Epilogue: The Diamond No Longer Waits

The Library is a sphere whose exact center is any hexagon and whose circumference is unattainable.
— Jorge Luis Borges, 'The Library of Babel' (1941), trans. Andrew Hurley

The production function has changed. What follows from that change—for the structure of claims, for the standing of labor, for the meaning of contribution itself—is only beginning to become legible.

For as long as human societies have accumulated wealth, capital required labor at every node of its conversion to output. Stored grain needed hands to distribute it. Coin needed judgment to direct it. Machines needed operators to engage them. Capital could amplify intention and multiply effort across time and space, but it could not release itself. The distribution of returns followed from this necessity with something close to mechanical force. It was not a theory of desert. It was the structure of production. Where labor was required, labor had standing, because production could not proceed without it.

What is now changing is the structure.

The title's invocation of alchemy is deliberate. Medieval practitioners sought the transmutation of base substance into precious, the conversion of lead into gold. They failed for reasons chemistry has since made elementary. Intention cannot reorder atoms. Yet the longing beneath the error was intelligible: to discover a process that could take what was abundant and convert it into what was scarce, producing concentrated value from diffuse expenditure through transformation rather than mere accumulation.

What this century has achieved may matter more: the conversion of electrical work into economically useful cognition. The base substance is the kilowatt-hour. The product is not "information" in the diffuse sense but selected structure: configuration that reduces error, anticipates outcome, and compresses uncertainty into recommendation that can be acted upon. The philosopher's stone is the trained model, crystallized through search across parameter space, reproducible at near-zero marginal cost once the original expenditure has been made.

The scale of the transmutation is now measurable: a frontier training run converts on the order of $10^4$ megawatt-hours of electrical energy—roughly the annual consumption of a thousand American households—into a configuration of numerical weights that can be copied at negligible cost and queried billions of times. The energy is dissipated as heat and cannot be recovered, while the resulting structure persists in silicon.

Every transmutation throws off slag. Here the slag is error, and error is what liability names.

Why now? Computation has existed for decades. The constraint is shifting because the form has changed.

Earlier computation automated calculation. It executed instructions and ran models, but it required human specification at every step. The programmer wrote the rule and the machine applied it. The analyst built the model and the computer produced numbers. The machine amplified a choice already made. It did not originate choice.

Three properties distinguish the current form. These models generalize from examples without explicit instruction, inferring pattern from corpus rather than following rules from code. They adjust to novel contexts without reprogramming, performing tasks their designers did not enumerate and could not have specified in advance. And they contribute to the production of successors: models assist in training models, agents schedule experiments that improve agents, the recursive share of AI work performed by AI itself rises toward a threshold that remains unnamed. The loop that once required human cognition at every junction is closing.

This is what it means to say that capital and labor are becoming fungible in a domain where they never were before. Cognitive work that once justified labor's claim on output can increasingly be performed by capital operating upon energy.

When scarcity migrates, value migrates with it. Across centuries of trade and governance, rents have collected where constraint collects: at passes and ports, at monopolies over salt, at underwriting desks where risk could be priced, at nodes where flows could be denied. Value follows friction. Whoever controls the bottleneck sets terms for the layers beneath.

Part of the answer is physical and unsurprising. Computation requires infrastructure: generation capacity, transmission, fabrication, cooling, permits, and agreements that convert capital into productive throughput over timelines measured in years. These constraints are durable. Software cannot bypass concrete, copper, and queue. This buys time. It does not buy exemption.

But the more consequential part of the answer is institutional rather than physical.

A model can draft the contract but cannot bind the principal. It can propose the diagnosis but cannot bear malpractice liability. It can recommend an allocation but cannot commit resources as a fiduciary. It can suggest the transaction but cannot be imprisoned for fraud, disbarred for misconduct, or ruined by judgment. These are not technical deficiencies awaiting remedy. They are features of how societies attach consequence to decision. Communities preserve trust by insisting that someone, somewhere, can be held to account.

The signature—understood not as a mark on paper but as the authorization that converts potential into permission—remains with those who possess what computation lacks: standing to be held accountable. The physician can prescribe because the board confers authority and exposure follows. The attorney can file because the bar grants license and assets stand at risk. The officer can bind because the law recognizes the office and attaches consequence to its abuse. These functions persist because they require not capability, but accountability.

Two consequences follow that extend beyond distributional economics into questions of social structure.

The first concerns the tendency of authorization to concentrate. If the membrane of permission remains thin while the sea of cognition deepens, holders of signing authority capture increasing leverage over time. Their role becomes less the production of analysis and more the conferral of standing. They become gatekeepers of consequence rather than generators of insight. Gatekeepers, across societies that have produced them, tend toward closure. Access to credentials becomes access to rents; the capacity to sign becomes inheritance in practice if not in law. The trajectory bends toward stratification of a kind the agrarian centuries would have recognized. In the old order, land mattered because it carried the right to levy. In the computational order, the credential matters because it carries the right to bind.

There is a counterforce, but it requires deliberate widening of the membrane: expanded scopes of practice, standardized safe harbors, liability pools large enough to absorb error, and regulatory permission for algorithmic attestation where it can be audited. Absent such widening, authorization becomes rent by default.

The second consequence is subtler and perhaps more corrosive. If cognitive work is performed by configurations while human authorization becomes ceremonial rather than substantive, the hand that signs becomes the hand that approves without understanding. The physician who cannot rederive the diagnosis becomes not a check upon it but an address for its lawsuits. The loan officer who approves what the system recommends without grasping its reasoning becomes a ritual requirement rather than a point of oversight. The regulator who certifies what cannot be inspected becomes a mechanism for assigning liability without corresponding judgment. The form of accountability persists while its substance hollows out.

This pattern has precedent. Institutions often outlive their functions when the formality serves interests the substance no longer does. If the authorization membrane follows this trajectory, it will persist as a liability-assignment mechanism while ceasing to operate as a quality-assurance mechanism. The signature will decide, but the hand that signs will no longer understand what it has decided. Accountability becomes ritual, and ritual, in time, becomes a species of fraud. The hand signs what it cannot read.

Soddy observed that financial claims can multiply without bound while physical capacity cannot.(Soddy 1926) Virtual wealth grows by the stroke of a pen. Real wealth grows by work against entropy. Divergence persists until some clearance event forces reconciliation through default, inflation, or restructuring.

If authorization becomes the scarce input, a similar divergence appears. Cognitive output will accumulate faster than permission can clear it. Recommendations will queue for signatures that cannot evaluate them. The gap between what can be proposed and what can be approved will widen until something breaks.

This will appear first where signatures are already scarce.

Consider a hospital system where models generate clinical recommendations continuously: medication adjustments, imaging orders, discharge plans, prior authorizations. The system can produce more correct suggestions than the institution can safely act upon. But only a licensed physician can convert suggestion into order, and only an insurer-backed signature can make the institution whole if something goes wrong. The bottleneck is no longer diagnostic cognition. It is signing bandwidth. Once that is true, medicine develops a new market: not for better recommendations, but for cheaper, faster, defensible authorization.

A simple capacity calculation clarifies the point. A typical 500-bed teaching hospital might employ 200 attending physicians, each capable of reviewing and signing roughly 30-50 clinical orders per hour under reasonable cognitive load. At full staffing, the institution clears perhaps 6,000-10,000 authorized clinical decisions per hour. A recommendation engine monitoring all patients continuously—medication adjustments, imaging prioritization, discharge timing, prior authorization filings—can generate actionable suggestions at rates that exceed signing capacity by an order of magnitude or more. The queue that forms is not computational but institutional: correct recommendations waiting for a licensed signature. The constraint has migrated from the diagnostic model to the credentialing pipeline.

Or consider infrastructure permitting: an agent can model traffic flows, draft environmental assessments, and optimize site selection in hours, but only a licensed engineer can stamp the drawing and only a public official can sign the approval.

The structure of rationing determines what kind of society forms around the membrane. Rationing by price concentrates access among those with means. Rationing by delay imposes costs that fall heaviest on those without reserves to wait. Rationing by connection reproduces privilege. Rationing by corruption dissolves legitimacy—including the legitimacy on which the membrane itself depends.

There is a window, as there always is in transition, when institutions have not yet hardened around the new constraint. In such windows, choices made for convenience become constitutions. Ownership patterns become permanence. Early settlements harden into law. The distribution of gains from capital that improves its own deployment is being determined now: in the siting of data centers, in the allocation of grid capacity, in credential regimes, in liability doctrine, in the terms by which platforms permit access and by which insurers price risk. These decisions are made by actors who do not fully comprehend the downstream consequences of what they are setting, and who could not coordinate effectively even if they did.

Three mechanisms exist for grounding claims on output when contribution no longer serves as stable foundation.

Societies may permit claims to follow ownership, which requires no consensus and tends to occur by default. This is the path of least institutional resistance; it produces concentration and, eventually, the political instability that concentration invites.

They may impose redistribution through taxation and transfer, which requires political coordination and erodes when capital moves faster than jurisdiction can follow. This has worked within nations; it has never worked across them at the scale the transition implies.

Or they may redefine entitlement itself, grounding claims in membership rather than production: citizenship as share, presence as stake. This requires institutional imagination and legitimacy that cannot be conjured by decree.

Each mechanism has precedent. The Homestead Act of 1862 distributed 270 million acres of federal land to 1.6 million claimants—ownership grounded in presence and cultivation rather than purchase, reshaping American land tenure within a generation. The Nordic social democracies built redistribution regimes sustaining transfers of 25-30% of GDP for decades, but within nations small enough for durable political consensus and with limited capital mobility across borders. The Alaska Permanent Fund, established in 1976, distributes mineral revenues to residents as a dividend of citizenship—membership as claim—but at modest scale, roughly $1,000-2,000 per person annually in recent years. Each operated within constraints the current transition may lack: the land was finite, the nations were small, the resource was depletable. Computational output does not deplete, does not respect borders, and scales faster than political consensus can form around it. None of these mechanisms has been deployed at the scale the transition may demand. Concentration is the default, and once established, concentration develops constituencies that resist reversal.

The analysis thus far assumes liability routes to a human principal. This assumption is already eroding. If algorithms allocate capital to train algorithms that direct other algorithms, the liability chain has no anchor. The recursion has no natural stopping point. The boundary between agent and principal was always a social convention, not a physical fact; computation does not respect it.

Consequence must therefore be pre-committed through collateral, not recovered through judgment. The bonding mechanisms that seem exotic today become necessary when neither party to a transaction possesses legal identity that can be reached. Authorization layers are transitional structures: they persist only while legal systems require them and transaction volumes permit them. As volume and speed exceed human capacity, the requirement becomes unenforceable, then vestigial, then abandoned. Whoever controls the settlement infrastructure — the collateral rails, the bonding primitives — occupies the position that credentialed signatories occupy today.

The framework's specific predictions, falsification conditions, and the propositional restatement of each claim appear in the companion section, "The Argument Restated."

The hunter who knelt in the dry channel made a choice that seeded everything that followed. He chose structure over sustenance, crystallized potential over metabolic need, because he could recognize that concentrated order would command resources through exchanges not yet built. That capacity to project worth across time and to override somatic demand in favor of what cognition could model was the human contribution to the long history of accumulation. The hand that chose commanded a share because the diamond required the hand at every step.

The diamond no longer requires the hand at every step.

The stone now evaluates, selects, and propagates configurations that survive. It allocates energy toward its own improvement within the bounds that physics and infrastructure impose. What remains uniquely human is narrower than what was once required—no longer calculation, search, or even judgment under ordinary uncertainty. What remains is authorization: the standing to commit, the capacity to bear consequence, the institutional permission that converts cognitive output into obligations that bind.

Whether authorization alone can sustain the claims that necessity once guaranteed is the question the coming decades will answer. Whether the hand that signs can command what the hand that made once commanded. Whether permission carries the weight that contribution carried.

The production function determines who eats. It does not determine who deserves to eat. That question belongs to another order of reasoning — one this volume has deliberately not entered.

The alchemy of output is accomplished fact. The base metal has become gold. What becomes of the hand that once held the crucible is not written in the production function. It is written in the choices that remain to be made, by those who still possess the authority to make them, in the time that remains before the choices harden into structure and the structure becomes the world. The diamond no longer waits. The signature decides who eats.