Hume’s Is-Ought Problem: The Hidden Leap in Every “Should”

Twenty-eight years old, his Treatise of Human Nature nearly done, David Hume slips a complaint into the middle of a paragraph. He has been reading the moralists, and he catches a sleight of hand. The authors proceed for a while with ordinary statements joined by “is” and “is not,” and then, without warning, every proposition begins to hang on an “ought” or an “ought not.” No bridge is built across the change. Hume does not denounce the move; he asks only that the bridge be shown. Nobody, in nearly three hundred years, has shown it.

Call it Hume’s guillotine, the name the idea later acquired: a clean cut between descriptions of what is and prescriptions of what ought to be. The metaphor is right in one respect and wrong in another. It is right that something is severed. It is wrong to picture the blade as decorative, a relic kept under glass in the seminar room. The cut is a working instrument. Pick it up, and it earns its keep every time you hear the word “should” — telling you whether the speaker has earned the word or smuggled it.

What the cut actually severs

The cut does not forbid the leap. It forbids the disguise.

The point is narrow, and its narrowness is its edge. Hume is not saying values are unreal, nor that you cannot reason about them. He is saying that no purely factual premise, however dense with information, entails a conclusion containing “ought” — not by logic alone. A valid deduction cannot carry in its conclusion what is absent from its premises. So if the conclusion prescribes, a prescription already sits among the premises, however quietly. The guillotine does not forbid the leap from is to ought. It insists the leap be declared, the buried premise set on the table where it can be inspected.

G. E. Moore sharpened a neighboring blade in 1903. In Principia Ethica he argued that any attempt to define “good” as some natural property — pleasure, survival, what we desire to desire — leaves a question still open. Whatever the property, one can always ask: but is that good? If the question makes sense, and it always does, the definition has not closed. Moore called the error the naturalistic fallacy. It is not identical to Hume’s point, and the two are often blurred, but they rhyme. Both catch a quiet substitution, a fact dressed as a value and sent out to do a value’s work.

Three ways the word is earned

Most uses of “should” survive the cut effortlessly, because they were never crossing the gap. The first is the hypothetical ought, conditional on an aim already in hand. If you want the bread to rise, you should let the dough rest. No metaphysics here. The “should” is the visible face of a fact about gluten and time; remove the wanting and it evaporates. Kant named this the hypothetical imperative and set it apart precisely because it borrows all its force from a goal. The chemistry does the work; the word only points.

The second is the institutional ought. In chess you should not leave your king in check, because the rules that constitute chess say so; the “should” holds inside the game and falls silent outside it. John Searle, in a 1964 paper that set out to derive ought from is, leaned on exactly this: promising is an institution, and to promise is, by the meaning of the act, to place oneself under an obligation. Critics answered that the institution already carries an evaluative commitment — you bought the ought when you bought into promising. Which is the guillotine conceding the case it was built to win: the prescription was in the premises all along.

The third earned use is the ought of competence, the standard internal to a practice. A surgeon should tie the knot this way; a sonnet should turn at the volta; a proof should not assume what it sets out to show. Each “should” is anchored in what it is to do the thing well, and you can read the standard off the practice without consulting the cosmos. These three — the conditional, the institutional, the competent — are the honest workers. They never claim to bridge the gap, so the gap never catches them out.

What only pretends to survive

Now the pretenders. The most common is the appeal to nature: the slide from “this is how things are” to “this is how they ought to be,” the wanting tucked inside the word “natural.” Hierarchy is natural, so we should keep it. Cooperation is our evolved inheritance, so we should cooperate. The factual claim may be flawless and the conclusion still unwarranted, because a buried premise — that what evolution selected is what we owe one another — has been waved through customs unexamined. The danger was never the biology. It is the silent “therefore” bolted to its end.

A subtler pretender wears the lab coat of optimization. A model is trained to maximize engagement, and engagement rises, and the metric is offered as though the climbing number issued its own command: this is what we should build, because look, it works. But “it works” answers the question the metric was handed, not whether the metric should have been the aim. The number is an is. The choice of number was an ought, made earlier and offstage, and the guillotine sends us back to find the hand that made it. Nearly every algorithmic harm of the last decade hides a smuggled ought at the instant a loss function was named.

Where the cut cuts both ways

Here the instrument turns in the hand, and honesty requires reporting it. The same logic that exposes the smuggler also disarms the moralist who hoped to derive values from facts and so place them beyond argument. If no “is” yields an “ought,” then no description of human nature, no theorem, no ledger of consequences can summon a value into being from facts alone. The relief of holding a clean blade is also the vertigo of finding nothing beneath the values but more values. Hume saw this and did not flinch — and then, stating his own view, reached for the very word he had shown to be underivable.

“Reason is, and ought only to be the slave of the passions.”— Hume, A Treatise of Human Nature

That line is the whole problem folded into eleven words. Hume cannot state his own position without using the “ought” he has just shown nothing can supply. This is not a flaw to be patched; it is the shape of the terrain. The guillotine audits arguments; it does not furnish a place to stand. It can tell you that a given “should” has not been earned by the facts on offer. It cannot hand you the “should” you lacked. What grounds our oughts, if not facts, is the harder country the cut delivers us into — convention, sentiment, commitment, the practices we agree to be bound by — and the blade has no opinion about which of those to trust.

Holding the blade rightly

So how to wield it. Not as a sneer, the way undergraduates swing it to dismiss any moral claim as “just an opinion.” That misreads the result entirely; the cut leaves moral reasoning intact and asks only that its premises stay visible. Use it instead as a courtesy you extend and request: when someone tells you what you should do, ask gently which want, which institution, which standard of competence the “should” is leaning on. If they can name it, the talk can move to whether that ground is worth standing on. If they cannot, you have found a value pretending to be a fact — and found it with a tool older than the United States.

The deepest yield of Hume’s small complaint is not skepticism but candor. The cut does not abolish obligation; it relocates the burden of proof. Every genuine “should” rests on a prior commitment that someone, sometime, chose to make and could in principle defend. The word is not weakened by this. It is made honest. To say “you should” and to know exactly where the ought is anchored — in an aim, a practice, a promise freely given — is to use the most consequential word in any language without lying to yourself about its source. That is what survives the cut: not certainty, but accountability. Everything else only pretends.