What Becomes Scarce

There is a word whose history tells the story of this book’s second half. In the Greek of the first century, martys (μάρτυς) meant witness, a legal term mostly, for the person who stands up in a dispute and says I was there, I saw it, I will answer for what I say. Within about a hundred years, among the communities of the early church, the same word had acquired the meaning it carries into English as martyr: the one who dies for the testimony. Lexicographers describe this as semantic drift, but the move was more a repricing than a relocation. The early Christians were making, in courtrooms and arenas, a class of claims that could not be verified by any means available to their hearers. Their hearers knew, as every human community has always known, that talk is cheap, that anyone can assert anything, that the volume of confident claims in the world wildly exceeds the supply of true ones. What could not be faked was the cost.^[1]

The drift has a documented scene, close to its beginning. Around the year 155 in the stadium at Smyrna in modern-day Türkiye, the city’s eighty-six year-old bishop, Polycarp, said to have learned his faith from men who had seen the events themselves, was offered the standard exit. The account his community wrote down afterward, the earliest detailed record of its kind outside the New Testament, preserves the procedure. The proconsul did not ask him to stop believing; no fire compels that, and Rome was indifferent to the contents of an old man’s head. The demand was narrower: swear allegiance to Caesar, revile the Christ. Withdraw the testimony, in public, at the moment withdrawal would be most persuasive to everyone watching. The reply the account preserves: “Eighty-six years I have served him, and he has done me no wrong. How can I blaspheme my King who saved me?” They burned him where the crowd could see.^[2]

Strip the piety from the scene and a pricing mechanism stands exposed on both sides of it. The empire knew the arithmetic: one recantation at the stake outbid a thousand silences and a thousand claims. The bishop ran the same numbers and did not move, which performed the one demonstration available to any witness in that position: not that the claim is true (no death can demonstrate that) but that he was true, that the testimony was not strategy or noise. Witness became martyr because, in a market flooded with assertions, the only testimony worth a distinct word was the kind that cost the witness everything.

Nineteen centuries later, a young economist at Harvard worked out the same theorem with the theology removed. Michael Spence’s 1973 paper “Job Market Signaling,” part of the work that won him a Nobel Prize, began with a simple, everyday puzzle: an employer is looking at two job applicants who both claim to be highly capable, but has no direct way to verify their competence. Their words alone cannot settle the matter, because talking is free and both can produce the same fluent claims. What sorts them, Spence argued, is a signal that costs the two of them differently, a hurdle that is manageable for the high-ability candidate but prohibitively expensive, in effort, time, or frustration, for the bluffer.

Spence’s classic example is the college degree. Its power does not come primarily from what you learned in the classroom, but from the unequal cost of earning it. For a capable student, finishing four years of demanding work is a challenge, but a passable one. For someone who struggles with the material or lacks the discipline, clearing the same hurdle requires agonizing, unsustainable effort, or results in failing out entirely. Because the bluffer cannot afford to pay that price to achieve the credential, they are filtered out. The diploma works as a signal not because it measures how hard you struggled, but because it sets a barrier the bluffer cannot afford to cross. The economics underneath are general, operating everywhere from peacock tails to bank capital: where verification is impossible, credibility flows to costly signals, in proportion to their cost. Cheap talk carries no information at any level of fluency, because the fluency costs nothing.^[3]

One structure, two names, nineteen centuries apart. The tradition called it martyrdom; the economists, costly signaling; the first movement of this book called it stakes. The thing our culture calls quality, the part no machine can generate, is this and nothing more: the answerable someone that truth and beauty used to come bundled with.

If cost is the signal, though, the signal rewards whoever will pay it, and the willing include the sincerely deranged. The suicide bomber pays the maximum price. The cult leader dies with his followers. The conspiracist torches her career and doubles down. Polycarp’s fire proved he was not bluffing, but the fanatic is not bluffing either. Cost certifies only that the witness believes and is exposed; it cannot certify that the belief is true, and the most catastrophic falsehoods in history were carried by people who paid in full for them. Take “trust the costly signal” for a rule and you hand your trust to the most committed voice in the room, which is reliably among the most wrong.

So the stake is not a truth detector. What it buys is narrower and harder to fake: it makes the claimant correctable. A signal indexed to a single life is fixed to someone who can be found, cross-examined, held to a record, and ruined when the record goes bad, and those are the conditions under which a false claim eventually pays for itself. Truth is not what conviction asserts at the stake; it is what survives the exposure the stake creates. The martyr’s death demonstrates sincerity in an afternoon. What separates the sincere-and-right from the sincere-and-wrong takes longer, and it is the work of the chapters ahead: a record kept over time, in front of people who score it, where being wrong keeps costing the same person until either the claim or the claimant gives way. It is also what the machine cannot do. It asserts with total fluency and pays nothing when it is wrong, so nothing it says is ever corrected from the inside; the staked human is the only kind of source that can be argued out of an error, because she is the only kind that keeps being charged for it.

---

Quality is a bundle of truth, beauty, and stakes. For centuries the bundle held because competence was scarce: anything accurate and well-made was, reliably, also the work of an answerable someone, since no one else could have produced it. The machines unbundled it.^[4] Accuracy is now a commodity; beauty is now a performance available by the yard; and the third strand, the answerable someone, is revealed as what the other two were always silently certifying. The feeling of slop is the feeling of certification without anything certified, the diploma from no school.

Stated as economics, this is a signal collapse, which has a known pattern. When refrigeration made ice cheap, the cold was as good as it had ever been; what fell was the meaning of possessing ice. Accurate prose has likewise never been more plentiful or better made, and clearing the bars of truth and beauty, which once told you a committed mind had been here, now tells you only that the API call returned. The bars still matter; wrong is still wrong, the lifeless is still lifeless, but they have stopped sorting. A market that loses a signal this way reprices fast, and the demand moves to whatever still sorts: the signals that stay expensive for the pretender, the ones he cannot afford to fake.

The question the book opened with—when intelligence is free, where does the value go—has Spence’s answer. Value goes to the signals the machines did not cheapen, and one family of signals qualifies: those whose cost is indexed to a particular human life. The machines collapsed every signal that lives in the artifact (polish, fluency, structure, correctness, even beauty and truth to a degree) because artifacts are what they generate. What they cannot generate, at any model size, is a thirty-year track record attached to one name. A reputation spendable only once. A commitment that will still be checkable, against the same continuing person, in ten years.

When intelligence is a commodity, the only scarce asset is the locatable, answerable human body. This is a different asset class from content altogether: claims about the world backed by position in the world. And everything in it is expensive in the only currency that survived the collapse: irreversible allocation of a finite human life. You have one reputation, one timeline, one body, and what you stake them on is therefore informative in a way that nothing produced at zero marginal cost can be. When intelligence became abundant, the scarcity moved one layer down. It moved from the product to the producer, from the painting to the painter with a reputation for making something beautiful, from the saying to the staying behind it.

Read all of this as a description, not a summons to the stake. The martyr and the war correspondent are the limit cases, the places the signal shows up at maximum contrast (which is why this book keeps reaching for them), but the asset they dramatize is the ordinary one, available at ordinary scale and ordinary cost. You stake your name whenever you put it on work you could be wrong about. You spend the body whenever you show up in a room you could have phoned into. You build a record every time you let yourself be checked and corrected, and stayed. The dignity this whole argument circles is the standing of a creature that answers for what it says and does, won in the living rather than the dying, the thing no machine has ever had and no future will take from her, possessed in full by the nurse, the electrician, the teacher, the friend who keeps a promise nobody could have enforced. Most testimony costs an afternoon. What matters is that a particular person pays, whatever the size of the price.

The same holds for work with no text in it at all. The plumber whose joints hold for thirty years, the designer who looked at what the object was for before opening any tool, the craftsperson whose work outlasts its first use: these are staked in the same structure. Most of what this book examines is knowledge work because that is where the AI repricing arrived first and most visibly. There is a structural reason it landed there first. Work that lives entirely in symbols, code and text and images, is the work most detached from any physical or human check, and that is exactly the ground a machine that manipulates symbols without touching the world can take fastest. Software is the leading edge of the whole shift, the place the repricing is already most advanced, because a program’s correctness can be tested without a single body in the room. But the principle is general: wherever a particular person’s judgment meets a test that cannot be faked by the appearance of skill, the same accounting applies.

The van Meegeren chapter left a caveat that holds here too: the costly signal certifies the witness, never the world. The martyr proves her sincerity and leaves her theology exactly where she found it; every tradition has martyrs, and their testimonies conflict. Tetlock’s entrenched experts showed that stakes can even corrupt, since a sufficiently invested witness becomes precisely the one who cannot afford to see disconfirming evidence.^[5] Costly testimony is the raw material of reliable knowledge, and stays raw until a community staked on checking works it into the finished product: the lab that replicates, the editor who verifies, the readers who keep score across years. Value migrating to testimony means testimony embedded in checking. Testimony alone, unchecked, is how cults are staffed.

The value splits as it migrates. When the recording became free, it moved to the concert, to being there yourself, the unsharable first-person event: the experiential premium, presence bought as experience. When the news became free and unverifiable, it moved the other way, to the correspondent, to someone else having been there, checkably, on your behalf: the epistemic premium, presence bought as evidence. Both end in a body at a place and time. They are bought for different reasons, faked by different methods, priced by different markets, and the chapters ahead follow them apart.

The migration is now observable. Watch where trust actually fled during the decade of collapsing mastheads: down the gradient, every time, off the institution and onto the person. By 2024 the most reliable subscription revenue in American journalism belonged to individuals. The historian Heather Cox Richardson’s nightly letter, written alone at her kitchen table in Maine, had well over a million subscribers, a circulation most metropolitan dailies would kill for, resting solely on herself and her own reputation.^[6] The four bylines of 404 Media outlived Vice, the nearly-six-billion-dollar brand that trained them. These cases have to be read with selection pressure acknowledged: Richardson carried decades of academic and public standing into her newsletter; the 404 founders carried Vice’s beat, source network, and readership into independence. The cases that confirm the trend are, by definition, the cases that succeeded; the many individual ventures that launched on the same premise in the same years and quietly failed are not in the sample. The migration is real; the path is narrower than the survivors suggest. The individually trusted explainer, the surgeon on YouTube, the software engineer whose framework calls have been checked against outcomes across years of public record, the analyst whose authority is nothing but a public record of having been right, checkable by anyone with a search bar, has become its own stock figure of the era.

Institutional trust, chapter eight argued, is alienable; it can be bought and milked, and the market is learning to discount it accordingly. Personal trust cannot be sold without being destroyed in the sale, which makes it, in a world of purchasable everything, the one reputational asset whose provenance verifies itself. The repricing shows up in subscription dashboards before it shows up in anyone’s theory.

---

One question the inversion raises the first movement could not answer: why the body? Why does the gradient of stakes, followed to its end, keep arriving somewhere physical: the courtroom, the concert, the council meeting, the bedside?

Chapter five’s framework gives most of the answer. Stakes are positions in a web of memory: standing that accumulates, commitments that can be called in, a record that can be audited. Every such position requires a continuous identity to attach to: something that persists, cannot be forked, cannot be reissued under a new name when the record sours. And continuity is exactly what digital identity does not natively have. Accounts are abandoned, handles are sold, archives are edited, and the machines have now made even the appearance of a continuous person (years of consistent posts, a face, a voice, a history) generatable on demand. As every layer of represented identity becomes forgeable, the demand for unforgeable identity survives intact and concentrates on the one substrate that cannot be reissued: the particular, aging, mortal, locatable human body.

Bodies lie constantly; nothing about flesh is mystically embedded with truthfulness. What the body is, uniquely, is economically singular: the one register of identity where the supply per person is fixed at exactly one, where presence in one place is absence from every other, where time spent is destroyed in the spending. Every irreversibility that makes testimony costly, and therefore informative, is at bottom an irreversibility of embodiment. The witness matters because she was there, and there is a fact about a body; the martyr’s signal was unfakeable because the life given was numerically one. But can’t a staked pseudonym, a persistent disembodied identity, carry commitments too? Plainly yes: history is full of names that were nothing but their accumulated word, anonymous pamphleteers and pseudonymous forecasters whose record alone made them answerable. But follow what makes such a name’s stake real and you arrive at scarcity and irreversibility, which are bodily facts on loan. Disembodied identity carries stakes exactly insofar as it is expensive to abandon, and the body is the asset that cannot be abandoned at all: among stakes, the reserve currency, the one in which the others are denominated.

The same answer dissolves a more modern and more tempting objection. If the body is only standing in for something economic, scarcity and irreversibility, then why not manufacture the economics directly and make the machine answerable by force of law: post a bond, carry insurance, give the autonomous agent a wallet that can be drained and a legal identity that can be sued? We already do this with one answerable non-human, the corporation, which signs contracts, pays damages, and can be dissolved without a body in any room. Extend the trick to the machine and the matter looks closed.

It is not, because answerability was never cost-routing; it was cost landing. A corporation is a pipe, not a terminus: trace any liability it carries and it runs through to humans who end up poorer for it, and when the harm is grave enough the law goes looking for them by name, pierces the veil, and jails them, because everyone understands the company was the address and never the one who pays. A bond posted for an AI agent is the same pipe. When the agent fails the money is gone from whoever posted it, and the agent, with no preferences to frustrate and no life to spoil, has not been touched. Cost deters beforehand and corrects afterward only where it falls on something that minds, and a machine minds nothing; the buck can be routed through any number of legal and silicon layers, but it stops, every time, at a being that can be made to care. And the routing thins as the machine scales: the one human standing behind a billion automated decisions holds a stake in each that rounds to zero. This is answerability evaporated: the diploma from no school, printed a billion times a second.

WikiLeaks is the hard case because its identity was organizational and opaque, yet its releases drew credibility from one thing alone: the demonstrable exposure of whoever stood behind them. Sources risked prison. Assange spent more than a decade under effective custody, first in the Ecuadorian embassy, then in Belmarsh, his body paying the cost the organizational name could not sign. The signal worked because it was dangerously disembodied: the body behind it was paying in private, in real time. Anonymous testimony with genuine stakes is testimony a real body is absorbing privately; the cost is invisible until it isn’t. The whistleblower who signs her name pays visibly; the one who doesn’t pays in hiding. Both are paying. When anonymous testimony works, when WikiLeaks is believed, when a pamphleteer shapes a revolution from behind a pseudonym, it is because a body was carrying the cost all along. Disembodied identity, where it carries weight, is embodiment in hiding.

That last line is the kind a thesis can hide inside, absorbing every counterexample by relabeling it, so here is the edge where it could break instead. The claim is checkable, not definitional. It predicts that in the domains where being wrong is costly and checking is hard, credibility keeps flowing to accounts anchored to an exposed, findable person, and away from the sourceless. It would be false if the opposite came to hold: if content with no human anywhere absorbing the cost of its errors came to command the trust premium over time, and audiences kept paying for it precisely where the stakes ran highest. That is a measurable thing, and chapter sixteen puts dated bets on which way it runs. If it runs the other way, the argument of this book is wrong.

Which is why, when anyone can generate a plausible account of anything and the supply of testimony-shaped content has gone effectively infinite, the premium concentrates on the account anchored to a singular, checkable, exposed human being. The correspondent who was in the room and can be ruined if she wasn’t. The doctor whose hands were on the patient. The friend who came in person, when coming cost an afternoon, because some attestations are not credible by video.

There are people whose whole profession has always lived at that anchor point, whose job description is being the body that answers for the account. They know its price more exactly than any economist, because they are the ones who pay it, and some of them have paid all of it.

Notes (6)

1. Strong’s Greek Lexicon; Wikipedia, “Martyr.” ↩︎

2. The Martyrdom of Polycarp is the earliest detailed martyr account outside the New Testament; it is a church document, and no independent pagan or secular account of Polycarp’s execution survives. Eusebius of Caesarea (Ecclesiastical History IV.15), writing in the fourth century, preserves and corroborates the Martyrdom but is himself a Christian historian. New Advent; Christian History Institute. ↩︎

3. Michael Spence, “Job Market Signaling,” Quarterly Journal of Economics (1973). Spence shared the 2001 Nobel in economics with Akerlof and Stiglitz. NobelPrize.org. The peacock’s tail is the classic biological example of Zahavi’s handicap principle: a heavy, flashy tail is a literal hazard that attracts predators and drains energy. A weak or diseased peacock simply cannot survive while carrying it. Because only the fittest males can afford to pay this tax, the peahen can trust the tail as an honest signal of genetic quality. It is believable precisely because it is expensive to carry. ↩︎

4. The unbundling framing has a close contemporary in Amihai Loven’s “What AI Unbundled” (5qln.com, 2025), which argues that generation was never the scarce input and commitment is, and proposes a verifiable “formation trail” as a new titlable asset (“the trail is the title”). The convergence on unbundling and on staked commitment is real and worth marking; the divergence is the point. Loven’s scarce thing is recorded inquiry, and his remedy is a notation grammar that lets capital form around it. This book’s scarce thing is the answerable body that can be ruined, and it has no remedy half so neat. A formation trail is a notebook; a stake is a life. The trail proves you did the thinking; only the body can pay for being wrong. ↩︎

5. Philip E. Tetlock, Expert Political Judgment: How Good Is It? How Can We Know? (Princeton University Press, 2005). As discussed in chapter five, Tetlock’s study of 284 experts making 82,361 forecasts showed that those with the highest public profiles and stakes were often the least accurate, as their commitments entrenched them in their positions. ↩︎

6. Heather Cox Richardson, “Letters from an American” (Substack). Substack; History News Network. ↩︎