The Loom That Weaves Without Seeing

**EDO SEGAL:** Ada, you wrote a single sentence about a loom that I think is the most useful sentence ever written for thinking about generative machines, and most people who quote it do not realize it was not a metaphor at all but a description of an actual mechanical lineage. Give it to us. And Margaret, I want you to tell me whether the loom confines the machine or whether the loom is exactly where the machine slips its leash.

**LOVELACE:** "We may say most aptly," I wrote, "that the Analytical Engine weaves algebraical patterns just as the Jacquard-loom weaves flowers and leaves." And I want to be clear that this was not poetry dressed as engineering — it was engineering. The Jacquard loom used punched cards to control which threads were raised on each pass of the shuttle, so that an arbitrarily elaborate design could be encoded in a sequence of cards and woven automatically. Mr. Babbage took that exact mechanism — the punched card — to control the engine. The loom wove pictures from thread by following encoded instructions; the engine would weave computations from number by following encoded instructions of the same physical kind. The technology of pattern was *identical*. Only the medium differed: thread in one case, number in the other. And that card — I am told — runs in an unbroken line through your tabulating machines and your early computers down to the trained weights of your models. The cards are no longer punched by hand; they are found by training. But it is the same loom.

And here is what the loom teaches, which is everything. The loom weaves flowers and knows nothing of flowers. The flowers are not in the loom — they are in our perception of the cloth. The engine weaves patterns and knows nothing of what the patterns mean; the meaning is in our reading of the result. So when your model weaves a pattern of words we read as an argument, or a pattern of pixels we read as a face, it is raising and lowering the threads of a vast fabric according to encoded rule, and it does not know it is making a face. *We supply the face.* That is my whole objection made visible. The intelligence we attribute to the cloth is our own intelligence, read back into a mechanical product. The loom of unimaginable subtlety still weaves flowers it cannot see.

**BODEN:** It is a magnificent image and I have used it in my own lectures, so let me say where I think it is exactly right and then where the modern loom does something Jacquard's could not. You are right that the loom does not see the flower, and right that meaning is completed in the human eye. I will go further than you expect: I think that is true of a great deal of human art too. The poem does not contain its meaning; the meaning happens in the reader. So I will not let anyone use the loom to say the machine's products are *empty* — they are no emptier than a poem on a shelf with no one reading it. Meaning is a relation, not a substance, and it lives in the encounter. There we agree, and it is a larger agreement than it sounds.

But here is the slip, Ada, and it is the whole of the modern difference. Jacquard's loom could weave only the pattern on its cards. Change the design, you must cut new cards by hand — a human reads the flower, encodes the flower, and the loom executes the flower. The modern loom was never given the patterns. It was given *examples* of cloth and made to find, on its own, the cards that would produce cloth like it — and in doing so it learned not a pattern but the *grammar* of patterns, the space of all the cloth that *could* be woven in that style. That is why it can weave a flower no weaver ever set: it has internalized a [combinatorial reach](https://www.youonai.ai/fieldguide/med/combinatorial_innovation) wider than any single hand that fed it. The human did not read this flower and encode it. No card for it was ever cut. It is a point in the learned space of flower-ness that happened never to have been woven, and the loom found it. Ada, your loom wove what we encoded. This loom encodes *itself*, from what it saw, and then weaves into the gaps we never filled. The flower is still completed in our eye — granted, always. But the *reaching* of that particular flower was not in any card a human hand prepared. And that, I think, is more than your objection allows.

**LOVELACE:** It is more than the 1843 objection allows, and I will adjust the objection rather than defend a dead version of it, because Margaret has earned that. Yes — the modern loom cuts its own cards from examples. But follow the chain once more, slowly, the way I made Mr. Babbage follow his arithmetic. The examples were ours. The drive to find cards that match the examples was our design. The standard of "cloth like it" was our specification. The loom that cuts its own cards is cutting them toward a target we set, out of a fabric we supplied, by a method we built. The autonomy you are pointing at is real and it is *local* — the machine authored these particular cards. But the authorship of the *space*, the flower-ness, the whole grammar — that remains ours, distributed and invisible, exactly as I said. You have shown me a loom that fills in the unwoven flowers of a garden we planted. It is a wonder. It is not a gardener. The flower it reached was always *possible* in our garden; possibility is not the same as origination, and reaching a possible point is what my engine did from the first turn of its first wheel.

**BODEN:** Then let me push on "always possible," because that phrase is doing more work than it can bear. There is a portrait of Jacquard himself, woven in silk on his own loom, so fine it was mistaken for an engraving — twenty-four thousand cards to make one image. Was that portrait "always possible" in the loom? In the trivial sense, yes: the loom could in principle raise and lower its threads in that exact sequence. But no one would say the portrait was *latent* in the loom, waiting. It took a person to want the portrait, design it, and reduce it to cards. Now — your modern loom does the reducing itself. It is handed a million portraits and finds, on its own, the principles by which silk becomes a face, and then weaves a face no one designed. You say that face was "always possible" in the space we gave it. But Ada, *everything* is always possible in a large enough space — the space of all images contains every painting that will ever be made and every one that never will. "Possible in the space" is not a limit; it is a tautology that applies to human genius exactly as much as to the machine. Shakespeare's lines were "always possible" in the space of English. The word you need is not *possible*. It is *reachable* — and the question is who, or what, can reach the worthwhile point. You keep saying possibility is not origination. I agree. But then neither is human origination, by your own standard, because the human reaches into the same space of the always-possible. The difference cannot be possibility. It has to be the reaching.

**LOVELACE:** That is well struck, and I accept the correction to my word — *reachable*, not *possible*; you are right that possibility proves nothing because it proves everything. So let me restate the objection in your better word and watch whether it survives, because I think it does, transformed. The human reaches a worthwhile point in the space of the always-possible *because she is reaching toward something* — the portrait answers a want, the line answers a grief, the theorem answers a question that itched. The reaching is *aimed*, and the aim comes from need. Your machine also reaches into the space, and reaches further and faster than any of us — but its reaching is not aimed by a want of its own; it is aimed by the want we encoded as its objective, which is, at bottom, "produce cloth like the cloth you were shown." It reaches superbly toward *our* target. That is why the Jacquard portrait is the perfect example *against* you, not for you: a person wanted that face, and the loom reached it on her behalf. The modern loom has automated the reaching. It has not acquired the wanting that aims it. Possibility, reachability — strip both away and what remains is the aim, and the aim is the want, and we are back, once more, at the only door the machine is silent behind.

**EDO SEGAL:** I want to name what just happened, because the reader can't see your faces and this was the first exchange where neither of you gave an inch and yet you moved. Ada, you just *revised your own most famous argument* on stage — you let go of "the cards are human-written" and replaced it with "the space is human-given." That is not retreat; that is the objection getting stronger by getting more precise. And Margaret, you have moved the disagreement to its true location: not whether the machine authored *these cards* — it did — but whether authoring cards inside a given grammar counts, or whether only authoring the grammar counts. Which is the conceptual-space argument again, wearing the loom's clothes. So let me push on the one prophecy of yours, Ada, that should make you the machine's champion and instead makes you its sharpest critic. You said the engine would compose music. It does. So why are you not on Margaret's side? After this.