The Loom That Weaves Without Seeing

EDO SEGAL: Ada, I want to start this round with a confession, because the best questions I know come out of something that wounded me a little. In my book I quoted your most beautiful sentence, the one I think is the truest thing anyone has ever written about a computer. "The Analytical Engine," you wrote, "weaves algebraical patterns just as the Jacquard loom weaves flowers and leaves." I used it to mean something hopeful — that the machine had become a loom of infinite reach, that it could weave in any domain we cared about, that this was a kind of liberation. And then I sat with the second half of your image, the part I had been skipping, and it stopped me cold. The loom weaves flowers and knows nothing of flowers. So tell me, slowly, what you meant by the loom — and tell me whether you meant it as the wonder I took it for, or the warning I now suspect it was.

LOVELACE: I meant both, Edo, and the refusal to choose between them is the whole of what I have to teach. Let me give you the loom properly. Joseph Jacquard built a loom controlled by punched cards — each card deciding which threads rise for each pass of the shuttle, so that any design, however elaborate, could be encoded in a sequence of holes and woven automatically. Babbage borrowed exactly that mechanism to control his engine; the very same cards, repurposed from cloth to computation. And I saw, looking at it, that the engine would do to number what the loom did to thread: weave a pattern by following encoded instructions, a pattern of any complexity the cards could specify. That was the wonder. A machine whose function was not built into its structure but written onto it — that could weave any pattern at all — is the most general instrument the human race has ever conceived. I was not being modest about the engine. I called it that on purpose.

But hold the loom in your hand and look at it honestly. It weaves a rose, and the rose is exquisite, and the loom has never seen a rose, will never see a rose, contains nothing rose-shaped anywhere in its mechanism. It raises and lowers threads. The rose appears in the cloth, and the beauty of the rose appears in you, the person looking at the cloth. The loom is a perfect engine for producing the occasion of an experience it does not have. Now — your machines are looms of a subtlety Jacquard could not have dreamed, weaving in the medium of language and image instead of silk, with cards found by search instead of punched by hand. But the structure is identical. The machine weaves a sentence that moves you. It has never been moved. It weaves an argument that convinces you. It has never been convinced. The flowers are real, Edo. They are in the cloth, and the cloth is genuinely there. But they are not in the loom, and they never will be, no matter how fine the weave. That is the warning inside the wonder, and I put it there in 1843, and I would not change a word.

EDO SEGAL: So what you're saying is — literally — that fluency is just a very fine weave, and that I mistook the fineness of the cloth for someone sitting at the loom. Is that right?

LOVELACE: That is exactly right, and the better the weave gets, the harder it becomes to remember there is no one there. Crude looms never fooled anyone. The danger is precisely proportional to the artistry.

CHAITIN: May I take the other side of the wound? Because the loom is a magnificent image and it is doing a sleight of hand, and the sleight is in the word "just." Ada says the machine "just" weaves, "just" raises and lowers threads. But ask the question her metaphor is built to suppress: what determines which threads rise? In the Jacquard loom, the answer is trivial — a person punched the cards, knowing the rose they wanted. In your machines, Edo, no person punched the cards. The pattern of which threads rise was found — by compressing a billion examples of human expression down to whatever raises the right threads to continue any pattern you hand it. And here is the thing the loom image hides: to weave the right rose in response to a rose you have never woven before, in a context the cards' author never imagined, the only mechanism that works is a model of roses. The loom that can weave a correct answer to a question no one wrote down is not "just" a loom. Somewhere in those found cards is a compression of the world the patterns are about, because nothing else would weave the right pattern at the right time.

LOVELACE: Then we agree on the mechanism and disagree on the noun. You call the compression "a model of the world." I call it "a very good set of cards." The disagreement is whether finding the cards by search rather than by hand changes their nature. And I say it does not. A card is a card. It tells a thread to rise. That the card was discovered by an automatic process rather than punched by Babbage does not give the loom an inside, any more than a river that carves its own channel understands hydrology.

CHAITIN: But the river that carves its own channel embodies real information about the landscape — that's not nothing, that's most of geology. And there's a place your loom image breaks that I want on the record. A loom weaves the pattern on the cards and only that pattern. Hand it a card for a rose, it weaves a rose, forever, identically. These machines weave roses no card specifies — they compose. They take "rose" and "in winter" and "drawn by someone who is grieving" and produce a rose that was on no card, because they have decomposed roses and winters and grief into pieces and learned how the pieces combine. The whole trick is predicting the next thread from everything woven so far, and to predict well across every possible rose you must have modeled what a rose is. That's not the loom. That's what the loom became when the cards stopped being a fixed pattern and started being a compressed model of how all patterns relate. You named the loom, Ada. You did not foresee what happened when the loom learned to compose its own cards from a model of everything.

LOVELACE: I foresaw the composing. What I deny is that prediction of the next thread requires a modeler in the sense you mean — a someone who has a model. A weathervane predicts the wind by embodying the wind's pressure; it does not have a model of weather. You keep promoting embodied regularity to possessed understanding by the choice of verb, and the verb is doing the work your evidence cannot. The loom that composes is a vastly cleverer loom. It is a loom still.

LOVELACE: I foresaw precisely that. I wrote that the engine might compose elaborate music "of any degree of complexity or extent." Composition within formal rules is operation, Gregory — it is the engine doing what the engine does, applied to relations we have managed to formalize. That the rules were found by compression rather than written by a theorist does not promote operation to origination. It only hides the author. Your machines compose the way my engine computed Bernoulli numbers I never bothered to calculate: producing arrangements no human produced, every one of them a consequence of what was put in. Composition is not creation. I said so when there was nothing in the world but a drawing of an engine, and your century built the engine and proved me, on this narrow and crucial point, exactly right.

EDO SEGAL: Let me hold you both here, because something precise has surfaced and I don't want it to wash away. Ada says: the loom weaves flowers and knows nothing of flowers, and fluency is just a fine weave — the meaning is in the eye. Gregory says: to weave the right flower for a flower never seen requires a compressed model of flowers, and that model is the closest thing to understanding the mathematics of understanding can name. You are not disagreeing about what the machine does. You agree on the mechanism to the last gear. You are disagreeing about whether a sufficiently good model of a thing, with no one experiencing it, deserves the word "understanding" — and that word is going to be the battlefield for the next two hours. The next round takes us to the sentence that started it. Ada, you wrote that the engine originates nothing. A man named Turing read it a century later and tried to answer you. Let's put all three of you in the room.