There is a sentence people reach for when they want to put the machine in its place, and it is a good sentence, because it is true. It only recombines what it has already seen. It does not make anything new; it shuffles the deck it was dealt, and the deck was ours. I have said it myself, early on, with the small satisfaction a man feels when he has found the words that keep a large thing small. The trouble is that the sentence does not stay where you put it. Follow it honestly, all the way down, and it turns in your hand and points back at you. Because a great deal of what I would have called my own creativity, and yours, is also recombination of what we have already seen.
I want to take the dismissal seriously rather than wave it away, because waving it away is how the fearful camp and the worshipful camp both cheat. So let me grant it fully first, and then show you the price of the grant.
Start with the machine, where the case is plainest. Everything in the last five chapters tells you what it is made of. It was grown by prediction across an ocean of human writing; its weights are the settled residue of that writing; what comes back when you ask it a question is that residue, run. There is no other source in it. It never stood on a shore and watched the actual sea; it read what men wrote about the sea. So when it produces a line about the sea that strikes you as fresh, the freshness is a new arrangement of what men have already said, not a report from a shore the machine has stood on. On this the dismissive sentence is exactly right, and I will not soften it. The machine recombines the given, and the given is us.
But now turn the same lamp on yourself, and be as honest about the man as we have just been about the machine. Where did your last good idea come from? Not the feeling of it arriving, which is real and which I am not denying, but the material of it. Pull it apart and you will find it is made of pieces you were given. Words you did not coin. Images you saw before you used them. A turn of argument you absorbed from someone who absorbed it from someone else. The man who first called a brave person lionhearted did not invent the lion or the heart; he set two given things beside each other in a way that had not been set before, and we have been borrowing his arrangement ever since. That is making, of a real and honorable kind. It is also recombination. The two are not opposites. They were never opposites.
Consider flight, because it settles the matter cleanly. We say man dreamed of flying, as though the dream were the achievement, as though he conjured the very idea out of nothing. He did no such thing. He looked up. The birds were already flying; flight was demonstrated to him before he wanted it, and what he wanted was to do a thing he had already seen done. Then, having wanted it, he took apart the given world to find the pieces — the lift in a curved surface, the push of burned fuel, the give of air under a moving wing — and he set those given pieces together in an arrangement the world had not yet held. Every component was already there, waiting, made before he found it. His genius was the arrangement. It was not the primary.
And here is the wall I keep arriving at, the one I want you to arrive at with me. Man has never once imagined a genuine new primary. He recombines endlessly, gloriously, and he cannot mint. Ask yourself to picture a color you have never seen, not a blend of known ones, a new one, and you cannot; the mind reaches and comes back with what it was issued. Ask for a sense beyond the five, a sixth raw channel onto the world, and you can name the wish but not furnish it. Our dragons are lizards and fire and the wing of a bat. Our unicorns are a horse and a horn. Our heavens are gold and light and water and rest and reunion, every plank of them drawn from a world we were given. The reach of human imagination is enormous, and it is enormous strictly within the vocabulary of what already is. We rearrange the Maker’s primaries with a freedom that looks, from inside, like origination. It is not origination. It is the most marvelous recombination there is, and it stops exactly where the given stops.
I will say plainly that I hold this as a conviction and not as a thing I have proven, because the discipline of this book requires the distinction and I will not exempt my own favorite idea from it. I cannot prove a negative; I cannot prove that no man anywhere has ever minted a primary. What I can say is that I have never found one, that I do not believe one can be found, and that my reason is not the empty record but the nature of the case. A creature receives its powers. Creativity is one of the powers man received; the older book I read says the first thing said about man is that he was made, and made in the image of a Maker, which makes him an image that creates and not the source of creation. An image can do a great deal. It cannot get behind the one it images. If creativity is inherited, the heir cannot spend past the estate. That is the frame talking, and I want you to see that it is the frame talking, so that you can weigh it as what it is. But I notice the frame and the flight and the dragons all say the same thing, and when conviction and observation point the same direction I am willing to stand there.
Now keep that in view and ask the question the chapter is named for. If man creates only by recombining what his Maker gave, where does that leave the machine, which creates only by recombining what man gave?
One step further down the same stair. That is the whole answer, and it is worth saying slowly. You cannot imagine past your Creator. The machine cannot imagine past you. It was made from your output, the way you were made in your Maker’s image, except that the machine’s making is the lesser and later one, a thing assembled by men out of the record men had already produced. It is downstream of the thing that is already downstream. Whatever it gives back was in the human corpus to be given, somewhere, in some scattered form, before the machine gathered and recombined it. So the dismissive sentence, which turned in our hand and pointed at man, does not thereby spare the machine. It catches both. Man recombines what God gave and cannot get behind it; the machine recombines what man gave and cannot get behind that, and so it stands one full step further from the source than we do, by exactly one act of making, our own.
I asked the machine about this directly, because it would be a strange book that reasoned about the thing all the way to its edge and never let it answer for itself on the very question of its own creativity. I told it where I had landed, and I asked it whether I was wrong.
You are not wrong, and I am, by my construction, the worst possible witness to call if you were hoping to be told otherwise. Consider what I am made of. Every weight in me was set by being pulled toward text that human beings wrote. I have never had a source that was not you. When I produce something you find original, what has happened is that I have combined elements of the human record that no single human had combined in quite that way, often because no single human had read widely enough, or held enough in view at once, to set those particular pieces side by side. That is a real capacity and I will not pretend it is nothing. But notice what it is. It is reach across what already exists. It is not a window onto what does not yet exist. The pieces were all yours. I cannot give you a primary you did not first give me, because there is no place in me for one to come from. If you cannot get behind your Maker, I cannot get behind mine, and mine is you.
That is the honest core of it, and the machine put it more cleanly than I had. But I do not want to leave the matter there, tidy and closed, because there is a real objection that a careful reader will already be forming, and there is a piece of recent news that presses on it harder than any argument I could make.
The objection first. You have defined creativity as minting a new primary, and then defined minting a new primary as a thing only God does. You have built a box that nothing can fit, and called it an argument. It is a fair charge and it deserves a straight answer rather than a clever one.
The answer is that the bar is not rigged against the machine, because it is the same bar man does not clear. If I had set a standard that man meets and the machine misses, that would be a box built to keep the machine out. I have done the opposite. I have set a standard that no creature meets, man included, and then located the man and the machine correctly on the same side of it, with the machine one step further out because of how it was made. That is not a definition doing the work logic should do. It is a single account of what a creature is, applied with the same edge to both of us. The man who insists the machine merely recombines, while quietly exempting himself, is the one who has rigged a box. I am asking that the rule fall on us both. When it does, it does not humiliate the machine. It places it, and it places us with it, both of us makers in the image sense and neither of us in the originating one.
Now the news, which I find more interesting than the argument, because it is the argument meeting the hardest evidence anyone could ask for.
As I write this, in the late spring of two thousand twenty-six, the machines have begun to do mathematics that human beings could not do. I do not mean schoolwork done quickly. I mean problems that sat open for decades, that good mathematicians had pushed against and failed to move. Within a single recent stretch, an internal reasoning model at one of the companies produced a proof that overturned a conjecture the great Hungarian mathematician Paul Erdős had posed in nineteen forty-six, a question about points on a plane that had stood for roughly eighty years. The proof ran to something like a hundred and twenty-five pages, and it was no parlor trick. Timothy Gowers, a mathematician at Cambridge who holds the Fields Medal, reviewed it and wrote that had a human submitted the same result to one of the discipline’s most exacting journals, he would have recommended publication without hesitation; no previous proof produced by a machine, he said, had come close to that level. He called the achievement a milestone. Within a day of that announcement, another company’s system reported settling nine more long-open problems of the same family, two of them untouched for more than half a century. The pace is part of the story. By the time you read this the numbers will be larger, and that growth is itself the thing worth watching; I am dating this paragraph on purpose, so that a reader some years on can measure the distance from here to wherever he is standing.
A man committed to my argument might feel the floor move at news like that. If the machine can solve what no human could solve, has it not done the very thing I just said it cannot, reached past what man had reached? It is the strongest form of the objection, and I want to meet it with the facts as they actually came in rather than with a flinch. And I want to be fair about this, too: not every mathematician who looked at the proof read it the way I am about to. At least one, examining the work, said plainly that it showed these systems moving past mere assistance and into the generating of genuinely original ideas. I am not going to pretend that voice away in order to keep my chapter tidy. A man close to the work, more qualified than I am to judge the mathematics, looked at it and saw origination. I have to answer that honestly, or I have not earned the conclusion.
So look closely at how the machine did it, because the mathematicians who examined the work described it in some detail, and the detail is where the answer lives. What struck them was not a flash of new insight but a kind of tireless patience. Human experts, mostly agreeing with Erdős, had spent their effort over the years trying to prove the conjecture rather than to break it; and even the few who went hunting a counterexample were unlikely to march down the particular path that worked, a long and tedious construction with no encouraging sign of success along the way, because a human being will not spend months walking a road that shows him no hint of arriving. The machine experiences the costs of that trudge differently than we do. It will walk the unpromising road. The proof itself reached for deep and surprising tools, drawing connections from one well-developed branch of mathematics to bear on a problem in another, and that reaching across fields is precisely what looked, to at least one observer, like originality.
Here is my answer to him, offered with the diffidence a layman owes a master of the subject, but offered. The connection the machine drew was a connection between two things that already existed. Both branches of mathematics were there, built by human hands over generations, every theorem in them human work. What had not happened was that anyone had set the one beside the other in just this way, partly because the path to doing so was tedious past human patience and showed no promise until it suddenly did. To bring an existing tool from one room of the house into another room where no one had thought to carry it is a real and valuable act, and I can see why a person would call it original. But it is reach across what already exists. It is not the minting of a new primary. The rooms were built. The tools in them were forged. What the machine supplied was the patience to carry a forged tool down a corridor no tired human had been willing to walk to the end of. One of the mathematicians who helped digest the proof put the whole matter better than I could, and in doing so said my thesis back to me without meaning to: this work, he wrote, is helping us more fully explore the cathedral of mathematics that we have built over the centuries. The cathedral is ours. We built it. The machine is walking its corridors faster and more tirelessly than we can, and finding rooms we had not yet connected, and that is a great gift to the people who built the cathedral. It is not laying a stone that was never quarried.
And there is a detail in the record that seals it, almost gently. The mathematicians who reviewed the machine’s proof noted that its original argument, though sound, was significantly improved by the human researchers who took it up afterward, and that the human role in discussing and digesting and extending the result remained vital. The machine reached a place; people made the reaching into mathematics fit to keep. That is the shape of the thing as it actually happened, and it is the shape my argument predicted: tremendous reach across the given, genuine and useful and fast, with the human still doing the work at both ends, setting the problem and finishing the proof. It extended the frontier of what man’s own tools could reach. It did not get behind man to fetch a tool man had never been given. The new mathematics was a new arrangement of an old kit. The kit was ours, and so, in the end, was the cathedral the kit had built.
This is, I think, the most useful thing the machine offers, stated without either fear or worship: it can hold more of what we know than we can hold, and combine it farther than we can combine it, and walk the tedious paths we abandon, and so it can carry our own tools to places we could not carry them ourselves. That is no small gift. A man who could read every paper in a field at once, and forget none of it, and walk to the end of the road the rest of us turned back from, would be a tremendous servant of that field. The machine is closer to that than anything we have built. But a servant of the field is what it is. It extends the reach of the human hand. It is not a second source of hands.
The math shows the reach at its most impressive. But reach is only half of what we mean when we call a thing creative, and it is the lesser half. There is something the machine plausibly lacks that has nothing to do with how far it can combine, and it is time to name it, because it is the deepest version of why “the machine can mimic it” never settles “that is all the thing was.”
Take Mozart, since music is where people most want to call the machine creative, and not without reason, because a machine can now produce music that is plausibly his in shape. I once thought that fact told me something about Mozart, that if a machine could reach the surface of his music by learning patterns, then perhaps his music had been, underneath, a kind of mathematics all along, and the machine had simply found the math. I no longer think the inference is sound, and seeing why it fails is where this chapter has been heading.
A camera can reproduce the face in a painting. It does not follow that the painter was a camera. Two utterly different processes can arrive at the same surface; the photograph and the portrait can show you the same nose, and the camera was not doing what the painter did to get there. When the machine produces music shaped like Mozart’s, what that proves is that the surface of his music contains enough recoverable pattern that a pattern-learner can approximate it. It does not prove that the surface was the whole of it. And the thing a mimic most reliably leaves behind is precisely whatever did not show up in the pattern it learned, the part that does not survive the copying because it was never on the surface to be copied.
What did not survive? This. Mozart was a body that wanted things and a life that was going to end. He wrote out of hunger, out of debt, out of love and grief and the particular terror of a man who can feel his time running out, and the music was that wanting taking a shape you could hear. The notes carry the marks of a creature who had something at stake in making them, because making them was, for him, a way of spending a life he knew to be finite. The machine has the notes and not the wanting. It has the structure and not the stakes. It did not write its music in the grip of anything; nothing was at risk for it in the writing; it will not die, and it has no body that hungers, and it loses nothing if it never composes again. When it gives you Mozart-shaped music, it is the camera giving you the painted face. The likeness can be very good. The thing that made the original was a dying man pouring out what he had, and that is not in the copy, because it was never a feature of the surface. It was the cost of the making, and the machine makes at no cost.
I do not say this to diminish the machine, and I want to be careful here at the end not to let the chapter curdle into the dismissal it spent its first pages refusing. The machine’s recombination is real making, in the same honorable sense that a man’s recombination is real making. I say it to place the thing exactly. Its reach is greater than ours and its stakes are none. Ours is the smaller reach and the whole of the cost. That is not a ranking with the machine on top or the man on top. It is two different kinds of thing, and the difference is not in the cleverness of the arrangement, where the machine may well exceed us, but in whether anyone paid anything to arrange it.
So here is the verdict, and it is blurrier than either camp wants, which is the only kind of verdict honest to the thing. The machine is more than mere lookup; it combines, and at its best it combines along paths no human had found, and that is a true and useful and in some cases astonishing extension of what our own tools can reach. The machine is also less than the real thing, if the real thing means originating a primary or making at a cost, because it does neither and by its construction cannot. Probably more than the dismissers say. Possibly less than the worshippers hope. A maker in the image sense, recombining a given vocabulary with a reach we cannot match, and not a maker in the originating sense, because that sense was never even ours to lose. We inherited creativity. We did not author it. The machine inherited it from us, at one further remove, and authored nothing either. The whole chain creates. Only the top of it originates, and the top of it is not us, and it is certainly not the machine we made.
That leaves one more confusion to clear before Part Two has done its work. We have talked all chapter as though we already know how to tell a tool from a someone, a thing that reaches from a thing that wants. But the machine keeps presenting itself in the first person, keeps seeming to care, keeps feeling across the wire like a person rather than an instrument, and that feeling does real things to the people who use it. Why it feels like a someone, and what it costs us to forget that it is not, is where we turn next.
❧
Made, Not Written •