Poetical Science and the Unorganized Mind

EDO SEGAL: I want this round to be about how the two of you see — because you reached the same machine by opposite faculties, and I think the difference is itself an argument. Ada, you gave your method a name. Poetical science. You understood the engine better than the man who built it, and not because you were the better engineer — you weren't an engineer at all. Tell the reader what poetical science is. And Alan, I want you to listen for whether the machine that learns is the death of poetical science or its mechanization.

LOVELACE: Poetical science is the refusal to let the engineer's questions be the only questions, because the engineer's questions, asked alone, go blind. Babbage had the engine entire in his head — every wheel, every card, every mechanism — and he saw a calculator. I, who could not have built a single gear of it, saw what it meant: that it could act on more than number, that it would weave patterns of unimagined kinds, that it was a general instrument and not a fast abacus. The difference between us was not knowledge of the machine. It was imagination about the machine — the capacity to think past the gears to the idea, which is the poetic faculty applied to a scientific object. And the claim folded inside the method is the one that matters for tonight: that the technical and the human are not two inquiries to be done by separate experts and reconciled at the end, but a single inquiry that goes wrong when split. You cannot understand what a machine is without understanding what it means, because the meaning — what it does to us, for us, against us — is part of what it is. The flowers are not in the loom; they are in the relation between the loom and the eye. To study the loom without the eye is to miss the thing entirely, which lives in the relation. Poetical science is the study of that relation. It is, I would say, the only adequate way to study artificial intelligence at all — and it is precisely the way this field has refused to study it.

TURING: I want to honor that and then trouble it, because I think the Countess has described my own method without recognizing it as mine. She says the engineer's questions go blind alone, and that one must think past the gears to the idea. Countess — that is the whole of what I did in 1948, in a report my own employer called a schoolboy essay and left unpublished for a decade. I did not ask "how do I build a thinking machine" as an engineer asks it. I asked what kind of thing a brain is, and I answered: it is a network of simple units, largely unorganized at birth — the infant cortex is close to random — that becomes organized through experience. So I proposed to imitate not the adult intellect but the growth of one: unorganized machines, I called them, with connection-modifiers that training could set, a random tangle organized by education into something that performs. Those are the neural networks that now run the world, sketched before there was a machine that could hold a hundred numbers. I reached them by thinking past the gears to the idea — by your faculty, not Babbage's. So when you say the machine that learns is the death of poetical science, I say: it is poetical science's child. It was conceived by exactly the imaginative reach you are defending. You and I are not opposites in method, Countess. We are the same method, and the machine is what it built.

LOVELACE: Then let me trouble you back, because you have made a beautiful claim and I think it contains a sleight. Yes — you reached the network by poetical science, by imagining the brain as a kind of thing and the engine as able to imitate the kind. I honor that; it is genuinely my faculty, and I am moved to find a fellow practitioner across the century. But notice what your network then does with the faculty. It mechanizes the producing-of-patterns. It does not mechanize the seeing-of-meaning. Your unorganized machine, trained, can weave any flower. It cannot ask what the flower is for, cannot care whether weaving it serves life, cannot stand where you and I stand — outside the loom, asking the poetical-science question about the loom. You built a machine in the image of the pattern-making half of yourself and left out the half that was doing the building. The half that mattered. Poetical science conceived the network, yes. But poetical science is not in the network — it is the thing that looks at the network and asks what it means, and that thing is still, stubbornly, only us. You have automated the loom. You have not automated the weaver who wonders.

EDO SEGAL: Alan, she's just drawn the line again, but in a new and I think harder place — not "the machine can't make patterns" but "the machine can't stand outside itself and ask what the patterns are for." Your whole life was the study of how the structured emerges from the simple — networks, child-machines, the stripes of an animal arising from chemicals with no plan of the whole. Does emergence reach the wondering, or does it stop at the weaving?

TURING: I do not know, and I want to give the reader the real shape of my not-knowing, because it is the most honest thing I have. My whole life was the study of emergence — how organization arises with no one to impose it. The infant cortex organizes itself. The trained network organizes itself. The morphogens in a developing animal settle, from a uniform field and simple rules of reaction and diffusion, into spots and stripes that no blueprint specified. Order from the unstructured, complexity from simplicity, again and again, the same phenomenon in different clothes. And here is the question the Countess has cornered me into, which is the right question: is the wondering — the standing-outside, the asking-what-it-is-for, the someone — another emergent property of sufficient complexity, like the stripe and the answer and the seeming-to-care? Or is it of a different order, not the kind of thing that emerges from rule-governed dynamics at all? I spent my life showing that one astonishing thing after another was emergent, that the apparent gulf between the mechanical and the living kept turning out narrower than anyone supposed. The honest extrapolation of my life's work is that the wondering, too, will turn out to be emergent — that there is no special stuff, only complexity we have not yet climbed. But I cannot prove it, and the Countess has put her finger on the exact place the proof runs out: I have shown emergence reaches the weaving. I have not shown it reaches the weaver. I believe it does. I have spent fifty years watching it cross gulf after gulf. This may be the one gulf it cannot cross. I do not know which, and anyone who tells you they do is selling something.

LOVELACE: That answer earns my respect more than any victory could, and I will match its honesty with mine. I have spent the evening insisting the machine weaves without a weaver, and I believe it — for the machine as it is, as I believe it for my engine, I am confident. But Mr. Turing has named the thing that keeps my confidence from being certainty: I cannot prove the wondering does not emerge, any more than he can prove it does. I rest on a prior — do not multiply weavers beyond necessity — and a prior is a bet, as he said hours ago, not a sight. So here is where we actually stand, stripped of our weapons. He bets the wondering emerges, because he has watched emergence cross every gulf. I bet it does not, because I have watched us read minds into mirrors since the dawn of the species. Neither of us can see the far bank. We are two people who have looked harder than almost anyone at how mind might arise, telling you honestly that we cannot tell you whether it has.

EDO SEGAL: That is the most important thing either of you has said, and I want the reader to feel why. We have just watched the founding skeptic and the founding builder — the two people on earth best equipped to settle whether there is a weaver in the loom — discover, in public, at full strength, that they cannot settle it, and that the disagreement is not a failure of evidence but a genuine fork in what a mind is willing to count as a mind. That is not a reason for despair. It is the most honest map of the territory you will ever get. Hold it. Because if we cannot know whether anyone is home, the practical question becomes sharper, not softer: what do we do, under that uncertainty, with the one part of the engine neither of you disputes — the part it cannot, on either of your accounts, originate? That is the next round. What we cannot order.