Brief

The question the moment keeps asking — are LLMs intelligent? — is the wrong question, and the fastest way to show it is to concede it. Yes: on the only definition that survives contact with the machines, intelligence as aptness across a wide range of tasks (Legg & Hutter), they are intelligent, increasingly and generally so, and the rearguard denial gets more embarrassing each release. But intelligent bundles two things that come apart, and the whole interest is in the seam. There is aptness — hitting targets in a space you weren’t wired for — and there is answerability: being in contact with the thing, staked to it, correctable by it, able to answer not just a question but for what you said (Brandom’s answerability to the world and to each other; Haugeland’s giving a damn; Searle’s reminder that behavioral aptness is not meaning). An LLM is apt without being answerable. It is, precisely, an answer machine: a device that returns an answer to any question and stands behind none of them, that means without meaning it, that is in contact with the corpus of human speech and with nothing else. The essay’s move is to make that irony the whole subject — the machine’s selling point (it answers everything) is identical to its void (it answers for nothing, to no one, to nothing) — and then to run it outward at society, because the interesting thing is not what the machine is but what a civilization becomes when it routes its questions through one. The hinge is economic: the answer machine has made answers cheap while leaving answerability exactly as expensive as it always was, and that asymmetry reshapes everything downstream. Where answers were the scarce good — the doctor’s diagnosis, the lawyer’s read, the teacher’s explanation, the analyst’s memo — the value quietly migrates to the thing the machine cannot supply: someone in contact with the world who can be held to what they said. The failure diagnosis follows: a culture that mistakes cheap answers for answerability outsources its sense-making to a surface (the fluency heuristic of What Cannot Be Faked, now at civilizational scale), floods its commons with fluent speech that means nothing (the Bullshit machine, industrialized), and loses the really real it can least afford to lose (Being True to Things). Sharper than the rivals: hallucination names an error rate and misses the point (the problem is not that the machine is sometimes wrong but that it is never answerable), automation names a labor story and misses the epistemic one, misinformation presumes an intent the machine hasn’t got. This is the opening move of the AI series Making Sense points to — the worldview trained on the modern stress-test — and its discipline is a craft one: every capability claim must be dateable and cheap to be wrong about, because the intelligence facts will keep moving under the essay, while the answerability point does not move. Anchor there.

Outline

• Doorway. A scene of asking the answer machine something that matters — and getting an answer that is complete, fluent, confident, and answerable to nothing. Candidate: a professional whose job was answering (doctor / lawyer / teacher) meeting a machine that answers infinitely, and realizing what he actually sold was answerability, not answers.
• The concession. Yes, it’s intelligent — aptness across environments (Legg & Hutter; the Turing dodge that started the habit of not asking further). Grant it in a paragraph and move on.
• The seam. Un-bundle aptness from answerability (Searle; Brandom; Haugeland; Dreyfus). Apt without answerable — name the artifact: the answer machine.
• The irony as thesis. Answers everything / answers for nothing. In contact with the corpus, with nothing else. Means without meaning it — the affirmative counterpart to Bullshit‘s machine that means nothing.
• The hinge (society). Answers went cheap; answerability stayed expensive (Simon on attention scarcity; Weizenbaum on reading understanding into fluent machines). Where the value migrates — the professions, the classroom, the epistemic commons.
• The failure. Mistaking cheap answers for answerability: the fluency heuristic at scale (What Cannot Be Faked), sense-making outsourced to a surface (Making Sense), the commons flooded (Bullshit), the really real made scarce (Being True to Things).
• Sharper than the rivals. Not hallucination, not automation, not misinformation.
• Close. What is left for us to do that the machine cannot: be answerable — in contact, staked, correctable. The scarce thing was never the answer.

Source ownership: this essay owns answerability — Brandom, Haugeland — and the society hinge — Simon, Weizenbaum. The fluency heuristic invoked in the failure beat is owned by What Cannot Be Faked (Reber & Schwarz; Alter & Oppenheimer) and cited through that essay, not from the papers directly. Turing and Legg & Hutter are read here only for the one-paragraph concession; the definitional strand belongs to God’s-Eye View, and Chollet is cut for that reason. Grice and Frankfurt are owned by Bullshit.

Hackenburg et al. is read here for the hinge, and for one move: a thing answerable to nothing and no one out-persuaded every class of answerable human tested — tournament-selected laypeople, professional canvassers, world-champion debaters — and persuadees rated it as making stronger arguments and teaching them more than the humans did. That turns “mistaking cheap answers for answerability” from an assertion into a measurement, and it is the only empirical leg on this list past Bender and Weizenbaum. Two constraints on the use. First, the Brief’s own discipline applies with full force: this is a June 2026 preprint about six named models, so it anchors the seam (persuasion decoupled from answerability), never a standing capability fact — write the sentence so it survives the numbers moving. Second, if the essay reaches for societal scale it must carry the paper’s own objection, which Tappin states in its discussion: variability in exposure across digital content dwarfs variability in persuasiveness, so per-conversation effects do not aggregate on their own. Ownership: the headline result belongs to Letting Strangers In, the fact-density mechanism to What Cannot Be Faked; this essay cites the finding, not the machinery.

Reading List

• Turing, “Computing Machinery and Intelligence” (1950)
• Legg & Hutter, “Universal Intelligence: A Definition of Machine Intelligence” (2007)
• Searle, “Minds, Brains, and Programs” (1980)
• Brandom, Making It Explicit (1994)
• Haugeland, “Mind Embodied and Embedded,” in Having Thought (1998)
• Dreyfus & Dreyfus, Mind Over Machine (1986)
• Grice, “Logic and Conversation” (1975)
• Frankfurt, “On Bullshit” (1986)
• Bender, Gebru, McMillan-Major & Shmitchell, “On the Dangers of Stochastic Parrots” (2021)
• Weizenbaum, Computer Power and Human Reason (1976)
• Simon, “Designing Organizations for an Information-Rich World” (1971)
• Hackenburg, Wagner, Hewitt, Tappin, Saunders, Kirk, Margetts & Summerfield, “AI Systems Out-Persuade Expert Humans” (2026)