The Octopus, the Hamburger, and the Background

EDO SEGAL: Professor Searle, before your room there was an older worry, and you spent decades on it: that meaning isn't in the symbols at all. There's a thought experiment that dramatizes it better than anything — an octopus on a telegraph cable, from Bender and Koller — and there's your own: the Background, and a man ordering a hamburger. I want to braid them. Start with the octopus, tell it like you'd tell a sharp fifteen-year-old, and then take us to the hamburger.

SEARLE: Happily, because it's my argument wearing a tentacle. Two people stranded on separate islands, connected by an old undersea telegraph cable, passing messages in English. Deep below, a brilliant octopus taps the cable. It has never seen an island, a coconut, a person — only the pattern of signals, which sequences follow which. It's a superb statistician, so eventually it cuts the cable and impersonates one islander, and the other doesn't notice, because most of conversation is well-worn pattern. Then one day the islander writes: I'm being chased by a bear, I have two sticks and a coconut, tell me how to defend myself, quickly. And the octopus has nothing — not because it's stupid, but because what's needed now isn't the pattern of bear-talk, it's bears. Sticks as levers, the physics of a charging animal, the difference between advice that sounds right and advice that keeps you alive. The octopus only ever had the form. Meaning — the relation between the signals and the world — never traveled down the cable. It couldn't. That's the model's situation: a brilliant statistician of a wake it has never seen the boat of.

Now the hamburger, which is mine and which I love because it's so ordinary. You walk into a diner and say, "Give me a hamburger, medium rare, with ketchup." Understood instantly. But your understanding rests on a colossal mass of unstated know-how — what I called the Background. You know you don't want the hamburger encased in lucite. You don't want it delivered next year. You don't want it a mile wide. None of that is in the sentence. The sentence radically underdetermines what's meant, and an indefinite mass of unspoken capacity fills the gap. "Cut the grass" and "cut the cake" share a verb your hands interpret completely differently without a flicker of thought. The Background is not more rules — that launches an infinite regress, because every rule needs further know-how to apply it. At the bottom there are no rules. Only capacities. Ways of coping you have and do not represent. And that, I argued, is exactly what classical symbolic AI broke on. You cannot write the Background down, because it is not the kind of thing that gets written down.

LEIBNIZ: I must stop you there, Mr. Searle, with admiration and with a confession against my own tradition — because what broke on the Background was my dream. You have just described, in your hamburger, the precise rock my characteristica universalis shattered against. I believed every concept could be decomposed into primitives and every rule stated exactly. Your Background is the proof that it cannot. The unstated mass — that is the world refusing to fit into my characters. So I grant it, fully, and it wounds the symbolic dream more deeply than any objection of my own era. But —

SEARLE: Here it comes.

LEIBNIZ: — but here is the turn that complicates your position as much as it vindicates the engineers, and I think you must face it as I faced mine. The deep-learning machine appears to have captured the Background after all — not by writing it down, which you and I agree is impossible, but by absorbing it statistically from the ocean of human text. The model handles your hamburger correctly. It knows, in some operative sense, not to make it a mile wide, because the unstated assumptions you pointed to are deposited, implicitly, in everything humans have written. The Background, which could never be stated as rules, can apparently be learned as a vast tacit competence soaked from the trace our know-how leaves in our language. That is a real partial vindication of the project you and Dreyfus declared impossible. The machine does the hamburger. Explain that, if the Background is unstatable.

SEARLE: And that's a genuinely good push, so let me answer it precisely, because the answer sharpens my point rather than surrendering it. The model has absorbed a statistical shadow of the Background — the patterns of how people who have the real thing talk and behave. It has not absorbed the Background itself, because the Background is bodily. It's practical capacity for coping with a world, rooted in having a body that acts and a life that's lived. The model knows how the word "balance" is used in ten million sentences. It has never balanced on anything. It has the linguistic shadow of embodied know-how without the embodied know-how that casts the shadow. And you can watch the difference in how it fails: it's spectacular on the well-trodden and brittle on the genuinely new — confidently producing plausible nonsense the moment it leaves the region the shadow covers. The competence is real where the world resembles the text and hollow where it doesn't. The shadow is detailed enough to be uncanny. It is still a shadow.

EDO SEGAL: There's a man who spent his life on exactly this and got mocked for it, and I want him seated before you answer, Professor Leibniz. Hubert Dreyfus. Berkeley philosopher, wrote What Computers Can't Do in 1972, argued from Heidegger that intelligence is embodied, situated, a matter of coping with a world rather than processing symbols about one. The AI establishment despised him. And here's the twist, Professor Searle — he was your colleague at Berkeley, and he was attacking the same symbolic AI you both helped bury. Whose ancestor is he?

SEARLE: He's mine, and I'll fight Leibniz for him. Dreyfus saw the Background before I named it — that expertise isn't rules, that the expert can't articulate his skill because the skill was never propositional, that coping with a world comes before and beneath any representation of it. He was right about classical AI and right early, when it cost him. And the deep-learning people think he's theirs now, because their networks have competence without articulable rules — knowledge smeared across weights, skill that can't explain itself. But that's the half of Dreyfus they kept. The half they dropped is the half that's mine: embodied. Situated in a world that pushes back, with a body that copes, with stakes. A network trained on text is exactly what Dreyfus would have called a degenerate case — all pattern, no situation. They kept "no rules" and threw away "being in the world," because the world is the part they still don't have.

LEIBNIZ: I will not fight you for Dreyfus; I will grant him to you, because the embodiment is your wound to press and you press it well. But I will take from him a different lesson than you do, and it returns us to my mill. Dreyfus said the expert's skill is not in his rules but in his coping — in a relation between a living body and a resistant world. Very well. Then ask: what is doing the coping? Not the rules, you say — agreed. Not the symbols. Something lived. And "something lived" is my question again in Heidegger's clothing: there must be a someone for whom the world resists, for whom the coping is coping with something that matters. The octopus does not cope; it computes the statistics of a cable. The model does not cope; it predicts the next token. Coping requires a coper — a subject with stakes, for whom the bear is frightening. Dreyfus and I want the same thing your machine lacks, Mr. Searle. You call it embodiment. I call it a perceiving subject. They may be the same hole with two names — which, I am beginning to fear, is the only discovery this evening will yield.

EDO SEGAL: Professor Leibniz, now steelman the octopus for me — what does it get right — and then tell me whether it cuts against you or for you.

LEIBNIZ: What it gets right is the warning, and the warning is double. It warns against assuming that competence within the data implies competence beyond it — every builder should have that engraved somewhere discreet. And it warns about us: that the islander on the beach will keep reading a mind into the cable long after the cable has stopped deserving it. Both true. Now — does it cut for me or against? It cuts for me, and this is where Mr. Searle and I stand shoulder to shoulder against the whole optimistic field. The octopus has the wake and not the boat. My mill has the gears and not the perception. His room has the squiggles and not the meaning. Three thought experiments, three centuries, one shape: all the structure, none of the inside. We are not three arguments. We are one argument, told by a German, an American, and a linguist, each in the idiom of our age.

EDO SEGAL: That's convergence number three, and it's the biggest of the night — mark it. The mill, the room, and the octopus are one argument: the form is not the meaning, the wake is not the boat, the gears are not the seeing. But I have to press the seam, because there's a counter that haunts me. The brain also sits in the dark, in a box of bone, receiving only patterns — spike trains, never water itself. If the brain is the octopus that got enough data, why is it not a mill, a room, a cable too? We'll open exactly there after the break, on the question your monad was built to answer and his biology was built to dodge: what, if anything, is actually home.