Vault

Chris vs. a Deontologist — “One Should Not Torture Unicorns”

Chris runs the unicorn line (“no agents ⟹ nothing to be right or wrong about”). The deontologist answers with a conditional universal and borrows priors to apply it. The exchange slides from existence to knowledge, and the universal he keeps turns out to be about suffering, not unicorns.

Date: about 2026-10-13 (recounted by Chris on 2026-10-20, “a week back”) Source: Chris’s account in session. No transcript. Quotes below are Chris’s paraphrase, not verbatim. Participants: Chris vs. an unnamed deontologist Links: Chris vs. Mullally (origin of the unicorn line), Chris vs. Claude — Non-Uniqueness (the Anne Frank / maxim-description exchange, [19–20]), Constitutive Middle — Dialectics, Relational Objectivity (Rand’s “of value to whom and for what?”), Morality Framework, Deontology vs. Consequentialism

The arc

Chris: morals are not absolute; they need an agent and something real. Killing a human can’t be unjust if there are no humans. The deontologist rejects this outright.
Chris: “one must feed unicorns.” The deontologist calls it weak, but offers “one should not torture unicorns” as a universal.
Chris: we don’t know what a unicorn is, so we don’t know whether it can be tortured.
Deontologist, appealing to priors: we both basically know what a unicorn is (horse-like, one horn). The meanings don’t match 1:1, but there’s enough overlap to rule out dragons. We know not to torture animals, so it’s universal that we shouldn’t torture unicorns.
Chris extracts a concession: on a reasonable reading unicorns are magical, so we don’t know whether they feel pain. The deontologist dismisses it.
Chris, afterward: “what else can we do? Everything is new. Can we dismiss everything due to uncertainty?” The priors move fits his own morality is a probability game. Does it reach his claim about the non-existent?

Analysis

1. The argument drifted from existence to knowledge. Chris opened with nothing there ⟹ no moral fact. By step 3 the issue was can’t know ⟹ can’t judge. Priors answer the second claim, not the first, so the original claim was never refuted. Recall trap: don’t follow the drift. Bring it back to “nothing there.”

2. His universal is a conditional over a property. Spelled out: for any being that can suffer, don’t torture it; unicorns probably can. The duty attaches to capacity to suffer, which is a real property of real beings. Unicorns are covered only on loan, by analogy with horses. The step-5 concession admits the duty depends on that property actually being present, which is Chris’s position. A conditional over an empty class is vacuously true (“all unicorns are purple”), so granting it costs nothing. Line: “I’ll grant the conditional. It’s about suffering, not unicorns, and suffering only exists where something can suffer.” This is the cyanide parity move: “cyanide is poisonous to unicorns” has no truth value until there’s a unicorn physiology.

3. The priors move helps Chris’s framework, not his. Applying a rule by analogy is induction, and induction can be overturned. A rule that needs priors before it can be applied is not applied absolutely. It also leaves deontology’s home ground: “animals feel pain, unicorns resemble animals” is empirical reasoning, not reasoning from duty alone.

4. Uncertainty doesn’t mean anything goes. The vault’s stance (moral foundations) is that missing certainty doesn’t make every option equally good. Two supports:

The costs of error are lopsided. Sparing a being that can’t feel pain costs almost nothing. Torturing one that can costs a lot. Expected value gives his conclusion without an absolute.
Priors have failed in both directions. Sometimes a real similarity was denied: Descartes’ animal-automata, and infants operated on without anesthesia into the 1980s. Sometimes a false similarity was asserted. The error each time was a prior that never got updated, not the use of a prior. So the right response is to keep updating priors, not to discard them.

Categorical, subdivided — the question re-asked

Chris connected this to the Anne Frank case: if categorical imperatives must be subdivided this finely, when do they stop differing from conditional ones? This was asked before, as Non-Uniqueness [19–20], on 2026-10-10. The answer then:

Two axes, not one. Specificity of content (“lie to this guard”) is a different axis from source of bindingness (does the ought depend on wanting something?). Subdividing moves along the first axis and never touches the second. A rule about one act can still bind regardless of what you want.
What collapses is the test, not the distinction. Universalizability stops filtering anything once maxims can be made arbitrarily specific (the maxim-description problem). O’Neill’s reply: a maxim is your actual reason, not any true description of the act. The endpoint of unlimited subdivision is Dancy’s particularism.
Chris’s own view is functionally categorical. It is conditional on being an agent, which every possible addressee meets.

What the unicorn case adds: subdivision ends at properties. The refined rule is for anything with P, don’t do X (P = can suffer; P = is using my honesty to hunt an innocent). The rule can be categorical while whether P holds is empirical and uncertain. So:

Level	Deontologist’s claim	Chris’s claim	Who’s right
Rule form (the conditional)	universal, binding on every agent	conceded; “functionally categorical”	both
Instantiation (is there a P-bearer?)	silent	no P-bearer ⟹ no fact	Chris
Application (does this thing have P?)	“we know”	a probability judgment	Chris

Where the equivocation charge is fair: carrying the rule’s certainty over to its application. “The conditional is universal” is true. “So we know not to torture unicorns” doesn’t follow, because the second sentence borrows the certainty of the first. That slide is the thing to name, and step 4 is an example of it.

This is also Mullally’s split. “Constant principle, varying application” (Mullally, last open question) is the same table, with application named as the varying part. On this reading Chris and the refined deontologist agree about form and differ about (a) what grounds the conditional (agent structure vs. bare duty) and (b) the cases where the maxim fails even though nothing bad happens (Kant: wrong even if trust didn’t collapse, non-uniqueness [36]).

The cost to Chris, unchanged from [20]: he still owes an account of which features count as P. The unicorn case gives one answer in his own vocabulary: P is whatever the conditional’s antecedent names, and the antecedent is read from the structure of the relata. That is O’Neill’s move without Kant.

Open Questions

Is “no P-bearer ⟹ no moral fact” just the vacuous truth of the conditional restated, or a substantive claim? If it’s only the former, the deontologist can concede it and lose nothing, and the real disagreement is level (a), grounding.
Does lopsided error cost (§4) give a rule (precaution for possible sentience), and does that rule hold up against Pascal’s-mugging cases (tiny probability, enormous claimed stakes)?
Which level does the word “morality” name: rule form, or application? Most of the absolute-vs-relative fight may be a disagreement over that.