The purist's model of spaced repetition
I am the sort of spaced repetition enthusiast who likes what I'll call the "purist's model" of spaced repetition, where:
- You learn something and write a good flashcard about it;
- Your memory of it starts to degrade (if only probabilistically);
- Software guesses when you'll forget it and shows it to you before that happens;
- Human and machine keep interacting like this, efficiently and indefinitely. The system is (i) a human decaying more and more slowly, prodded less and less frequently by (ii) a software system.
This is a good model, but it has problems:
- You also encounter those facts in everyday life. (Often you're learning them precisely so that you'll evoke them in everyday life.) The system, then, contains more than the human and the human's software.
- You might encounter those facts on other cards (as I've discussed before). The action of the software system tends not to be as clean and regular as the purist's model assumes.
- The facts can change: again, the system includes more than the human and the software. Yes, there are various (real and imagined) schemes to keep cards up to date, but even if the software adjusts perfectly, the empirical-theoretical underpinnings of the purist's theory are weaker in these cases. (In practice, I just change the answer on my "What is the capital of Indonesia?" card and don't worry much about whether and how the card's history should still count.)
Even with perfect use, then (and use tends not to be perfect), the purist's model is more of a helpful ideal than a perfect description of how spaced repetition functions.