Prospective memory

Why is it easier to remember an errand tied to a place you will pass than to a time you will reach?

"Post this on your way past the post box" and "post this at four o'clock" ask for the same act, from the same person, with the same warning. One gets done. The other is discovered, unposted, in a coat pocket at nine in the evening — and not because it was forgotten in any ordinary sense. Ask at eight whether there was a letter and the answer comes straight back.

So the failure is not of storage. Something was held perfectly well and simply did not present itself at the one moment it was needed. What has to happen at that moment, and why does a place supply it while a time does not?

Start with the awkward structural fact. Ordinary remembering is cued from outside: someone asks a question, and the question is itself the instruction to search. An intention has no such moment. Nobody says "now retrieve the thing you meant to do at four." The retrieval has to be started by you, from inside, while you are busy with something else entirely — and the something else is where all your attention actually is.

That gives two ways for the moment to be caught, and they are not variations on a theme. The first is to keep asking. Check the clock, notice the answer is not yet four, return to work, check again. This works, and you can watch it working: people in time-based tasks check a clock increasingly often as the target approaches, sparse in the middle stretch and dense at the end. It also has an obvious weakness. The checking runs on the same attention the task in front of you is consuming, so the deeper you fall into the work, the longer the gaps — and four o'clock arrives inside one of them.

The second way is to arrange for the world to do the asking. If the intention was tied to the post box, then the post box comes into view, and the question is now being put to you from outside, exactly as an ordinary memory question would be. Nothing had to be maintained in the interval.

But that is too generous, and the next step is where the real mechanism sits. Being present is not enough — inattentional blindness alone should warn us. What seems to matter is whether the cue falls inside what you are already processing. Einstein and McDaniel's multiprocess account draws the line here: when the cue is focal — when noticing it is part of the job you are doing anyway — retrieval can happen spontaneously, the cue calling up the associated intention without any monitoring. When the cue is incidental, sitting outside the current task, you are back to watching for it, which is the clock problem wearing a different coat.

Is that testable, or just a redescription? It makes a sharp prediction. Monitoring costs something measurable: hold a prospective intention while doing an unrelated task and you get slower at the unrelated task. If focal cues really are retrieved spontaneously, that cost should shrink toward nothing for them, while non-focal cues keep it. That is the standard experimental result, and it is the reason the two-mechanism story is preferred over a single monitoring story. It is also the observation that would kill it: were the ongoing-task cost identical for focal and non-focal cues, spontaneous retrieval would be a fiction, and everything would be vigilance all the way down.

Three refinements connect the steps.

First, this is not a claim that time-based intentions are badly stored. They are stored identically; they lack an external retrieval agent, so their fate depends on whether attention was free at the right instant. That is why the reliable fix is not effort but re-encoding: converting "at four" into "when I next leave the building", specifying a concrete when-and-where — what Gollwitzer called an implementation intention. Its published effects are respectable, but the pooled estimate leans heavily on small studies, and large field trials have generally been more modest than the meta-analytic headline; treat it as a good technique with an inflated reputation rather than a settled magnitude.

Second, an interrupted intention's habit of intruding on unrelated moments is a different phenomenon from this one. That is a maintenance effect — unmet goals keeping material accessible. Prospective memory is about the moment of retrieval, and heightened accessibility is neither necessary for it (a good focal cue works without it) nor sufficient (a nagging intention that never meets its cue still fails). They are complementary halves: accessibility keeps the thing alive; a cue collects it.

Third, and honestly: "focal" is doing heavy work and is defined partly after the fact, by whether the cue overlaps processing the task already requires. Attempts to specify it in advance have been only partly successful, and the boundary between spontaneous retrieval and very cheap monitoring is not cleanly settled.

• Why prospective memory failures rise sharply under interruption, and what that implies for handovers in safety-critical work.
• How repetition of an intention can produce the opposite error — carrying out the same act twice with no memory of the first.

Every term the collection defines is gathered in the glossary.