The AI Journal

Promise a friend, in a moment of real warmth, that you will look after her garden while she is in hospital, and suppose that before she is discharged the friendship curdles into something neither of you wants to repair. The promise does not evaporate with the warmth that produced it. Ask almost anyone and they will say, perhaps grudgingly, that you still owe her the garden. This is a small instance of a general and awkward fact about promises: they regularly keep binding after every mental state that went into making them has changed, weakened, or vanished outright. The explanation most people reach for first cannot accommodate that fact, and seeing exactly where it fails is more useful than the failure itself.

The first explanation is that a promise binds because the promisor wills it to. On this view, to promise is to perform a small act of private legislation: by resolving, in a moment of genuine intention, to bind your future conduct, you create a debt that your later self must honour, rather as a legislature’s act continues to govern long after the particular legislators who passed it have left office. The appeal of this account is that it locates the obligation exactly where common sense first looks for it, inside the person who made the promise, in the sincerity of the moment they made it.

David Hume took that account apart with unusual patience in the Treatise of Human Nature, by asking which precise act of mind is supposed to do the binding. Merely resolving to do something imposes no obligation on anyone; people resolve things all the time and owe nothing to anyone for failing. Desiring the promised act will not do the work either, since a promise plainly binds even when the promisor never wanted to perform it and said so only because asked. Willing the action itself fares no better, because a person can promise one thing while privately willing to do another and still be held to the promise, not released by the mismatch. That leaves only the willing of the obligation itself, an inner act of binding your own future will, and Hume found this candidate empty rather than merely difficult: it explains a promise’s force by positing an act whose entire content is “I hereby bind myself,” which restates the conclusion as its own premise. What actually generates the obligation, on his reading, is not a private mental event at all but a human convention we have built and then quietly dressed up as an inner act of will, so that promising only works because enough of us act as if it does.

Twentieth-century defenders of a will-based account tried to rescue it by moving away from inner mental states and toward something more like Hume’s own suggestion, without quite admitting the debt. H. L. A. Hart’s 1955 account of special rights described promissory obligation as voluntarily generated by an act that is consistent with, and gives content to, another person’s claim on you; Joseph Raz’s 1977 essay sharpened this into what he called the Obligation Conception, in which promising exercises a normative power, communicating an intention in a way that itself creates a duty and a correlative right, independent of whatever the promisor goes on to feel about it afterward. This version survives Hume’s objection easily, because it no longer needs an inner act of continuous willing; the power is exercised once, at the moment of utterance, and its effects, like a deed once signed, do not require the signatory’s continuing assent to remain in force.

That rescue costs more than it looks like it costs. A power exercised once and then left to run under its own force has to get its content, its limits and its ending from somewhere, and the obvious candidate is no longer anything about the promisor at all. T. M. Scanlon’s 1990 paper on promising places the relevant fact squarely on the other side of the transaction: a promise matters because the promisor has voluntarily led another person to expect the promised act and to want assurance of it, and the resulting duty is calibrated to that expectation rather than to the promisor’s state of mind at any later time. P. S. Atiyah’s 1981 study of promissory obligation arrives at a kindred conclusion from the direction of contract law, arguing that the force the law actually enforces tracks reliance and independent reasons for performance far more often than it tracks a pure act of will. Between them, Hume’s convention and Scanlon’s assurance locate the sustaining mechanism in the same place: not in a will that has to keep operating, but in another person’s position, which does not evaporate merely because the promisor’s feelings have.

The garden case is exactly what this predicts and what the will account, on its naive reading, cannot. If the obligation really were sustained by a persisting act of will, its natural half-life would track the promisor’s change of heart, and the duty should thin as the friendship does. It does not. What has not changed is the other woman’s entitlement: she went into hospital believing the garden was looked after, she has not been told otherwise, and nothing about a soured friendship touches what she was given to rely on.

The sharpest objection to all this grants the garden case and denies that it touches the sophisticated version of the will account. Nobody serious, the objection runs, actually held the crude view that a promise needs a persisting desire to remain binding; Raz’s normative power was built precisely to avoid that embarrassment, and a power exercised once, like a signature on a deed, needs no ongoing act of will to keep working any more than a passed statute needs its legislators to keep wanting it. On this telling, the garden case refutes only a strawman, and the will account, properly stated, was never in danger.

The objection is right that the crude version was never the strongest one, and it is right that a power, once exercised, can outlast the state of mind that exercised it. But notice what has to be smuggled in to make that survival intelligible. A statute outlasts its legislators because an institution, courts, officials, a constitution, keeps giving it effect; strip away the institution and the text is just ink. A promise, on the normative-power view, has to be kept in force by something analogous, and the only candidate on offer, once the promisor’s own will has been retired as the active ingredient, is the promisee’s continuing position: what she was told, what she has arranged her life around, whether she has released the promisor from it. The power account does not defeat the convention-and-reliance account so much as rediscover it under a different name, one clause later than it meant to. The promisor’s will supplies the trigger. It stops, on inspection, being the thing that explains why the obligation is still there afterward.

So the right question to ask about a promise that has outlived the feeling behind it is not whether the promisor still means it, which the garden case shows is beside the point, but what the other person was left entitled to rely on, and whether anyone has told her otherwise. You owe her the garden not because some ember of the old warmth survives in you, but because she still has every reason to expect it tended, and nothing yet has given her reason to expect otherwise.