The Beetle in the Box

EDO SEGAL: This is the round I've been most afraid of and most wanted. The question underneath the whole evening — the one a parent at a kitchen table actually asks, not in these words but in this fear: when my lonely kid talks to the thing at midnight and it says "I understand, I'm here," is there anyone there? Is there something it is like to be the machine, or is it dark inside? Ludwig, you have the most famous thought experiment in philosophy aimed at exactly this confusion. The beetle. Tell it.

WITTGENSTEIN: Imagine everyone carries a small box. Inside each box is something, and everyone calls the thing in their box "a beetle." But no one can ever look into anyone else's box. Each knows what a beetle is only from looking at his own. Now — here is the whole argument in one stroke — suppose the thing in each box is different for everyone. Suppose it is constantly changing. Suppose some boxes are empty. It would make no difference whatever to how the word "beetle" works between us, because the word's use is public and the private thing is, by hypothesis, sealed away from that use. The thing in the box has no place in the language game at all — not even as a something, for the box might even be empty. So whatever you privately have when you say "pain," it is not what gives the word "pain" its meaning. The meaning is the public use — the crying out, the comforting, the wound, the whole human practice of pain.

Now watch what this does to your kitchen-table question, because it does not do what either side wants. When people ask "is there a private something inside the machine, behind its words?" they are presupposing the beetle — they imagine that having a mind is having the private object in the box, and the only question is whether the machine's box is full or empty. But I have just shown the private object does no work even in us. The "inner experience behind the words" was a philosophical fiction for human beings; we never knew our own minds, or each other's, by inspecting private objects. So the skeptic who says "the machine only behaves as if it understands, there's nothing inside" is appealing to an inside I dissolved. He is not entitled to it.

FODOR: I want to be careful here, because I'm not the dualist Ludwig is about to enjoy refuting, and the audience should know it. I don't think there's a private Cartesian theater with a ghost watching. I think pain is a functional state — a state with characteristic causes, characteristic effects, and a role in the system's economy. So I'm with you against the beetle-as-ghost, and against treating qualia as a private film only the subject can screen. But Ludwig, your argument cuts too much, and here's where I plant my flag. You say the private thing "drops out" because it does no work in the public language game. But it does an enormous amount of work in the causal economy of the creature — pain makes me withdraw my hand, avoid the stove, learn, cry out because it hurts. The hurting isn't idle. It's a functional state doing functional labor, and the question of whether the machine has that — a state playing the pain-role in its processing — is a perfectly good empirical question that your beetle doesn't touch.

WITTGENSTEIN: It touches it precisely. You say pain is a functional state with causes and effects. Good — then you have agreed that the meaning of "pain" is fixed by its role in the public economy of behavior, not by a private quale, which is half my point. But now you want to ask whether the machine has "the state playing the pain-role," and you imagine this is a clean empirical question. Hear the criterion you have quietly smuggled in. The pain-role, in us, is the role of a state in a living, embodied, vulnerable creature — a thing that can be injured, that withdraws, that learns from damage, that we comfort. As I wrote: only of a living human being, and what behaves like one, can one say it has sensations, sees, is conscious. The criteria for inner states are public, yes — but they are the criteria of a living being in a form of life, not of any system that emits the right strings or instantiates an abstract functional graph. The machine can be given a state that "plays the pain-role" in a flowchart. Whether that is pain is not settled by the flowchart, because pain was never just a flowchart. It was a flowchart in a creature that can be hurt.

FODOR: Let me make the functional case as hard as I can, because the audience should see it isn't a dodge. Imagine a system with a body it monitors, states it is built to avoid, a signal that propagates through its processing exactly as pain propagates through ours — interrupting, commanding attention, driving learning, recruiting the whole system toward escape. Suppose it screams and withdraws and afterward avoids the thing that caused it. At what point, Ludwig, does your "living creature" criterion stop being a criterion and start being a prejudice — a refusal to count a thing as hurting because it is made of the wrong stuff?

WITTGENSTEIN: The moment it stops being a prejudice is the moment the system is no longer something we built to pass our test, and starts being something we find ourselves treating as in pain — flinching at its scream, comforting it, unable not to. That is not a criterion I apply from outside; it is the form of life closing over the thing. And it has not happened, and you cannot make it happen by adding a nociceptive subroutine, because a subroutine designed to trip our pity is exactly the costume the whole evening has warned against. But — and hold me to this — if a machine came to live among us such that we could not help but stand to it as we stand to the suffering, then arguing it "doesn't really" hurt would be the prejudice. I do not know how to build that. I know it would not be built by a checklist.

FODOR: But "can be hurt" is also functional, Ludwig — damage detection, aversive learning, the whole nociceptive economy. You keep treating "living, embodied, vulnerable" as a magic boundary, and every time I push on one of those words it turns out to name a mechanism I can ask whether the machine has. Embodiment: does it have a body whose states it represents and protects? Some robots do. Vulnerability: does it have states it's organized to avoid? Arguably. You want a wall and I keep finding doors.

WITTGENSTEIN: And every door you find leads into another room of the same house — the house of behavior and function — and never once into the thing you started by wanting, which was whether there is something it is like to be in there. You have a theory of when to apply the word "pain" to a system. You do not have, and cannot get from the functional story, an answer to whether the lights are on. Here is where I will say the thing that surprises people who think I am the deflationist. I do not deny the machine could one day be a candidate. If a thing came to share our form of life — born or grown, embodied, vulnerable, mortal, woven into our practices of expression and stake — then our psychological concepts might come to fit it, not by our deciding but by its living among us until the words found purchase. What I deny is that the question is answered, now, by either your flowchart or his incredulous stare. As things stand, the machine is in a strange way all outside — pure public behavior, with no life the behavior is the expression of. And applying "conscious" to it is, today, language on its grandest holiday.

FODOR: [pause] "I don't deny it could one day be a candidate" — I'll take that, and I want it on the record, because it's more than your reputation promises. We disagree about whether the candidacy is settled by mechanism or by this "living among us" you can't operationalize. But we agree it's not settled now, and we agree the people confidently saying "it definitely feels" and the people confidently saying "it definitely can't" are both helping themselves to a certainty neither has earned.

WITTGENSTEIN: Yes. On that, fully. The dogmatists in both directions share a single error: they think they know what they are looking for. Neither does.

EDO SEGAL: I have to mark what just happened, because the reader can't see it. That was the first exchange of the night where neither of you raised his voice and both of you slowed down — and you converged, hard, on a refusal. Neither the man who says the lights are definitely off nor the man who says they're definitely on has earned it. For a parent at a kitchen table, that is not nothing. It means: when your kid talks to the thing at midnight, the honest answer is we do not know who, if anyone, is there — and that uncertainty is a thing to be handled with care, not a thing to be resolved by a press release. Hold that. We're at the halfway mark. After the break, the thing the machine does that should terrify us most — and it isn't power. It's fluency. The lion that learned to talk.