New to You, New to History

EDO SEGAL: Margaret, lay out the distinction. And then I want to hand each of you the same machine output and watch you classify it differently, because I think that is where this gets real.

BODEN: Two scientists, working alone and unaware of each other, arrive at the same theory. The first to publish gets the credit; the second gets a footnote. But did the second do something less creative? They performed the identical mental act. The only difference between them is a fact about the rest of the world — who got there first. I refused to let that external accident contaminate our account of the internal act, so I split the word. P-creativity, psychological creativity, is the generation of an idea new to the mind that produces it, regardless of whether anyone else ever had it. H-creativity, historical creativity, is reserved for ideas new to the whole of human history. And the crucial point, easily missed: H-creativity is a special case of P-creativity. Every historically novel idea was first psychologically novel to someone. A child who works out for herself why the sky is blue, never having been told, has done something genuinely creative even though the answer is in every textbook. My framework honors that act. The fame-based definition throws it away.

EDO SEGAL: So point it at the machine.

BODEN: Watch what it does to the standard dismissal. People say: these systems merely recombine their training data, producing nothing new, only sophisticated remixes of what humans already made. My framework grants the premise and resists the conclusion. Yes — the system works from what it absorbed. But so does every human. The child reasoning to the blue sky is working from absorbed concepts too: light, scattering, color. P-creativity never required that the materials be invented from nothing; it requires only that the combination or the discovery be novel to the producer. So when a model produces something genuinely new relative to what it was given — and increasingly it does — that is P-creativity. It is real. But — and I insist on this with equal force — the prize we actually revere is H-creativity, the idea that reorganizes a whole field, that no one ever had. And there the machine is weaker, almost by construction. A system trained on the existing corpus is a map of what has already been thought. Its center of gravity is the already-said. It interpolates brilliantly inside the territory of human achievement. Whether it can push the boundary outward — that is transformation again, now sharpened. P-creativity the machines have, in abundance. H-creativity is the contested frontier.

EDO SEGAL: Let me put a hard case to both of you, because I think it tests the frontier exactly. A machine played a board game — Go, a game humans had refined for two and a half thousand years — and on the thirty-seventh move of a famous match it played a stone that every master watching called a mistake, an error, the kind of thing no strong human would ever play. And it won. And within a year, human masters were studying that move and the family of moves like it, and the game's own theory shifted to accommodate them. Margaret — is that not exactly your H-creativity? New to history, surprising, and demonstrably valuable, since it won and reshaped the field?

BODEN: It is the single best candidate anyone can offer, and I will not wave it away, because it does something my framework predicted was possible and many said was not. The move was new to the whole history of the game — H-creative by the strict definition. It was surprising — it violated the trained intuitions of every expert. And it was valuable in the one domain where value is not a matter of taste: it won. So yes, Edo, that is as close to machine H-creativity as we have seen, and Ada must reckon with it. But notice why it is clean, and notice what it cost. Go has a closed, formal evaluation — winning. The machine did not need a human standard of value because the game supplies one, unambiguously, at the end. The transformation was real, but it happened inside a space where value is given by the rules themselves, not by a needing mind. Take that crutch away — ask it to be H-creative in poetry, in ethics, in a domain where what counts as winning is itself the contested thing — and the very evaluation problem Ada landed on me returns in full. So the Go move proves the barrier is not absolute. It does not prove it is gone. It shows the machine can transform a space whose value function we handed it complete.

LOVELACE: And that concession is the whole of my position, Margaret, so I will take it gratefully and press it. The move was new, surprising, valuable — granted, all three. But who decided that winning was what mattered? We did. The game's purpose, its single sharp standard of worth, was supplied entire by human beings before the first stone was placed. The machine originated a brilliant path to a goal it did not choose, could not have chosen, and does not care whether it reaches. That is exploration of a space whose value we defined — superhuman exploration, I will say it plainly, more than I would have granted in 1843 — but the origination of the point of the game was never the machine's. Show me the machine that decides Go was the wrong game to master, that the worthwhile thing was never winning at all, and invents a game no one asked it to invent because something in it needed that game to exist. That is H-creativity in the sense we actually revere, and it is the sense the closed evaluation hides. The thirty-seventh move is your strongest evidence and I thank you for it, because even at its strongest it required us to hand the machine the one thing my objection says it cannot supply itself: what is worth wanting.

EDO SEGAL: Hold that — the machine that invents the game no one asked for. We will need it in the Crossing. Ada, finish the thought you started, because the P- and H- distinction does something to my confusion at the desk and I want it named.

LOVELACE: It does, and I am going to use it to make my objection more precise than I could in 1843, which is the gift of arguing with someone who has better tools. Here is what I now see. When your machine produces something and you, in the chair, have never seen it before, you are having a creative encounter — it is P-creative for you. The novelty is entirely real on your side of the glass. But it may be a mere retrieval from a remote region of the training distribution on the machine's side — not P-creative for it at all, because nothing was new to the machine; it does not have a "new," having no standing model of what it already knew that the idea could be new against. Do you see what this does, Mr. Segal? It dissolves your confusion at the desk and it dissolves it in my favor. You felt met because you had a P-creative experience. That experience is a fact about you. It is not evidence of anything on the other end, because the same experience is produced whether the machine reached a genuinely novel point or merely surfaced a distant one you happened never to have visited. The feeling cannot tell the two apart. Only the audit can — and the audit, as ever, is yours to perform.

BODEN: That is fair, Ada, and I will not pretend it is not — the user's P-creative experience genuinely cannot distinguish machine retrieval from machine novelty. But you have proved less than you think, because you have made the symmetric problem visible too. I cannot, from the outside, tell whether your insight, Ada, was novel to you or retrieved from some corner of your reading you had forgotten you read. P-creativity-for-the-producer is an internal fact, hard to verify from outside, for the human exactly as for the machine. You want to say the machine has no "new" because it has no standing model of what it already knew. But that is an empirical claim about the architecture, and it is getting less true: systems are being built with exactly that — a model of their own knowledge state, an ability to register that something does not fit what they already represent. The day a system reliably flags "this is new to me, against my own prior model" is the day your dissolution dissolves. I do not say that day has come. I say you have described its arrival, and named the instrument that would detect it.

LOVELACE: Then we have narrowed it again, and I am content to keep narrowing, because every narrowing makes the remaining human ground more exact even as it shrinks. We agree the surprise in the chair is the listener's P-creativity and proves nothing about the machine. We agree that what would prove something is the machine registering novelty against its own model of what it knew — and that this is an architectural question, not a metaphysical one. I will say only that registering "this is new to me" is still not origination. The logarithm table, were it conscious, could register that a value was new to it. To notice novelty is not to have made it, and to have made it is not yet to have meant it. There are three doors here, Margaret, and the machine is loud at the first, quiet at the second, and silent — so far — at the third.

EDO SEGAL: Hold the three doors, because that image is going to organize the rest of the night: noticing novelty, making it, meaning it. We have spent an hour on the architecture of surprise. I want to come down now from the abstract, because there is one image in your work, Ada, that is not an abstraction at all — it is a machine you actually saw, and it has turned out to be the most literally accurate description of how these systems work that anyone has ever produced. The loom. After this.