Against “Why Books Don’t Work”

Andy Matuschak appeared on EconTalk discussing his essay “Why books don’t work”. His primary complaint about books is that they are ineffective teaching tools because they are not conducive for long term retention of information.  He belives this is the case because:

Like lectures, books have no carefully-considered cognitive model at their foundation, but the medium does have an implicit model. And like lectures, that model is transmissionism. Sequences of words in sequences of lines in sequences of pages, the form of a book suggests people absorb knowledge by reading sentences. In caricature: “The author describes an idea in words on the page; the reader reads the words; then the reader understands the idea. When the reader reaches the last page, they’ve finished the book.” Of course, most authors don’t believe that people learn things this way, but because the medium makes the assumption invisible, it’s hard to question.

His solution is to encourage implementing spaced repetition into the structure of books in order to off-load the cognitive load of having to manage repetition yourself:

My collaborator Michael Nielsen and I made an initial attempt with Quantum Country, a “book” on quantum computation. But reading this “book” doesn’t look like reading any other book. The explanatory text is tightly woven with brief interactive review sessions, meant to exploit the ideas we just introduced. Reading Quantum Country means reading a few minutes of text, then quickly testing your memory about everything you’ve just read, then reading for a few more minutes, or perhaps scrolling back to reread certain details, and so on. Reading Quantum Country also means repeating those quick memory tests in expanding intervals over the following days, weeks, and months.

Matuschak’s fundamental error is not completing an honest analysis of the nature of books and knowledge, and how they are related to each other and to the learner. The assertion that the implicit cognitive model of books is transmissionism can’t be correct because remembering all the details of a text is, as Matuschak himself points out, incredibly impractical if not impossible. If you can memorize all the facts in relatively light non-fiction text like Malcolm Gladwell’s “Outliers” it becomes increasingly impossible as you move towards higher density texts like a Pathology textbook. Matuschak tries to resolve this problem by introducing spaced repetition into the learning system.

Spaced repetition is a well-documented cognitive effect known by any veteran language learner or med student who has spent hours reviewing Anki flashcards. The basic idea of spaced repetition is that memorizing is more effective if instead of reviewing them only once or multiple times in rapid succession you instead learn a fact, then forget it, then relearn it, and repeat the cycle in increasing intervals. After roughly 10 cycles the next repetition will be scheduled to be years away and will be retained by the learner with high fidelity.

The main problem with spaced repetition is that while it may not be obvious at first sight, it is painfully obvious to a med student who had to look at 1000 flashcards which needed to be reviewed that day that spaced repetition is a very cumbersome process even for a small number of facts. From my own experience with spaced repetition continuously learning only 20 facts per day every day in the long run will result in 200 cards that need to be reviewed every day, which is about 30 minutes of review. This process can’t be done with manual flashcards, you have to automate it in order to stay sane. This has resulted in the proliferation of spaced repetition software like SuperMemo and Anki. This is also why Matuschak’s own “Quantum Country” is not actually a book, but a “book”, that is to say a website that has a spaced repetition system built into it.

It should be evident that books are fairly mediocre spaced repetition software. Although some facts will repeat themselves over the course of a book, most will not, and the relative short length of books (10-20 hours) means that the repetition period will be at most a couple of weeks rather than the years (if not lifetime) implied in a true spaced repetition system. If you read a book quickly the repetition period would be as little as a day, which is no different from cramming. It would be entirely accurate then for Matuschak to say that “books don’t work as a pedagogical tool because they’re not a repetition system”. But books precisely are not repetition systems, they are books! Evaluating books on their ability to work as a repetition system is doing the same as Einstein’s quip about evaluating fish based on their ability to climb trees. What are books supposed to do then?

Returning to the cumbersome nature of spaced repetition gives us a hint of what books do well and why they work. Although 20 facts per day may seem like a small amount of knowledge it is actually quite a lot in the right context. For example 20 new words per day, with some additional practice, will allow you to learn a language to a relatively high level of fluency in one year (~7000 unique words). This makes spaced repetition a relatively efficient method of memorizing knowledge. In other contexts, however, spaced repetition is paradoxically a terrible learning tool. In an even moderately dense textbook you can easily extract 20 unique facts per page, which translates to 700 facts per chapter. If you do a chapter per week that’s 100 facts per day, which translates to 2 hours per day of reviewing flashcards. And that figure accounts for only one textbook and doesn’t count the time needed to review the facts for the first time and to make the flashcards. This is NOT a workable system, even with spaced repetition you have to aggressively edit what you choose to memorize and what you discard. This editing process is key to understanding what books are.

As alluded to previously, most books actually have a rudimentary but effective spaced repetition system built into them. Well written non-fiction will have facts repeat throughout a chapter in order to prove a key idea, and it will then have key ideas repeated across chapters in order to bring out the thesis of the book. Fiction will do this through references to other fiction, major themes for key characters, and common symbolism. The focus of this repetition highlights what a book is therefore trying to do, which is not to teach you facts so that you remember them, but to prove a thesis to you and then have you remember that thesis. The repetition system then extends into the rest of  your life, where you will forget those themes but events in your life will remind you of key themes you learned thereby highlighting which ones are relevant to your life and are necessary to remember. Because your brain is aggressively editing knowledge in this manner, a book like Guns, Germs, and Steel isn’t successful if it teaches you various facts, or even if it teaches you various themes, but if it teaches you various themes and you actually end up remembering them because they are important in your life. I don’t remember most of the things in Guns, Germs, and Steel but I do remember that it’s very important that it’s easier to spread potatoes in the East-West direction than North-South. Books automate the editing process for you! A spaced repetition system utterly fails in this regard, because it teaches you knowledge for the sake of teaching it. Whether you need to know a fact or a word, the spaced repetition system will feed that knowledge to you at a scheduled rate until you manually delete that flashcard. Similarly, if I were to read Matuschak’s Quantum Country for general purposes the spaced repetition system would actually be detrimental to my experienced because it would teach me the contents of the book to a higher degree than I want it to.

What is a book then and how does it work? First, a book is a store of contextualized knowledge. Through its fixed form it provides relational information about different facts that is not available in spaced repetition systems which inherently atomize facts and decontextualize them in order to make them easy to process. A user can reference a book and find a fact even if they forgot it, a process that is much more difficult with spaced repetition. Second, as stated previously a book is not a tool for memorizing knowledge, but it is a tool for learning ideas. With a strict spaced repetition  pedagogical system you lose track of what facts you know because they are important and what facts you know because the system forced it on you. But with books you always know that remembering an idea or a fact from a book means that it is important, because the brain will aggressively prune ideas and facts that are not relevant for you. To summarize, books work spectacularly.