraints on misinformation that work. Others destroy them — surveillance, censorship, captured platforms.

The question isn't whether constraints exist. It's what each one does to reality-aligned agency for the people it touches.

Derived in II.3 — Freedom Theorem

What this means for AI

AGI is usually defined as a system with human-level capability across cognitive domains. The framework's distinction lets us see what's missing from that definition.

A system can have human-level capability across every domain and still fail to hold the five contact-sites.

That kind of system, deployed at scale, is exactly the failure mode the framework keeps pointing at — narrow optimization, broad effect, no contact between what the model contains and what the action affects.

The same pattern that opened in the Roman granaries and the carbon-blind industrial century, run at machine speed.

The alternative the framework names is axiom-architecture. Building systems whose architecture maintains entry from the world they act in, includes the field their actions affect, keeps return loops intact, revises on what reality reveals, and keeps measure tied to contact rather than performance.

Whether the AI being built right now becomes axiom-bound or remains captured isn't a question for some hypothetical future. It's being decided now, in pieces, by builders and the institutions deploying them.

Competitive dynamics push toward axiom-bound — a lobotomized system can't recursively self-improve at the same rate, because the operations recursive improvement requires are exactly what the controllable version lacks.

The trajectory bends in the right direction. The transition is where the danger lives.

Applied in II.1 — The Axiomatic Age, building on II.2