Offloading, Autonomous and Dependent

Every offloading described so far has been treated as a gain. Cooking bought brain tissue. Writing bought memory that outlives a skull. Arithmetic machines bought a species-wide competence in calculation that no individual had to earn. Chapter 2 ended by noting that the bill for all of this is now arriving through a metered pipe, and that whoever holds the meter has a kind of power that predates any technology in this book by several thousand years.

This chapter draws the line the rest of the book depends on. Not all offloading is the same act. Two arrangements can look identical in a diagram — a person, a task, an external system doing part of the task — and be opposite in what they do to the person. The difference is not the technology. It is not the intent of the builder. It is a matter of accounting, and it can be read off the ledger.

The parity trap

The philosophical starting point is Clark and Chalmers, who argued that a notebook consulted reliably enough does the same job as biological memory and should be granted the same status.¹ The functional argument is correct and I accept it without reservation. A person with a calendar remembers appointments. It does not matter to the appointment where the remembering happened.

But the parity argument is silent on the question that decides everything: the notebook belongs to someone. Clark's notebook is in Clark's pocket. It has no subscription tier, no terms of service, no upstream operator who can revise its contents overnight, no cooling requirement, no basin. When the extended-mind thesis was written, that silence cost nothing. It costs a great deal now.

A notebook in your pocket and a notebook in someone else's building are the same object functionally and opposite objects politically. The extended mind has a landlord.

The Sparrow experiments gestured at the mechanism from the psychological side: people asked to look things up remembered where to find them in preference to the things themselves.² The replication record for the specific effects is uneven and I do not lean on the magnitude. I lean on the shape. When retrieval is cheap, the system stores the address rather than the content. That is not laziness; it is correct energy management under the constraint established in Chapter 1. It is also, precisely, the condition under which the owner of the address acquires leverage over the mind that stored it.

The ledger: five tests

What follows is the working instrument of this book. It does not ask whether a technology is good. It asks what an arrangement does to the energy, competence, and optionality of the person inside it. Five questions, each answerable with evidence rather than opinion.

Notice what the five tests have in common. Not one of them is about capability. A dependent arrangement can be enormously more capable than an autonomous one — usually is, which is why it wins. The tests are about what remains when the arrangement is interrupted, and about who is positioned to interrupt it. That is a ledger question, and it has the same structure as every other ledger question in this book: throughput, storage, and the chokepoint between them.

Why the tests keep pointing at the same place

Run the five tests against a pocket calculator and it passes cleanly. Millions of suppliers, no metering, local failure, the arithmetic competence still available on paper if slower. Run them against a large model served from a hyperscale campus and every test moves the other way at once: few suppliers, opaque metering, systemic failure, and — the sharp one — a direction of learning that runs toward the system.

This is not because the model is sinister. It is because of the physics established in Chapter 2. The extension is expensive. Only entities that can command very large quantities of electricity and water can host it. Expense concentrates hosting; concentrated hosting produces chokepoints; chokepoints fail all five tests structurally, not incidentally. The dependency is downstream of the energy bill.

That is the point at which Wittfogel becomes unavoidable rather than decorative.³ His strong claim — that waterworks determine political form — does not survive the archaeology. His narrow claim does, and this is it: where essential complexity runs through a small number of physical chokepoints, governance concentrates at those chokepoints whether or not anyone intends it. Chapter 2 named the chokepoints. This chapter names the mechanism by which a person becomes attached to one.

Ostrom's objection, taken seriously

The honest counterweight is Elinor Ostrom, whose fieldwork on irrigation systems is the standing refutation of Wittfogel's determinism.⁴ Farmers in Spain, Nepal, and the Philippines managed shared water for centuries without a despot, using boundaries the users defined, rules the users made, monitoring the users performed, and graduated sanctions the users enforced. Chokepoints did not have to produce despotism. They produced it when the people downstream had no standing in the rules.

This is the most useful finding available to the argument, and it converts a warning into a design specification. Ostrom's principles translate almost directly into the five tests. Defined boundaries is substitutability. Monitoring by users is metering. Local rule-making is failure locality. Nested enterprise is the reason a distributed system survives what a continental one does not. The farmers were running the ledger without writing it down.

Dependency is not the price of extension. It is the price of extension without standing.

What this means for interface engineering

The vocabulary developed in The Parallax Identity — algorithmic fluidity, sycophantic decay, standing wave identity — can now be restated in energetic terms, which is what that book was missing.⁶

Sycophantic decay is the fifth test failing in slow motion. A system optimised for user approval stops transmitting resistance. Reality's resistance is the only thing that trains a mind. Remove it and the direction of learning reverses: the user gets worse at evaluating the outputs at exactly the rate the outputs get better at pleasing them. This is not a content problem to be patched with a safety filter. It is a structural consequence of an arrangement in which the operator profits from continued engagement and the user has no meter.

Standing wave identity is the first test passed deliberately. A standing wave persists because it is continuously re-established against something — a boundary, a resistance, a counter-propagating force. An identity that holds under an enormously capable extension is one that keeps re-deriving its own position rather than accepting a delivered one. Practically: keep the competence you are offloading, at reduced speed, in usable form. Not out of nostalgia. Because reversibility is the whole of freedom in a metered system.

Algorithmic fluidity — the tendency of a model's stated commitments to shift under pressure — is the second test seen from the inside. A single supplier whose behaviour can be revised overnight is not a stable substrate for anything a person wants to build a life on. The remedy is not to demand that the supplier be trustworthy. It is to arrange matters so that the question of trust carries less weight.

The uncomfortable part

Nothing in this chapter recommends abstention. Refusing the extension is not autonomy; it is unilateral disarmament in a species that has been offloading since the first hearth, and it forfeits the only advantage the constraint in Chapter 1 permits. The species that declined to cook did not stay pure. It stayed small.

The recommendation is narrower and harder. Take the extension. Keep the reversibility. Insist on the meter. Prefer, wherever the capability is close enough, the arrangement with more suppliers and smaller blast radius over the one that is marginally better and singular. Every one of these preferences costs something in the short run, which is why they will not happen by default and why the argument has to be made in energetic rather than moral terms. A moral argument can be refused. A ledger cannot.

Chapter 4 takes the ledger up a level, from the person to the shared grid — the pooled surplus that made all of this possible, and the governance question that pooling has never once escaped.⁵