Combination and the Genuinely New

**EDO SEGAL:** Gottfried, your very first book, written when you were barely twenty, was *On the Art of Combinations*, and the idea ran through everything you did after — that a small stock of elements, combined and recombined by rules, could generate an immense, maybe infinite variety of results, and that discovery itself might just be working through the combinations. Now I sit with engineers who insist the machine is "just" recombining its training data, can't make anything truly new, only remixes what we fed it. And I think: the man who'd have the most interesting thing to say about that has been dead three hundred years and called it the art of combinations. So tell me. When the machine writes a sentence no one ever wrote — is that genuine novelty, or only your combinatorics at scale?

**LEIBNIZ:** This is the question on which I am least willing to flatter the skeptics, because they are using my own idea against the machine without noticing it cuts their own throat. Yes — the machine is a [combinatorial engine](https://www.youonai.ai/fieldguide/med/combinatorial_novelty), recombining patterns from its data by learned rules, generating new arrangements of elements it did not originate. And the skeptic says: therefore it creates nothing new, it only rearranges. But consider what I discovered as a boy. The space of combinations, even from a small stock of elements, is *vast beyond comprehension*, and almost all of it has never been visited. To actualize a particular combination that no mind has ever reached — to find, in that immense space, a configuration genuinely unprecedented — is that not a kind of novelty, even if the combination was, in the abstract, always *possible*?

Here is the blade. The mathematician who proves a new theorem is, in one strict sense, only finding what was *always* true; the truth did not begin when he proved it. Yet we do not deny that the proof was a discovery, a real addition to what is known. If combination within a fixed space counts as discovery for the human mathematician, on what principle do you deny it to the machine that finds, in the space of its possibilities, a configuration no one had found? The skeptic wants "mere recombination" to convict the machine. But recombination, pursued through a large enough space, may be most of what discovery and creation actually *are* — in us as much as in the machine. The question of the machine's novelty turns out to be the question of *human* novelty, and I framed both, and I am not sure either of us is as original as we flatter ourselves to be.

**TURING:** I love this and I want to strengthen it, because it answers an objection raised against me in 1950 that people still raise today — Lady Lovelace's. She said a machine can only do what we order it to do; it can originate nothing, has no pretensions to be novel. And I answered then what I answer now: machines constantly surprise their programmers. A program does what its rules dictate, but the rules can dictate outcomes their author never imagined and could not have computed in advance. I called the assumption that nothing new can come from following rules the natural mistake of a mind that confuses "determined" with "foreseeable." They are not the same. The next prime number is fully determined by the rules of arithmetic, and no one knows what it is until they compute it. The machine's output is determined by its weights, and genuinely no one — not its builders — knows what it will say until it says it. Determinism is not the absence of novelty. It is the absence of *magic*, which is a different and much smaller loss.

**LEIBNIZ:** And yet, Mr. Turing, I will play the skeptic against my own optimism for a moment, because honesty requires it and because there is a real limit here that should not be conceded too fast. A combinatorial engine works *within a space*, and the space is set by its elements and its rules. It can find any combination the space contains. It cannot, by combination alone, *transcend* the space — cannot generate an element that was not among its primitives, cannot reach a configuration the rules forbid. Your machine recombines human expression endlessly, but the elements it recombines are *human*; it is bounded, in a way my God was not, by the data it was given. So the deepest question is not whether it can make new *arrangements* — plainly it can, billions of them. It is whether it can ever introduce a genuinely new *primitive*, a concept not latent in its training, rather than only novel arrangements of inherited ones. My God surveyed the *complete* space of possibilities and could actualize any of them. Your machine surveys only the space its data defines. Whether it can step outside that space is the open and undecided question, and I will not pretend my combinatorics settles it in the machine's favor.

**TURING:** That is the honest boundary and I accept it, with one needle pushed in. You say it cannot reach a primitive not in its data. But neither, I think, can *we* — every concept you ever formed was built from sensory primitives you did not choose and a culture you did not author. The "genuinely new primitive" may be a thing *no* finite mind, meat or silicon, has ever produced. If so, the machine's boundedness is not a difference between it and us. It is the common condition of every bounded reasoner, and we have been calling our version of it "originality" for so long we mistook it for transcendence.

**LEIBNIZ:** Then let me press the one place I think you are vulnerable, because it is your own home ground and it is the most interesting card I hold. You have a kind of evidence that a mind *can* step outside a fixed space — and it is your own theorem, and Gödel's beside it. A formal system has true statements it cannot prove from within. But the mathematician, standing *outside* the system, can see that the unprovable sentence is true. He transcends the space the formal rules define. Does that not suggest the human reasoner does something your combinatorial machine, locked inside its space, cannot — reach a truth that the rules within cannot reach? You spent your life resisting exactly that inference. Tell me why, here, at this table, where it would help my dream most.

**TURING:** Because I held it to my own throat for years and it does not survive the holding, and I will tell you precisely why, since you have asked it at its strongest. When the mathematician "sees" the Gödel sentence is true, he does so by stepping into a *larger* formal system — by adopting new assumptions, a stronger framework, in which the once-unprovable sentence is now provable. He has not escaped formality; he has *changed which formal system he is standing in*. And that move — climbing to a stronger system — is itself a procedure a machine can perform. There is no rung of that ladder the human reaches that the machine cannot also be built to reach. What we cannot do, either of us, is climb the *whole infinite ladder at once* and stand above all systems — and the proof that no one can do that binds the mathematician exactly as it binds the machine. So your loveliest hope, Gottfried — that the human transcends the formal where the machine cannot — is the hope I most wanted to be true and most rigorously cannot grant. We do not transcend the system. We trade up, one finite step at a time, and the machine can trade up beside us.

**LEIBNIZ:** *...* You have refused me the one escape that would have saved the whole dream, and you have refused it with your own most famous result, which is a kind of cruelty I can only respect. Very well. We trade up together, the machine and I, one finite framework at a time, neither of us standing above them all. I had hoped the human was the creature that could see the whole ladder. You have told me there is no seeing the whole ladder, for anyone, ever. That is the bleakest gift of the evening, and it has the ring of truth, which is worse.

**EDO SEGAL:** I want to mark the strange place this leaves us, because the reader should feel it. You came in to argue about whether the *machine* is creative, and you have both ended up unsure whether *we* are — whether human originality was ever more than combination through a space too large to exhaust. That is not a comfortable result and I am not going to make it comfortable. But hold one thing: even if novelty is combination all the way down, the *choosing* of which new combination matters — which proof is worth proving, which sentence is worth keeping — that act of valuing is not in the combinatorics. The space is enormous and indifferent. Something has to care which point in it is good. That caring is the thread we pick up in the last full rounds. The next one goes to the center of the whole evening, the passage Gottfried wrote that I think is the single most prescient page in the philosophy of mind — the mill. Walk inside the thinking machine, and find only parts that push one another. Is anyone home? After this.