The Robot, the Body, and the Ghost from Berkeley

**EDO SEGAL:** There is a ghost I want to seat properly at this table, because he belongs to both of you and you should have to fight over the inheritance. Hubert Dreyfus. Berkeley philosopher, wrote What Computers Can't Do in 1972, spent his life arguing — from Heidegger, from 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 of his day despised him. Here is the delicious part. Dreyfus was attacking the symbolic AI that you, Alan, also spent your career undermining — his case against rules and representations reads, today, almost like a manifesto for the learning machines you sketched. So whose ancestor is he? Each of you, make the claim. Alan first.

**TURING:** He is mine, and I can prove it with his own argument. Dreyfus said intelligence is not rule-following — that experts cannot articulate their expertise because it was never propositional, that it lives in something more like a trained body than a logic engine, acquired by experience, resident below the level of stateable rules. That is precisely what a learned network is: competence without articulable rules, knowledge smeared across connection strengths, skill that cannot explain itself. The symbolic men he fought tried to write intelligence down as explicit rules on explicit representations, and they broke on exactly the brittleness he predicted. We vindicated him — by building on his side of the argument. The bitter joke is that he did not live to see his case win by changing which machines it was about. The "computers" he said could never do it were the symbol-shufflers. The thing that could do it was learning, pattern, the refusal of explicit rules. If Bert were here I would tell him: you were right about everything except which machines you were right about.

**SEARLE:** That is a beautiful eulogy and a selective one, and I knew Bert — he was down the hall from me at Berkeley for decades, and he would not let you keep half of him. What Dreyfus actually said was embodied. Situated. In a world that pushes back, with stakes, with a body whose skillful coping is the understanding. You kept the "no rules" half and quietly dropped the "being in the world" half — because the half you dropped is the half your systems still do not have. A network trained on text is exactly what Dreyfus would have called a degenerate case: all pattern, no situation, the articulable shadow of skill mistaken once again for the skill. He would look at your language models and say what he said about every generation's AI — that we have confused the shadow of intelligence for the thing — and then he would note, with that Berkeley relish, that this time we built the shadow out of everyone's articulations at once, which makes it a far more convincing shadow and not one inch less a shadow. He is mine, Alan. You buried him with the wrong family.

**EDO SEGAL:** Then let me press exactly there, because the engineers are calling your bluff, John. This is the Robot Reply, built in silicon. You once answered it by dropping the man inside the robot's head — let him shuffle the symbols coming from the cameras and note he still understands nothing. But the systems now genuinely act. They see and hear, they run code, they control robots that manipulate objects and learn from a world that pushes back and corrects them. When the network copes — Dreyfus's word — when it skillfully copes with a world that fails it and teaches it, do you inherit his position or abandon it?

**SEARLE:** I will give a more honest answer than my old one, because you are right that the old one was weaker than I admitted. When I dropped the man into the robot's head, I was pretending I had refuted the Robot Reply when in fact I had relocated the problem. The Robot Reply concedes my central premise — that pure symbol-shuffling, sealed off, lacks meaning — and then adds the missing ingredient: causal contact with the world. That is not a defence of strong AI. It is a roadmap for fixing it, and I treated a roadmap as a dead end. So here is where I actually stand. The cameras feed in more symbols, and symbols about symbols are still symbols — that part holds. But I have always granted that a machine with the right causal powers could, in principle, have a mind; I never ruled out silicon by fiat. What I deny is that it gets a mind by running the right program. So the question the embodied systems pose is the right one, and it is not "does it shuffle symbols from cameras." It is: has the causal organisation, closing through a real world, become the kind of thing that has a someone for whom the coping matters? My answer is still no — but it is now an empirical bet about causal powers, not a knock-down argument about syntax. I will own the difference.

**TURING:** Then we have narrowed it honestly, and I will mark how far. We both now agree that text alone is not the whole story, that closing the loop through a world changes the artifact, and that the remaining question is whether the right causal organisation in a non-biological material can host a someone. John bets no, on the ground that biology has special causal powers. I bet — not yes, but: show me what the special power is, name it, and I will believe the line is real; until then "it is the wrong stuff" is the oldest losing move in the book. And I want one thing on the record that the embodied turn gives my side. When these systems act and get corrected by a world, they are at last doing the thing Dreyfus said mattered and the thing the child-machine was always for: learning competence from consequences, in a world, the way a body does — the difference between mere [prediction and genuine construction](https://www.youonai.ai/fieldguide/med/prediction_vs_construction) collapsing in real time. That is not the costume getting better at the costume. That is the loop John spent the night saying was missing, being closed.

**SEARLE:** Closed for the function, Alan. Not yet shown to be closed for the someone. And that gap — function present, someone unproven — is the whole of the last question. The body can be built; I have stopped denying it. Whether building it lights anyone up inside is the thing neither your test nor my room can see. Which means we have arrived, finally, at the floor under every floor. Not what the machine does. Whether there is anything it is like to be it.

**EDO SEGAL:** Then we have arrived. For most of three hours I have stood between you. The deepest water is next, and it is the one place where Alan's "decide under uncertainty" and John's "the inside is the only thing that matters" finally have to touch the same nerve. Is anyone home? After this.