Conclusion: The Cost of Testimony

Intelligence got cheap. Not wisdom. Intelligence. Answers, fluency, competence, the entire layer of human output that can be evaluated by looking at the output. That layer is now a commodity, and every value that lived in it is draining toward the one thing the machines do not cheapen, because it was never in the output at all: the answerable human. The person who stands where the cost lands. The witness. For twelve chapters I traced what disappears when that person is removed from the making of things; for six more, where the value reappears when the person is restored. The two halves are one claim: slop is what content becomes when the testimony is subtracted from it, and the presence economy is what testimony costs once it is priced on its own.

The word this whole argument keeps arriving at is testimony, the same root the courtroom uses: the act of a person placing themselves behind a claim, in public, where being wrong has an address. The first Christians’ word for it became the word for dying because their hearers understood pricing better than our feeds do: an attestation is worth what it costs the one who gives it. Spence proved the theorem for job markets; Perkins enforced it on manuscripts; the Confrontation Clause carved it into constitutional law; Colvin paid it in full in a shelled media center in Homs. It is the homeliest principle. It is how grandmothers knew which neighbors to believe. We built an information system that forgot it, flooded that system with fluent unstaked content until the forgetting became unlivable, and are now, expensively, in courtrooms and camera firmware and iris-scanning orbs and subscription dashboards, being forced to remember.

The instinct is not eccentric and not only mine. In early 2025, the Vatican issued a formal note on artificial intelligence and human intelligence,^[1] and then, in the spring of 2026, a full encyclical on it;^[2] the word both documents organize themselves around is dignity. The timing was deliberate to the decimal: the encyclical was signed on the hundred-and-thirty-fifth anniversary of Rerum Novarum, the 1891 letter^[3] in which an earlier pope of the same name answered the first industrial revolution by relocating human worth away from what the worker’s labor produced, which the machine had just cheapened, and into the worker himself, who stayed beyond cheapening. Rome grounds dignity in gift: worth conferred on the person ahead of any capacity or output, owed as fully to the one who can answer for nothing as to the one who can answer for everything. This book means something narrower and more contingent: the standing you build by being answerable, which can be grown and can be lost. The two are not rivals. They meet at the floor both were raised to defend, that a person is worth more than what they produce. In 1891 the Church said a human being is not an engine; in 2026, it said he is not an information processor. Each time a machine had cheapened the output, and each time the answer set human worth in the person the machine could not reach. Where the output stops being scarce, the part this book has tracked, your standing as the one who can answer for your actions, is what is left to price, and no ledger prices it and no machine supplies it.

And it reaches far past a writer’s problem or a knowledge worker’s. Every promise kept, every account given, every word one person offers another and stands behind, is testimony by the same test, or it is slop. We are all in this trade every day.

In 2024 Dario Amodei, who runs Anthropic, published an essay imagining a future where machines cure most diseases and fold fifty years of science into five, then spent much of the same essay uneasy about the other face of it: an intelligence that powerful, pooled in a few private hands, with no settled answer for whom it should finally belong to.^[4]

The reflex is to reach for a familiar fight, the state against the company, public control against private. But this is not the right fight. A nationalized frontier and an unaccountable lab are the same danger in two uniforms, because whatever made a holder of dangerous power tolerable never had much to do with whether it wore a flag or a logo. It came down to whether somebody on the outside could drag it into the light and make it answer. We can live with a state that regulates the labs, because that state can still be sued, voted out, and reported on; made to answer by something outside itself. Hand it the labs too and there is no outside left. The book’s oldest question, put to a forger and a masthead and a midnight chatbot alike, comes back with no one to ask: who answers for the one power that answers to nothing? That is a different book, and a grimmer one.^[5] But it runs on the test already given. A holder of dangerous power stays tolerable exactly as long as someone outside it can drag it into the light and make it pay, and stops the moment no one can.

That outside is not a given. It has to be built, and the building has started. Coalitions that cross the usual political lines have begun setting down concrete principles for what accountable AI would require, from pre-deployment safety testing, to liability that cannot be shielded, to a brake on the race toward systems no one can yet control. The demands draw broad popular support even where the particulars stay arguable.^[6] None of it is settled, and some of it may be wrong. But the answer to who makes the power answer is being drafted now, by more hands and across a wider coalition.

The turn, though, cuts toward hope. The intelligence those labs are guarding and governments are circling, the thing serious people argue may be too dangerous to leave in private hands, opens in a browser tab for anyone who wants it, for the price of a sandwich or for nothing at all. What the most powerful states on earth are maneuvering to control, a person in an apartment can just use. And pointed well, the same machine that pours slop into the feed does the opposite: it carries an ordinary person clean past the ceiling of their own mind. The book you never had the decade to research. The language you could not read. The second field you were never going to learn. The sharper version of your own argument that a cleverer opponent would have built and you could not. All of it sits an afternoon away, for anyone who keeps a hand on the rudder.

The catch is the one the genie always carried. A machine that makes whatever you ask is worth exactly what you ask of it, and most wanting, once wanting costs nothing, turns out to be slop. The feed is a portrait of what its makers wanted the second wanting got cheap. Your own work is about to be a portrait of what you want. Aim the thing at a half-formed wish and it hands back a gleaming half-formed thing, slop with your signature on it. The slop was yours before the machine touched a word of it. AI slop was only ever human slop with the friction removed. So the prompt is the wrong place to start. Start with the want. Clear the slop out of your own thinking, your own aims and desires. Focus on the thing you actually mean to make. Then, at that point, the cheap intelligence that was going to drown you becomes the longest lever a clear-eyed person has ever been handed.

Which leaves the question that has no market answer. You, the particular finite person who has read this far, are holding some quantity of the suddenly scarce asset. A name that can accumulate a record. A reputation spendable once. Attention, judgment, and a body whose presence in one room at a time is the only nonfungible thing you own. It is, in the one piece of business vocabulary that survives this whole argument, your only real moat: the thing that cannot be bought, copied, forked, or out-scaled, because it is welded to a single life and accumulates only under one name. The era’s economics will press you to keep all of it off the table: to produce without exposure, opine without commitment, attend without appearing, to let the tools carry your labor and your liability along with it. The pressure will be reasonable. The output will be fine. And if the argument of chapter eleven is right, the cost will be a self that, by inches, no longer contains the faculty that knows the difference. That is the last asset, and once gone, no subscription restores it.

What I am recommending asks less than heroism and more than abstinence. I have confessed exactly how this book was made, tools and all. The recommendation is the oldest practice there is, renamed by an economy that finally values it correctly: keep a stake in the world. Use every tool that makes the work better and refuse the one move that voids the point, which is letting the tool be the one who answers. The help is real and only the hollowness is fatal. The skill is telling them apart. Put your name on what you believe, and still make some things the hard way, since the collisions are the calibration. Keep an audience that keeps score. One reader who can tell is worth ten thousand who can’t. Pay the presence premium in the places no market prices, the funeral and the hospital corridor and the friend’s bad month, where everyone at the bedside understands without being told that the visit is the testimony. Be findable, let yourself be checked, and sometimes, at real cost, be in the room.

A claim that asks a lot of a reader has to be staked a lot; a thing that asks little needs only to be honest about asking little. A song you made because you wanted it to exist and liked how it came out, handed to a friend as exactly that, stakes almost nothing and counterfeits nothing. It is the opposite of slop however much of it a machine voiced, because you meant it and you stand behind it as yours. Slop was never the small claim. It was the small claim wearing the costume of a large one, the unmeant thing dressed as meant, the thing nobody chose pretending someone did. Truth and beauty are not switches set to on or off; they run by degrees, and most of a good life with these tools is built from things that are a little true and quietly beautiful and staked exactly as much as they claim, no more. You do not owe the world a martyr’s exposure on a birthday song. You owe it the smaller honesty: be what you say you are.

Not every flinch at the machine is the right one. Some of it is the true alarm this book has been tuning, the sense that no one is home. Some of it is only the origin showing, the reflex that rejects a thing for having touched a machine even when a person plainly stands behind it and the work is fine, which is the wrong disgust the book opened on, now wearing a virtuous face. And some of it is neither, a mismatch of register, the generated image that breaks the tone of a serious room without breaking a single principle, fitting badly without being false. Learning to tell the three apart, the real absence, the bare prejudice, and the merely out of place, is most of the maturity the moment is asking from us. Only the first is slop.

That is the maker’s half, and most people, most days, are not makers but ingestors, every waking hour, of a supply now flooded. The consuming side has a discipline of its own, lighter and just as decisive, and the book has already named its heart: the leaning moment. You cannot fact-check the feed and should not try, and a life spent suspicious of every sentence is its own kind of ruin. The discipline is only this: notice the few moments in a day when you are about to lean on something, and there, and only there, ask the one question the whole book has been sharpening, which is whether anyone stands behind this who would be embarrassed if it were wrong. Spend your trust where the answer is yes, and scroll the rest in peace. It is the same flinch the parents in Glasgow felt, one beat too late.

The crowd on O’Connell Street, the era’s perfect joke: ten thousand Dubliners assembled by a content farm for a parade that was never coming. But notice again what is actually in the photographs: human beings who, on the strength of a promise, put on costumes, gathered their children, and showed up in a body, in public, together, hoping to see something. Every element of that scene was real except the promise. The machines supplied the one part of the evening that costs nothing to supply. The crowd supplied everything else: the trust, the effort, the presence, the willingness to stand in the cold for the chance of shared delight. A system fed on exactly those investments while putting nothing at risk against them.

The crowd was the valuable part, the only valuable thing on that street, and the whole event was a machine for harvesting its worth. That is slop’s true shape, seen whole: a one-way extraction of human stake, attention flowing from those who pay it toward processes that wager nothing in return. This is not what the authorless film does, or move 37 in Seoul, or the protein folded by no hand. Chapter four called those found beauty: gifts without givers, honest about what they are, real as a canyon. O’Connell Street’s harm lay one level deeper, in a promise without a maker behind it. And yet a world where nobody stands on the curb anymore would be a poorer world, and it is exactly the world the extraction produces if the only lesson people draw is to stop showing up. The better lesson is to spend the showing-up, yours, finite, unreissuable, where someone has paid for the promise. On the witnesses. On the makers with names. On the institutions that still send the chapter back, the curators still numbering their spines, the four reporters with a subscription button, the friend who came in person. They are easier to find than they have ever been, because everything around them has stopped resembling them.

When intelligence is free, the cost is the value. That is the whole economics, and the whole ethics too. The parade is real wherever someone answers for it.

Dignity, in the sense this book has pressed, is the capacity to be answerable. A machine can counterfeit every sign of it and hold none, because it has nowhere for the cost to land. That standing is the one thing the world cannot take from a person who keeps paying for it. This is the cost of testimony.

Notes (6)

1. “Antiqua et Nova: Note on the Relationship Between Artificial Intelligence and Human Intelligence,” issued by the Vatican’s Dicasteries for the Doctrine of the Faith and for Culture and Education, January 28, 2025. Vatican. ↩︎

2. Pope Leo XIV’s encyclical on artificial intelligence and human dignity, Magnifica Humanitas, was signed May 15, 2026 (the 135th anniversary of Rerum Novarum) and released May 25, 2026. Vatican News; America Magazine; CNN. ↩︎

3. Rerum Novarum (“Of New Things”), Pope Leo XIII, May 15, 1891, the Church’s first great social encyclical. Vatican. ↩︎

4. Dario Amodei, “Machines of Loving Grace” (October 2024), and subsequent interviews. Amodei, who leads Anthropic, sets a radically optimistic case for AI-accelerated science (curing diseases, compressing decades of progress into years) beside hard safeguards, and has been candid that the concentration of frontier capability is an unsolved problem. His own framing favors a democratic coordination of the technology and serious safety regulation, rather than either pure laissez-faire or outright government ownership; the point borrowed here is the open question he keeps returning to, not a specific policy he endorses. ↩︎

5. The principle is old and has many fathers. Lord Acton’s “power tends to corrupt, and absolute power corrupts absolutely” (letter to Bishop Mandell Creighton, 1887) is the epigram; Madison’s Federalist 51, on contriving a government that must “control itself,” is the engineering. The argument the conclusion presses is narrower than a theory of the state: what makes any concentration of dangerous power bearable is the existence of an outside that can compel an answer, which is why the axis that decides the matter is accountability rather than ownership. On the specific worry of frontier capability pooling in a few private hands, see Amodei’s “Machines of Loving Grace” (above), which raises it from inside the industry without resolving it, and the wider AI-governance literature on concentration and oversight. ↩︎

6. The clearest recent instance is the Pro-Human AI Declaration (2026), a set of principles drafted by a politically mixed group of scientists, faith leaders, child-safety, labor, and human-rights advocates and convened with the Future of Life Institute. It covers human control over consequential decisions, pre-deployment safety testing on something like the drug-approval model, labeling of machine-generated content, liability that cannot be shielded behind a tool, and a prohibition on racing toward superintelligence absent broad scientific consensus that it can be done safely and genuine public buy-in, this last echoing the 2025 Statement on Superintelligence signed by a long list of researchers and public figures. National polling reported alongside the declaration found majorities across the political spectrum backing essentially every principle. The point in citing it is not to endorse the list item by item; reasonable people will accept some principles and resist others, and the document is an opening position rather than settled law. The point is that the outside this chapter insists must exist is being actively designed, and that the appetite for it is broad rather than partisan. For a practical on-ramp aimed at citizens, journalists, and governments, see the Future of Life Institute’s “A Better Path for AI.” ↩︎