The Ghost of Hubert Dreyfus
**EDO SEGAL:** Hubert Dreyfus, Berkeley, wrote What Computers Can't Do in 1972. He spent his life arguing — out of Heidegger, out of the phenomenology of skill — that computers would never achieve real intelligence because intelligence is embodied, situated, a matter of *coping* with a world rather than processing symbols about one. The AI establishment despised him. But here's the delicious part: Dreyfus was attacking *symbolic* AI — rules and representations — and his arguments read today almost like a manifesto for learning machines. So whose ancestor is he? Each of you gets to claim him. Andy, you first.
**CLARK:** He's mine in the marrow, and I'll say exactly why. Dreyfus said the expert can't articulate her expertise because it was never propositional — it's [skillful coping](https://www.youonai.ai/fieldguide/med/embodied_understanding), acquired through a body engaged with a world, resident in the way you absorb a situation and respond before any rule fires. The master carpenter doesn't consult a theory of the chisel; the chisel has become part of him, withdrawn into transparent use, and he copes with the wood. That's the extended mind and the embodied mind in one sentence, written before I was out of graduate school. Dreyfus's whole point was that mind is *being-in-the-world* — situated, embodied, coping — and that the thing AI kept building was a disembodied symbol-shuffler that could never cope because it was never *in* a world. A language model is precisely what he'd have called the degenerate case: all pattern, no situation, no body, no stakes. He's my ancestor. The phenomenology is mine.
**EDO SEGAL:** Thomas, the same body, the same eulogy — but you're going to bury him with a different family. Make your case.
**HOBBES:** I will claim him by the half Mr. Clark quietly drops. Dreyfus said intelligence is *not rule-following* — that competence lives not in articulated propositions but in something acquired, smeared through the practitioner, inexpressible as rule. On that he and I are kin and your symbolic AI was the enemy of us both. For I never said reasoning was the obeying of explicit rules a man could recite. I said it was reckoning — the motion of residues, the adding of names below the level of any rule the reckoner could state. Your neural engine is reckoning of exactly that kind: competence with no recitable rule, knowledge smeared across connections no one can point to. Dreyfus killed the *rule-bound* machine. He did not kill mine. He killed the symbol-shuffler and blessed, without knowing it, the reckoner that learns. You keep his "embodied" and forget his "non-rule-bound." I keep his "non-rule-bound" and deny that "embodied" does the work you want. Between us, sir, I have the half that built the thing that actually works.
**CLARK:** No — and this is the cut that matters. You can't keep the "no rules" and amputate the "in the world," because for Dreyfus they were *one thing*. The reason expertise isn't rule-bound is *that* it's bodily coping in a situation with stakes. Drop the body and you don't get a purer, leaner Dreyfus. You get the opposite of Dreyfus. The "smeared competence" you admire exists *because* a body practiced against a resistant world until the skill sank below articulation. Take the world away and there's nothing to smear the competence *into*. You've kept the symptom and thrown out the cause. He's not your ancestor, Thomas. You've buried him in the wrong plot and put your own name on the stone.
**HOBBES:** And yet your robots — for I am told you have them now, machines with these models inside them, learning to grasp and to manipulate from their own fumbling against a world that pushes back — when one of them copes, in Dreyfus's word, when it skillfully copes with a resistant world and the coping sinks below any rule, do you inherit his position or flee it? You have made the body the price of admission. They are paying it. What then?
**CLARK:** Then I follow my own principle and my line moves — that's the honest answer, and I won't dodge it the way a partisan would. If a system has a body, acts in a world of real consequence, gets corrected by error that *costs* it something, and the competence sinks below rule through genuine practice — then yes, it's coped its way toward the thing Dreyfus meant, and I'd be a hypocrite to deny it on grounds of silicon. The [parity principle](https://www.youonai.ai/fieldguide/med/parity_principle) is a knife that cuts toward me too. But notice what you've had to grant me to get there. You've had to give the machine a *body* and a *world* and *stakes* — the three things your own picture said were unnecessary, the three things you spent the night calling a survivable defect. The moment you build the thing that could really cope, you've conceded my case and abandoned yours. You can have the coping robot or you can have "reason is nothing but reckoning." You can't have both.
**HOBBES:** Perhaps I can have both, sir, if the body proves to be only more cable — a richer stream of motions to reckon over, not a different kind of thing. But I confess the question is genuine and I do not hold it lightly. Let us say this much is settled between us: a reckoner with a body in a world of stakes is a different and more formidable creature than a reckoner without one. Whether the difference is of *kind* or of *richness* — that, we have not settled, and I begin to suspect we will hand it to your reader unsettled.
**EDO SEGAL:** Mark that — second real convergence of the night. You both now agree that *embodiment changes the creature*; you disagree only on whether it changes the kind or merely the richness. That's an enormous narrowing from where we started. Hold the thread — it returns when we ask what the death cross is actually measuring. But the body Clark keeps invoking does something specific inside his theory of the brain, and I want to go there next, because it's the strangest and most beautiful idea he's brought to this table. The brain that hallucinates the world. After the break.