Miscellaneous statistical stuff

“Spaced repetition” helps one remember facts by creating discrete flashcards which one tests oneself on at increasingly distant ‘spaced’ time periods, repeating the fact just before one probably would have forgotten it; using software to track & automate tests & review scheduling, spaced repetition can scale to hundreds of thousands of discrete items.

If spacing out facts can help one remember by repeating items just before they are forgotten, is there any use for an “anti-spaced repetition” with the opposite method of repeating items only after they are probably forgotten?

I can think of two: first, it could be used to plan consumption of media such as movies by eg. tracking one’s favorite movies of all time and scheduling a rewatch whenever one is predicted to have forgotten enough to make them novel & highly enjoyable again. Second, and more interestingly, it could be used as a serendipity generator by allowing efficient processing of notes or excerpts or old writings.

In rereading such materials many years later, one often gains a new perspective or learns something useful because one forgot something: one didn’t understand something about it at the time, or new material has radically changed one’s interpretation, and since it’d been forgotten, no use could be made of it. Unfortunately, using spaced repetition to memorize such material, while ensuring any serendipitous connections get made as soon as possible, would be radically infeasible for bulky items (a single lengthy text excerpt might correspond to hundreds of discrete items, quickly overloading even SRS systems) and for almost all items, useless. One can justify rereading old material once or perhaps twice, but not many rereads nor full memorization. But rereading haphazardly is likely to inefficiently cover some material many times while neglecting others, and such rereads will often be far too early in time (or—a lesser concern here—too late).

Instead of spaced repetition, one would instead use anti-spaced repetition: each item would be tracked and reviewed and its expected forgetting time predicted, as in spaced repetition, but instead of scheduling a review before forgetting, a review is scheduled for some time (probably long afterwards) after forgetting. The total number of reviews of each item per user lifetime would be set to a small number, perhaps 1–4, bounding the time consumption at a feasible amount.

Such an anti-spaced repetition system could be used with hundreds of thousands of notes or clippings which a person might accumulate over a lifetime, and enable them to invest a few minutes a day into reading old notes, occasionally coming up with new insights, while ensuring they don’t waste time reading notes too many times or reading notes they likely already remember & have exhausted.