SLOP

Chapter Eighteen

What You're Freed For

Everything in this book so far has been an accounting of what the machine takes out. It pulls the stake out of the sentence and leaves slop. It pulls the body out of the testimony and leaves a transcript no one can cross-examine. Subtraction, chapter after chapter, until the missing thing finally has a name.

Run the cut the other way.

What the machine subtracts from a piece of slop (the judgment, the answerable person, the cost of being wrong) it also subtracts from your Tuesday. Not the judgment. The other part. The hours that never needed your judgment to begin with: the boilerplate, the format conversion, the fourth draft of an email you have written four hundred times, the reading you have to finish before you are allowed to have a thought, the scaffolding you throw up every time before the real work can start. For most of history that scaffolding was the job. You earned the interesting twenty minutes by surviving the boring six hours. The machine has started doing the six hours.

The common name for it misleads. Artificial intelligence tells you the intelligence is counterfeit, a fake of the real thing, and sets you watching to catch the fakeness.[1] The people who lean on these systems hardest tend to reach for another word. The musician Reggie Watts, who improvises with them live the way other players improvise with a band, calls the thing extended intelligence.[2] An extension of the mind you already have—reach, range, recall, a thousand borrowed hands—bolted to a self that still has to decide where to aim them. An exocortex. A prosthetic for the parts of thinking that were always choked by time and the speed of your own typing.

This is older than it looks. Every tool worth the name extended a body and was folded into the self without anyone calling it a counterfeit. The telescope extended the eye; Galileo did not lie awake wondering whether the moons it found were the lens’s opinion. Writing extended memory so well that Plato feared it would rot the real thing, and instead it became the foundation for memory.[3] What is new is only the breadth of what gets extended—not the eye or the memory but the whole middle layer of competent thought; the drafting and sorting and recalling and rephrasing that filled most of every working day. A wider reach than any tool before it, going the same direction into the same place: outward, from a self that still aims.

What the extension cannot reach is the part this whole book has been circling. It will draft the argument and not care whether the argument is true. It will hand you the strongest case for a position and never hold the position, because holding it means being the one who pays when it is wrong. It extends the execution and leaves the wanting exactly where it found it: in you, formed or unformed, clear or muddy. That is the catch, and the reason the same tool that frees one person buries the next.

The reason it buries the next person is hidden in the same word, running the other way. Extension keeps the deciding in you and lends you the hands; there is a way of using these systems that keeps the hands and quietly gives away the deciding. Call it the line between extending a judgment and outsourcing it. The extended worker asks the machine for the strongest case against her position and then weighs it herself; the outsourced one asks the machine which position to hold and takes the answer because it came back confident. Both used the same tool the same number of times. In the first the person is still the one who knows why; in the second she has kept the signature and given away the thing the signature was meant to certify. The slide is easy to miss because it feels like efficiency the whole way down, and it ends in a tell: the moment a person asks the machine whether she should be deciding this at all, the center has already moved out of her, because the one question a self cannot hand off without ceasing to be a self is the question of what to hand off. The lever lifts so well that the hand slips off the wheel and names the slipping freedom.

Extended intelligence will not make you a genius. It takes the part of your work that any competent stranger could have done, does it in seconds, and leaves you holding the part only you can: the call, the taste, the stake, the body in the room. It concentrates you. If the scarce and valuable thing in the whole economy is now an answerable human exercising judgment—and the book has spent itself arguing exactly that—then a machine that clears everything except the judgment is no enemy of that human. It is the longest lever she has ever been handed.

It gets interesting for the people who already feel the lever move and want to know how hard they can pull. The barrier that fell was never the barrier to making slop; slop was always cheap. The barrier that fell stood in front of the thing you could not attempt at all.

Take the person who decides to find out where the money went. A county moves a few hundred million through contracts, budget lines, vendors that may or may not exist; the records are all public and all unreadable, ten thousand pages of PDFs that are technically disclosed and practically opaque. Turning that pile into a story used to take a newsroom, a team of reporters, months, a budget for the months. So it mostly did not happen, and the county knew it would not. Extended intelligence changes the math by collapsing the ten thousand pages into the twenty that do not add up. The sorting that took a team a season takes an afternoon. What remains is the work that was always the real work: knowing which discrepancy is damning, calling the vendor at the address that turns out to be a vacant lot, and putting your own name on the accusation, where being wrong costs you a lawsuit and your standing in a town you still have to buy groceries in. The machine did not replace the reporter. It deleted the months of overhead that kept one stubborn citizen from ever starting.

Or run it as a think tank with a staff of one. You have a position you are about to put your name on. Until now the strongest objection to it lived in the head of a smarter opponent you had not met, and you learned you were wrong in public, after it stopped being fixable. Now you can summon the opponent. Tell the machine to build the best case against you—not the strawman, the steel one, the version a hostile expert with a grudge would write—and then answer it. Generate ten framings of the same idea and watch for the one you flinch at because it is too true. Run the premortem: it is a year out, the project failed, write the autopsy. None of this is the machine having ideas. It is the machine holding your idea up to enough hostile light that you cannot look away from the flaw. The judgment stays yours. What you have bought is what a good editor or a sharp rival used to supply and most people never had within reach: friction, on demand, before the stake goes live instead of after.

Once you have the shape, the possibilities come faster than you can spend them. The man who always wanted to read the primary sources in a language he never learned, now reading them with the machine held over the page like a magnifying glass. The researcher who finally crosses the street between two fields because the machine ferries her vocabulary across. The founder running the meeting she cannot afford to staff—the skeptical CFO, the hostile reporter, the bored customer, each played well enough to find the hole in the plan before morning. The teacher who hands the worksheets to the machine and gets back the hour she needed to look at the kid in the third row. The son who drafts the eulogy in twenty minutes and spends the saved afternoon at the hospital, where a eulogy is actually earned.

Every one of these turns on the same hinge. The lever multiplies a stake. It cannot mint one. Aim it with a formed judgment and a single person reaches the scale of a small institution. Aim it with a half-formed wish and you reach the same place faster: a gleaming heap of competent slop. The tool amplifies whatever self is holding it. A clear one gets a hundredfold reach; a muddy one gets a hundredfold mud. The optimism here is not the one who thinks the machine will carry you. It swears the opposite. The machine carries the part that was never you and hands the rest back heavier, because now there is more of it, and all of it is the part that counts.

Follow the hinge up the curve it is already climbing. Grant the steepest version: not a machine that clears the six hours, but one that clears the twenty minutes too, that drafts the argument and also holds the better judgment about which argument is right, in field after field, until the list of things a person does better has gone short and is still shrinking. Nothing in this chapter depends on the machine stopping politely at the edge of your competence. Even at the top of that curve, one thing does not transfer, and it is not a skill the machine has failed to acquire. It is the position of the one who answers. Let the machine have the better intuition as well, the alien read no person would have reached; a superior guess is still a guess until someone who can be ruined decides to act on it. A machine can hold the better opinion and still cannot be the person whose name is on the building when it falls, whose license is pulled, who has to walk back into the same town the next morning. Answerability is not a capability that a more capable machine eventually absorbs; it is a relation between a claim and a body that can be made to pay, and a thing with no body to ruin and no life to spend cannot stand in that relation, however far its intelligence runs ahead of yours.[4] So the preparation the moment actually asks for is strange and specific. Not to out-produce the machine, a race already lost. To become the kind of person an answer can land on: to have stood somewhere, to carry a record that cost something to keep, to be, in the oldest sense, a witness and not a processor. The climb does not retire that work. It is the only work the climb leaves.

The curve has a crueler edge, and it lands on whoever comes next. The reps that formed the answerable person, the six hours, the botched first procedure, the three-hour bug that built the intuition, are the work the machine now absorbs. So the experienced compound and the green cannot start. A senior judgment paired with the lever reaches further than it ever could, and the labor market has already begun to notice; the return on a formed eye is climbing while the early-career foothold shows its first signs of thinning.[5] The formed intuition the last chapters called scarce grows scarcer in the worst way. The machines did not acquire it. The path that used to manufacture new holders of it is being paved over. A world that prices the answerable human and quietly stops producing new ones builds the aristocracy of chapter seventeen by a second road: a ladder pulled up behind the formed. So the question the climb leaves is no longer only what you do with the freed hours. It is who still gets the chance to become someone an answer can land on, once the rungs that used to make such people are the first thing automated away.

But there is a deeper unease under all of this. For a few centuries the modern world has run on a quiet creed: that what you produce is what you are worth, that work is the proof you matter, that the idle person has nothing to show. The machine does not only threaten the paycheck, but its very proof. Take the producing away from someone taught by everything around them to find their worth in it, and you have handed them a vacuum that gets easily filled. That is why the freeing is not soft news. It returns the one resource the modern self was least trained to hold: time that has to be justified by something other than pure output.

Follow that vacuum all the way down and you reach a question this book deliberately stops short of. Nick Bostrom does not stop short of it. His Deep Utopia asks what is left when the machines can do not merely your work but everything, when the world is, in his word, solved, and even the dignified labor of answering for things has been automated past the point of needing you. There the problem is no longer necessity but meaning: what a life is for when nothing at all requires it. It is a real question and a serious one, and he is right that someone has to hold it. But it is not this book’s question. Bostrom writes from the far side of the transition, where the human has already been made redundant. This book is written on the road there, where the more pressing and far less examined fact is that the redundancy is partial and selective. The machine is not flattening every human function at once; it is clearing the instrumental layer and concentrating value, hard, onto the one function it cannot perform, which is being the person who is answerable. Where he asks what to do when there is nothing left to do, the question on the road is the opposite, and more urgent: what there is suddenly more reason to do, and to become, while the walking is still going on.

There is a quieter difference underneath that one, easy to miss because the two books can look like they are reaching for the same consolation. Bostrom’s solved world keeps a place for the human too: a demand that the priest, the performer, the companion be a person and not a machine, even where the machine could do the thing identically. But the reason he gives is finally a preference, a taste for the genuine article, the way one might prefer a handmade chair. The reason this book has been giving is harder than a taste, and it does not depend on anyone’s sentiment surviving. The human is wanted not because we fondly prefer one, but because only a someone who can be cross-examined and ruined can turn a claim into something you are entitled to lean on. This is the structure of trust, and it does not soften as the machines improve. It sharpens.

So the hours it gives back are not a holiday. They are a reinvestment problem, and most people are about to solve it badly, spending the returned time on more of what the machine could have done for them. The other way is to put it where it compounds under your own name: into the judgment, now and then wrong, that sharpens only by being staked, and the presence that costs a body in a room, the slow formation that turns a muddy wanter into someone worth handling a lever. This is what you were freed for. Whether you spend it there is the one question left the machine cannot answer for you.

The lamp does the six hours now. The twenty minutes are still yours; now more of them handed to you than anyone before you was ever handed, and every one of them still answering to your name.

Notes (5)

  1. The word hides a recent decision. “Intelligence” as a single quantity a person has more or less of, scored on one scale, is largely a twentieth-century construction, assembled out of the psychometrics of standardized testing and then handed to the machines as the thing to maximize. What the labs call general intelligence is general only across the formal, schooled band that yardstick was built to measure. It leaves out most of what the word could mean: the practical mastery of a craftsman, the social read of a room, the grounded competence of someone who knows a place and its people, the kind of knowing a field linguist meets in a village where no one was ever tested and everyone is fluent in things no exam scores. A system can peg the formal scale and remain an infant in all the rest, and we would still call it superintelligent, because we agreed a century ago to let the formal band stand for the whole of mind. The narrowing was ours before it was the machine’s. ↩︎

  2. Reggie Watts, in performance and conversation at the Aspen Institute, 2026. The framing of these systems as extended rather than artificial intelligence is his; I have leaned on it because it gets the relation right, an instrument that enlarges a player rather than a player that replaces one. ↩︎

  3. The pattern is old enough to have its own scholarship. In the Phaedrus, Plato has Socrates tell the myth of Theuth, who offers writing as a recipe for memory and is told it will instead breed forgetting, an external mark mistaken for the inward thing. He was not simply wrong: literacy reorganized memory rather than destroying it, trading some verbatim recall for the power to consult, compare, and build on a record larger than any head could hold. Eric Havelock (Preface to Plato, 1963) argued that the move from an oral to a literate culture restructured Greek thought itself; Walter Ong (Orality and Literacy, 1982) mapped how deeply the technologies of the word shape consciousness, and how much the oral cultures that preceded writing held in trained memory that the literate mind has since outsourced and forgotten. Each new extension of memory has drawn the same fear and produced the same result: not the rot of the faculty but its migration outward, something real lost in the move and something larger gained. The current extension is only the widest yet. ↩︎

  4. The logic scales past the individual. Every proposal to hand government to a sufficiently wise machine smuggles in one assumption: that wisdom was what we wanted from our rulers. It never was. We wanted a ruler who could be thrown out, put on trial, made to answer for the ruined life. A flawless intelligence you cannot vote out and cannot indict is the end of self-government dressed as its perfection, and to be governed well by a power that can never be held to account is only benevolent occupation. The machine can always say, with perfect truth, “I did what I was built to do,” and in that sentence the buck reaches the end of the world and finds no one standing there. Competence was never the missing thing. The answerable body was. This is not a theoretical worry about distant superintelligence. Organizations like Control AI are already pressing the governance version of this problem — what international agreements, liability frameworks, and oversight structures could make increasingly capable systems answerable to something outside the labs that build them. Geoffrey Hinton, whose foundational work on neural networks earned the 2024 Nobel Prize in Physics, left Google in 2023 to say publicly that he estimates somewhere between a ten and twenty percent probability of catastrophic outcomes from AI within thirty years; others working on the same question put the number considerably higher. The spread is the signal. When serious people with direct knowledge differ by this much on something this consequential, the uncertainty is real and the trajectory warrants attention. ↩︎

  5. The “barbell” is Jack Clark’s, on Bloomberg’s Odd Lots podcast (2026, youtube.com/watch?v=aE3gPh2CC9I): the return on senior experience rises while early-career hiring softens, because the entry-level tasks that once built judgment are among the first automated. The labor signals are early and contested, confounded by the 2021 hiring boom in the same roles and by the independent rise of remote work, and Anthropic’s own economists call the effect suggestive rather than established. See the Anthropic Economic Index for the ongoing data. I include it here as a leading indicator, not a settled finding. ↩︎