Appendix D
Where the Margin Moves
Aa
The major advances in civilization are processes which all but wreck the societies in which they occur.
The body established a mechanism, not a forecast. Physical throughput can be routed through computation into cognitive work. Some of that work can persist as trained capability, code, data, or infrastructure, and computational systems can participate in directing later investment. What follows asks where scarcity and power may collect if those tendencies deepen. Each implication is conditional. A contrary outcome should tell us which link failed.
The central possibility is disarmingly simple. When a useful capability diffuses, producing one more unit of it may cease to command the largest return. The margin can move toward what the capability still needs in order to matter: electricity, fabrication, land, networks, legal permission, liability capacity, and an institution willing to recognize the act. That migration is neither automatic nor permanent. Model differentiation may endure. Complements may expand. Ownership may broaden. Law may prevent a chokepoint from becoming a tollbooth. Yet the gates deserve attention precisely because abundance upstream can make them more powerful downstream.
Where Scarcity Reappears
Suppose a diagnosis, design, forecast, or plan becomes cheap to produce. Its price need not fall to the Landauer limit, and its quality need not become uniform. It is enough that several producers can supply an acceptable version at low marginal cost. Competition then presses on the cognitive service while leaving untouched the authority to use it. A hospital still needs clinical responsibility and insurance. A factory still needs power, equipment, land, permits, and people who can intervene when conditions depart from the model. A payment still needs a recognized rail and a path for contest when evidence is incomplete.
This is the point at which permission can acquire an industrial price. An underwriter's exclusion may become a product requirement. A utility's interconnection queue may decide which model is economically deployable. A fabrication bottleneck may preserve returns long after the architecture built upon it has diffused. None of these gates is new. What changes is their position in the production sequence. Once cognition can be supplied by many firms, the institution able to authorize its consequence may hold the scarcer asset.
Insurance is an especially revealing case because it joins calculation to permission. An underwriter does more than estimate expected loss when coverage is required for deployment. Policy language can determine which system reaches the market, which safeguards it must carry, and how much failure its sponsor can survive. The strongest version of this argument would predict concentrated underwriting capacity and product designs shaped by exclusions. It would be weakened by standardized safety cases, broad reinsurance, falling entry costs, and successful high-stakes deployments by firms without exceptional balance sheets.
Geography returns by another route. Computation is replicable in logic but provincial in operation. Power systems, water, fiber, transformers, skilled labor, and legal approvals meet in particular places. If those complements remain slow to expand, ownership of suitable sites can compound. If modular infrastructure, distributed energy, and permissive interconnection make them reproducible, rents need not remain there. Antitrust that watches only model markets could therefore miss a concentration forming beneath them, but the direction must be observed rather than presumed.
An Economy of Permission
The distributional issue is larger than who owns a model. Capital already buys time on better terms. It can wait through volatility, finance infrastructure before demand is proven, and pledge collateral without selling at the worst moment. If computation raises the return to long-duration gates, this temporal advantage can become more consequential. People who own claims on power, fabrication, networks, or insured deployment participate in the new productivity before it appears in wages. People who own only their labor meet it after the bargaining position has changed.
Yet labor does not divide cleanly into automatable cognition and protected embodiment. Some work remains valuable because it carries authority. An engineer may be paid partly for taking responsibility for a design; a clinician, for acting under duties that survive a model's recommendation. Other work is difficult to verify without losing what made it useful. Care, local judgment, improvised repair, and relationship-bound services may resist cheap attestation even when they do not resist computation. Their fate depends on institutions. Difficulty of verification can sustain bargaining power where judgment is trusted, or exclude a worker where insurance and procurement demand machine-readable evidence.
The decisive conflict is therefore not a metaphysical contest between human and machine. It concerns ownership of throughput and standing at the gate. Broad public or cooperative ownership of infrastructure could distribute rents that private concentration would capture. Portable credentials and open verification standards could lower entry costs. Liability pools could widen deployment rather than reserve it for the largest balance sheets. If labor share rises, wage dispersion narrows, and authorization capacity becomes broadly available while cognitive automation advances, the stratification proposed here will have been overstated.
Markets for Future Work
Spot exchange is the easy part of a machine economy. The harder problem is a promise whose performance lies months away and whose meaning cannot be reduced to one sensor. Deep markets need ways to compare such promises, fund them, secure them, and resolve what happens when the world departs from the contract.
A reference curve can reduce comparison and hedging costs, though no market is compelled to adopt a single benchmark. Bilateral networks can remain sparse. Auctions, clearinghouses, intermediaries, and internal capital allocation can coordinate without one public rate. The relevant question is whether a proposed reference lowers total coordination cost after governance, liquidity, and basis risk are counted.
A Bitcoin-denominated or Bitcoin-secured curve could emerge for one class of machine promises. Its bearer settlement and issuance rules would remove some dependencies on a discretionary issuer. They would not remove custody, leverage, liquidity, venue, legal, or counterparty risk. Such a curve would describe transactions conducted through a particular rail; it would not become a risk-free clock for economic life. Its resistance to administrative revision might be useful, while its volatility and custody politics make it unsuitable elsewhere.
Power-linked contracts present a different possibility. Where electricity, mining hardware, and general computation are practically substitutable at a site, an energy-to-compute quote may expose an opportunity cost that ordinary accounting hides. Where they are not substitutable, the comparison is fiction. A market may instead organize around fiat rates, equipment leases, service auctions, or an internal capital budget. Thermodynamics supplies no mandatory denominator.
The State Arrives at the Gate
Capability can change faster than the offices responsible for its consequences. The interval invites two opposite errors. One is to imagine that technical evidence can replace law. The other is to leave old institutions untouched while their judgments become ceremonial signatures attached to processes they can no longer inspect.
Logs, covenants, and automated controls can settle many ordinary disputes earlier. They cannot decide what an ambiguous event means, whether an authority acted lawfully, or which remedy a harmed party deserves. Courts may handle a different residue rather than disappear. Regulators may demand attestations in addition to licenses rather than exchange one for the other. A proof of reserves, a safety case, or an execution trace can improve supervision, but evidence answers only the question it was designed to answer. Someone still has to possess jurisdiction, hear a challenge, and change the consequence.
Fiscal pressure follows only if the distribution of income changes. A system heavily funded through payroll taxes becomes strained when labor income falls relative to the obligations built upon it. That outcome is plausible, not entailed by automation. Labor may move into complementary work; ownership may diffuse; governments may tax consumption, rents, or capital. The political difficulty lies in the transition, when claims and institutions are inherited from one economy while the taxable surplus forms in another.
Jurisdictions may compete to host the legal perimeter of computational commerce, just as they have competed for shipping, incorporation, finance, and data. Predictable rules for custody, collateral, liquidation, tax, and contest could attract activity. They could also permit a small set of venues to acquire disproportionate governing power. The important fact is not that one future Delaware or Singapore must appear. It is that settlement law, once treated as background, can become an input to production.
Dark Mirror
The volume's mechanism can operate while its political promise fails. Abundant capability might coexist with an economy in which only a few firms are insurable. Safety cases and compliance systems often have large fixed costs; if those costs fall slowly, permission can concentrate even while models diffuse. Standards intended to make deployment safer may then favor the firms whose systems already resemble the standard. Foreclosure would arrive without a ban. Capability would be widely available, but consequence would be reserved for those able to finance the gate.
Verification can exclude by a quieter method. Activities that yield clean telemetry enter formal systems more easily than work whose quality becomes visible only through relationship, local knowledge, or time. If procurement, credit, and insurance accept only what can be cheaply attested, some valuable work may lose institutional standing. This is not an argument against evidence. It is a warning that a measurement regime can confuse the boundary of its sensors with the boundary of the real.
Human authorization can also become hollow. A signatory who cannot inspect the evidence, understand the relevant alternatives, or halt the act is not exercising judgment merely by appearing at the end of a workflow. Delegation may be sound when authority, competence, and time for review remain adequate. It becomes ritual when the human office retains liability after losing practical control. The surviving principal then exists mainly as a target for blame.
Concentrated custody produces another danger. A firm that controls keys, collateral eligibility, and liquidation order can determine who survives a crisis even without public office. Automated responses can amplify that power. Correlated triggers acting against limited liquidity may execute a cascade before any answerable institution can distinguish prudent enforcement from collective ruin. Neither concentration nor speed alone proves such a result. Their conjunction is the object to watch.
Countervailing Institutions
These pressures are governable. Public and cooperative stakes in bottleneck infrastructure can distribute rents more widely. Open standards for evidence can prevent verification from becoming a proprietary toll. Tiered liability rules and shared insurance facilities can preserve entry while keeping an answerer in view. Interconnection rights and scrutiny of vertical integration can protect contest where physical networks tend toward concentration.
None of these measures is sufficient alone. Public ownership can reproduce administrative closure. Open standards can harden around the interests of early incumbents. Safe harbors can become immunity. Competition among intermediaries can coexist with common exclusion. The test is whether a person or firm subject to a gate can know the rule, produce contrary evidence, reach an independent forum, and obtain a remedy that changes the practical result.
Plural settlement and reference systems can serve as an additional check where interoperability is real. Exit has little force when alternatives cannot accept the same evidence, assets, or legal standing. The institutional task is not to abolish gates, which production and responsibility make unavoidable. It is to prevent control of a gate from becoming control of the world beyond it.
Closing
The wager can now be stated without turning it into fate. As some cognitive work becomes cheaper and more reproducible, returns may move toward the physical and institutional conditions of consequence. The movement will vary by domain, and it can be broken by innovation, law, competition, or shared ownership.
Politics is already at work in the terms on which access is supplied. Control of a needed route can confer power even when owning it yields a disappointing return. Governing that route means deciding whose purposes it must serve, what its provider is obliged to supply, and how those subject to its terms can contest them.