The Applicability of Spaced Repetition

Spaced repetition has a natural domain of applicability: information that is<br>systematically organized as an unambiguous key-value mapping with short keys and<br>values. The “Hello, world!” of flashcards is the NATO phonetic alphabet:<br>A → alpha, B → bravo, etc. Similarly, the periodic table can be thought of as<br>defining a collection of mappings: element name ↔ symbol, element name ↔ atomic<br>number, etc. You can just drill these cards and memorize the facts without a<br>prior step of understanding, or building a conceptual model.

Applying spaced repetition is trivial for this kind of information. That’s why<br>most people who use spaced repetition are either language learners or medical<br>students. In biology the main intuition you need is for “3D shapes bumping<br>around in Brownian motion”, which comes free with your human brain, and<br>afterwards it’s mostly just a lot of facts you have to memorize. Analogously<br>with language: you already have a language center, you just need to drill<br>vocabulary and grammar.

And the further you go from this domain, the harder it is to apply spaced<br>repetition.

Highly conceptual knowledge, like math, is hard to encode. You have to spend a<br>lot of time just understanding the information, and building a conceptual model<br>in your head, and then you start writing flashcards to solidify that model, like<br>taking tomographic cuts of some complex object. And coming up with questions<br>that make good flashcards (short, unambiguous, etc.) out of this highly abstract<br>knowledge is very hard. Often you have some deceptively simple fact, a simple<br>assertion, but there’s no good way to encode it as a flashcard, so you have to<br>encode “around it” by asking questions that assume or require that knowledge<br>(e.g. asking why X is true), and hoping that in drilling those, your brain will<br>remember the actual target.

In general, relational facts are easier to encode, since a binary predicate<br>like $\text{Property}(\text{Object}, \text{Value})$ readily becomes a<br>question. “Caffeine is metabolized by cytochrome 1A2”, in<br>Prolog, is $\text{Metabolism}(\text{Caffeine}, \text{CYP1A2})$, and<br>becomes “Q: What is the cytochrome that metabolizes caffeine? A: 1A2”. But how<br>do you encode stand-alone assertions like “all unitary matrices are<br>invertible”? You could encode that as a yes-or-no question, but that’s<br>useless, because rationally you can expect such questions to be biased towards<br>yes. Both “what is a property of unitary matrices?” and “what kinds of matrices<br>are invertible?” are useless because they have hundreds of possible<br>equally-valid answers, so they’re ambiguous. You have to be creative and find<br>all kinds of tricks and stratagems to encode around the knowledge.

Tangentially: this, I think, is why using AI to write flashcards is often<br>misguided. In highly systematized domains, you don’t need AI in the first place,<br>because there’s nothing for the AI to do except import a CSV into Anki. In<br>domains that are highly conceptual and abstract, you’re not memorizing a set of<br>objectively-knowable facts, you’re trying to solidify a private, internal mental<br>model that you build by reading and thinking and solving problems. You can give<br>the AI all kinds of general rules on how to write good flashcards, but the AI<br>can’t look into your mind and know which facts are salient for you, what you<br>already know, which micro-volumes of knowledge can be encoded lightly with just<br>a few flashcards, and which things need more shoring up and consequently more<br>coverage.

Can this situation be improved, or is this just an intrinsic limitation of<br>spaced repetition? I don’t know. But it seems reasonable to think some limited<br>gains are possible. I think not a lot of people are using spaced repetition on<br>these more “conceptual” domains, and (by the rule that most people in a<br>community are lurkers) even fewer of those people are writing, in detail,<br>to share their knowledge. Plenty of people have written about how<br>to write good flashcards in general, what I want to read is closer to case<br>studies where someone sits down with a text (or, even better, a textbook) and<br>describes the process by which they turned that text into flashcards, like<br>this from Michael Nielsen. From a corpus of similar case studies we<br>might derive general rules for, not how to write effective flashcards, but how<br>to encode complex, conceptual knowledge into question-answer form.

Published<br>17 May, 2026

Previous

Notes on the Hantavirus Outbreak

Next

None