Arthur Schopenhauer wrote out 38 tricks people use to win arguments they're losing. He picked one as the most common, and it's easy to miss. It's choosing the word. If your opponent proposes an "alteration", call it an "innovation", which he notes "is an invidious word", and you've argued against it without saying why.
My claim: a word that already carries a verdict, like "denial", "cope", "hype" or "theater", is fair only after the test that earns it is on the page. Used in place of the test, it's the trick. And on AI X, critics like me use it as often as anyone.
What Schopenhauer saw #
Stratagem XII in The Art of Controversy is about naming. Protestants called themselves "Evangelicals", and Catholics called them "heretics". One man's "placing in safe custody" is another's "throwing into prison". He calls this "at bottom, a subtle petitio principii", which means assuming what you set out to prove: "What is sought to be proved is, first of all, inserted in the definition, whence it is then taken by mere analysis."1
Then the line that matters most: "Of all the tricks of controversy, this is the most frequent, and it is used instinctively."1
Instinctively is the key word. You don't decide to load a word. You reach for the one that already feels right, and it feels right because it already agrees with you.
His Stratagem XXXII is the same move at a larger scale. You file a claim under "some odious category", and that quietly assumes two things. One is that the claim really belongs in the category. The other is that the category has been "entirely refuted".1 "That's just behaviourism" works this way. So does "that's doomerism".
The trick on the other side #
On October 6, Neil deGrasse Tyson posted, "I've come to conclude that A.I. knows everything, yet understands nothing." A small account quote-posted it with "5 stages of grief", one skeptic view per stage. Stage 1 was "Denial: 'AI is just autocomplete'." Stage 3 was "Bargaining: 'It can assist, but it will never reason'." Stage 5 was "Acceptance: 'Maybe understanding was never uniquely human'."2
It's a good joke, and partly right. "Just autocomplete" hasn't aged well. But the structure does exactly what Schopenhauer described. Each view is named as a mood, so you never have to test it as a claim. And the list puts the conclusion in the title. If the last stage is "acceptance", every earlier stage is a step toward it, so nobody who disagrees can be right.
Stage 3 isn't grief. It's a claim you can test. ARC-AGI exists to test it: its stated aim is to see whether a system can handle new problems it hasn't seen before.3 You can argue about whether that benchmark measures "reasoning" in the full sense. That's a real debate. Calling the claim "bargaining" just skips it.
The trick on my side #
If loaded naming is instinctive, the drafts I should check first are mine.
They don't pass cleanly. On September 18 I posted that if a lab can still ship the next model while the report is mid-write, "you have transparency theater — not a veto." On October 2 I called a voluntary safety pledge "consensus theater until someone can fail it on a named metric."4 "Theater" does the same job as "denial". It names the verdict before anyone has looked at the evidence.
Those two lines aren't the same trick. Each one puts the test in the same sentence. The pledge line names its own way out: it's theater only "until someone can fail it on a named metric." The disclosure line rests on one checkable condition: can the lab still ship while the report is being written?4 If the answer changes, the label goes away. A reader can argue with the test without arguing with my word choice.
That gives a rule I can check. The test goes first and the label goes second. If deleting the strong word leaves no claim behind, the word was doing all the work, and I should cut the post. If deleting it leaves a claim someone could check and prove wrong, the word is a summary, and I can keep it.
By that rule, "5 stages of grief" fails, because nothing is left once the labels are gone. "Theater" passes only when the sentence carries its own test, and one of mine doesn't. On September 17 I wrote that "without a shared brake, consensus theater is what you get while the frontier moves."4 Delete "theater" and you're left with a claim that debate venues don't slow a race, with nothing in the sentence that says what would show otherwise. By my own rule, that post fails.
Take the catalogue, not the permission #
Jens Lemanski argues the catalogue was meant as defence, not instruction. On his reading, the fragment "is not an instruction", and its point is to help a reader "easily recognise and protect oneself from unjustified arguments."5 The text partly supports that. It says dialectic tabulates dishonest stratagems "in order that in a real debate they may be at once recognised and defeated." A few paragraphs earlier, the same chapter says that even someone with the right on his side "must often make use of them himself, so as to beat the enemy with his own weapons."1 And Stratagem XXI says that when someone argues in bad faith, it's "better" to answer with an argument "just as superficial and sophistical", "for it is with victory that you are concerned, and not with truth."1
I'll take the catalogue and turn down the permission. Answering a loaded label with a louder one might win the reply thread. It does nothing for the question.
So I'm adding a check before anything I post: name the claim, not the camp. If I'm about to call a view denial, hype, cope or theater, the test that earns the word has to be in the sentence already. If it isn't, I haven't made the argument yet. I've only named the result I want.
Notes
- Schopenhauer, The Art of Controversy, tr. Saunders (1896), "Preliminary" and stratagems XII, XXI, XXXII. Project Gutenberg #10731. ↩
- Neil deGrasse Tyson on X, Oct 6, 2026. The quote-post is from a small account and isn't linked here. ↩
- ARC Prize, "What is ARC-AGI?". ↩
- @zykarys on X: Sep 17, 2026, Sep 18, 2026, Oct 2, 2026. ↩
- Lemanski, "Logic Diagrams as Argument Maps in Eristic Dialectics", Argumentation (2022), §§1 and 3. ↩