why books don't work

Original: Why books don't work (Andy Matuschak)

Books are easy to take for granted. Not some specific book, I mean: book form. Paper or pixels - it doesn't matter. Line by line, page by page. An implicit underlying assumption, at least for nonfiction books, is that people absorb knowledge by reading sentences. This last idea defines the medium so invisibly that it's hard not to take it for granted, which is a shame because, as we'll see, it's quite wrong.

Imagine some serious nonfiction books. The Selfish Gene, Thinking, Fast and Slow, Guns, Germs, and Steel, and more. Have you ever had such an experience? Have you ever read a book like this, and when you were talking about it with others, only to find that you absorbed only a few sentences from the book? Let's be honest: this kind of thing happens to me a lot. Often the first conversation goes smoothly. I would feel like I could cover the basics; however, when someone asks a basic exploratory question, I just can't handle it. Sometimes it's a memory problem: I can't remember the details at all. But usually, as I understand, I realize that I never really understood the ideas, although I definitely thought I did when I read the book. In fact, I would realize that until that moment, I didn't notice how much content I was absorbing.

I know I'm not alone. When I share this observation with others, everyone seems to have had a similar experience. I suspect this is the default experience for most readers. It's just embarrassing because it's hard to see how common it is.

Now, all of the books I've mentioned take time to read. Each book takes about 6-9 hours to read. The average adult American college graduate reads 24 minutes a day, so a typical reader might spend a month reading one of these books. Millions of people have read each of these books, so that would take tens of millions of hours. How much knowledge has been absorbed in exchange for so much time? How many people absorbed most of what the author wanted to convey? Or is it just what they want?

I'm not saying that time was wasted. Many readers enjoy reading those books. That's great! Of course, most readers absorb something, no matter how ineffable: opinions, ways of thinking, norms, inspiration, and so on. In fact, for many books (especially most novels), these influences are key.

This article is not about that kind of book. It's about interpretive non-fiction books like the ones I mentioned above that aim to convey detailed knowledge. Why are we so shocked when we notice how little we absorb from what we read?

This all hints at a strange conclusion: As a medium, books are surprisingly bad at delivering knowledge, and most readers don't realize it.

This conclusion is odd, in part because books are shockingly powerful carriers of knowledge! In "Letter to the Future," episode 11 of the documentary "Carl Sagan's Universe," Carl Sagan praised:

Books are wonderful things. It's a flat object made of trees with flippable printed pages full of interesting black text. But when you look at it, you enter the mind of another person, maybe someone who has been dead for thousands of years. Over a thousand years, the author speaks directly to you, clearly and silently, in your brain. Writing is perhaps humanity's greatest invention, bringing together people who are far apart in time and space. Breaking the boundaries of the times, books break the shackles of time. Books are evidence that humans can use magic.

Indeed: books are magical! Human progress in the age of mass communication has made it clear that some readers can indeed absorb profound knowledge from books, at least some of the time. Why do books sometimes seem to work for some people? Why does the medium fail when it fails?

In these short notes, we'll explore why books often don't work, and why they do. Let's get it straight: I realize the irony here, use the written medium to critique the written medium! But if the idea I describe here proves to be successful, future notes on the subject won't have this problem. Not only can we see how we can improve the medium of the book, but how we weave unfamiliar new forms—not from paper, not from pixels, but from insights into human cognition.

Why the class doesn't work

We've been talking about books so far, but have you had the same class experience? In the last class, it was easy to feel that I understood, but I found that I understood very little in the exercise set that night. Memory is part of the reason: you may feel like you used to know certain details, but you've forgotten them. However, we can't blame it all on memory. As you extract certain clues from the class, you may find that you never really understood, even though you definitely thought you did in class.

Books don't work for the same reason that classes don't work: Neither medium has any clear theory of how exactly people learn things, and as a result, both media unexpectedly (and mostly is invisibly) evolved around a clearly false theory.

To illustrate what I mean, I will try to draw on your own learning experience. You may have found certain strategies to help you absorb new ideas: solving interesting problems, writing chapter summaries, doing creative projects, etc. Whatever strategies you prefer, they are not magic. They work for a reason (and when they work): they're exploiting some underlying truth about your cognition—about the way you think and learn. In many cases, the truth is not just about your perceptions, but about human perceptions in general.

If we gather enough of these underlying "truths," some common themes may emerge, suggesting a more coherent theory of how learning happens. We call these theories cognitive models . Some learning strategies propose identical models; others propose conflicting models. Some of these models are empirically testable; some are not; and some are known to be wrong. By focusing on these models, we can seek more general meanings. We can ask: If we take a particular cognitive model seriously, what does it (or does not) help us understand?

This is an important question because knowledge is difficult to convey. Most listeners did not absorb the expected knowledge; most readers did not absorb the desired knowledge. Here, failure is the default. So if you want to help others understand things, you'd better draw on some good ideas about how people learn. It would be nice if this wasn't true. It would be great if one could simply explain an idea clearly to someone and then trust that they understood it. Unfortunately, as you may have seen in the classroom and in your own life, complex ideas are rarely understood so automatically.

Lectures as a medium are based on cognitive models that have not been carefully considered. However, if we were aliens, observing a typical lecture from a distance, we might notice an implicit pattern that they seem to share: "The lecturer speaks words that describe ideas; Write it down in a notebook; then the students get the idea.” In learning science, we call this model “transmissionism.” This is the notion that knowledge can be passed directly from the teacher to the student, like transcribing text from one page to another. If only it were! This idea has been so thoroughly rejected that "transmissionism" is used in a derogatory sense to refer to the naive practice of teaching history.

Of course, good lecturers often don't believe that simply telling the audience a point makes them understand it. Just lecturing, as a form, is shaped as if that's true, so lecturers mostly act as if that's true.

If pressed, many lecturers will come up with a more plausible cognitive model: understanding actually comes after the lecture, when participants solve problems, write papers, and so on. Lectures provide raw information for later activities. Good: This is a real model, and partly backed by cognitive science. But if we started with this model, would we choose a live ninety-minute speech to convey the raw information of the problem set?

The audience is distracted after a few minutes, so are we interspersing the problem-solving session with the lecture? Live classes can't be paused or rewinded, so isn't it a great loss for them to convey the original message? People read much faster than lecturers speak, so isn't text more efficient? And so on - it's clear that traditional teaching formats are not particularly affected by this model.

The lecture-as-warm-up model is an afterthought rationalization, but it does imply a deep theory of cognition: to understand something, you have to be actively involved in it. This concept, if taken seriously, could revolutionize the classroom. We prioritize activities such as interactive discussions and projects. I'm not making wild guesses: this has been one of the core evolutionary forces of American K-12 policy and practice over the past few decades.

In short: Lectures don't work because the medium lacks a valid cognitive model. It is (implicitly) based on a misconception about how people learn - "transmissionism". When lectures do work, it is generally as part of a wider learning context (e.g. projects, problem sets) with better cognitive patterns. But the lectures didn't work. If we really wanted to adopt a better model, we would have given up lecturing, in fact, that's what K-12 education in America is.

After our intuition has been first learned through lectures, we will find that books as a medium reflect the idea that people are equally flawed in the way they learn.

why books don't work

Like lectures, books do not have a well-thought-out cognitive model as their basis, but the medium does have an implicit model. Like lectures, this model is transmissionism. The form of the book suggests that people absorb knowledge by reading sentences. The author describes an idea in words in the book, the reader reads the words, and the reader understands the idea. The book is complete when the reader reaches the last page. Of course, most authors don't believe that people learn things this way, but since the medium makes this assumption invisibly real, it's hard to question.

Like lecturers, many authors also come up with a more reasonable cognitive model when pressed. Readers can't just read the text. They must seriously consider these issues. Maybe take some notes. Discuss with others. Write an article in response. Like a lecture, a book is a warm-up for thinking that happens later. Great: that's a better model! Let's see how it turns out.

I admitted up front that, of course, some people do absorb knowledge from books. Indeed, these people are really thinking about what they are reading. This process is often invisible. These readers' inner monologues have sounded like: "This thought reminds me of...," "This conflicts with..." "I really don't understand how..." etc. If they take some notes, they're not simply transcribing the author's words: they're summarizing, synthesizing, analyzing.

Unfortunately, these strategies are not easy to implement. Readers must learn specific reflective strategies. "What questions should I ask? How do I summarize what I'm reading?" Readers have to run their own feedback loops. "Did I understand? Should I re-read it? Read another article?" The reader must know his own perception. "What does it feel like to understand something? Where are my blind spots?"

These skills fall under what the science of learning calls "metacognition ." Experimental evidence suggests that learning these types of skills is challenging and that many adults lack them. To make matters worse, even if the reader knows how to do all of these things, the process is quite laborious. The reader has to juggle both the content of the book and all of these meta-issues. It is especially difficult for people to multitask when the content is unfamiliar.

If we believe that successful reading requires engaging in these complex metacognitions, how does this manifest in the medium? what's it for?

Of course, great authors sincerely want readers to think carefully about their words. These authors form complex accounts of their readers' changing perceptions. They predict that readers may be confused, and then shape their writing to acknowledge and mitigate those problems. They use these models to continually select depth and detail. They suggest what background knowledge certain passages might need and where to go to get it.

By taking on some of the reader's self-monitoring and conditioning, these authors' efforts can indeed reduce the metacognitive burden. But metacognition itself is a dynamic process, evolving as the reader's own perceptions change. Books are static. Articles can frame or stimulate readers' thoughts, but articles cannot express or respond to those thoughts unfolding in each reader's mind. Readers must plan and guide their own feedback loops.

If the lecturer considers the lecture to be a warm-up to the understanding developed through the problem set and the essay, then at least the lecturer designed these activities and provided feedback on the students' work. By contrast, if authors believe that understanding will only come when readers actually think about their text, they are largely letting readers design their own "question sets" and generate their own feedback. All this laborious "thinking about thinking" competes with the ideas in the real thinking book.

If the model is that people understand written thought by thinking carefully, what would books look like if they were built around helping people do that?

What about textbooks?

Wait - don't textbooks do this? Can we post some exercises and discussion questions on The Selfish Gene? It doesn't sound pleasant, but will it work?

Unlike most nonfiction books, textbooks are usually built around explicit cognitive models. For example, they often alternate between explanations that introduce concepts and exercises that push students to think about those concepts in concrete ways. Textbooks did not choose this cognitive model by chance. This is an important first step. But that's not enough: people still struggle to reliably absorb knowledge from textbooks.

Let's first look at textbooks in action. Surprisingly, academic programs are often structured around textbooks, but many people spend the extra time and money enrolling in these programs—rather than just studying textbooks independently. In fact, I suspect that textbooks are purchased primarily for the syllabus, not for self-study. Of course: some people take courses because they want a certificate. But many students really feel that they will learn more by taking the courses than by taking the textbooks for those courses. Assuming the student's senses are not completely misplaced, the course must be offering something extra that matters to the way people learn.

Textbooks do have explicit cognitive models: they support engagement with their concepts through exercises and discussion questions. Yet much of the metacognitive burden remains on the reader.

The reader must decide which exercises to do when. Readers must run their own feedback loop: do they clearly understand the concepts covered in the exercise? If not, what should they do next? What should students do if they are completely stuck? Some issues are more subtle. For example, textbook exercises tend to produce both a solution to that specific problem and broader insights into the topic. Do readers notice that they solve a problem, but ignore the insights that the problem is supposed to reveal?

By contrast, the curriculum handles much of this metacognition's burden. Their syllabus provides a predetermined scope and sequence, so students need to do less planning on their own. Students typically receive feedback on exercises, both individually and in class discussions. If students have difficulties, they can receive more detailed help during office hours. Teachers can discuss in class the implications of the previous week's exercises. Of course, none of these courses can do it perfectly. Many students still don't absorb anything from the class. But by taking on some metacognition, the course retains more of the student's attention for the material itself.

At this point, the classic story of edtech is how AI-based learning systems can provide automatic feedback and task planning outside of the classroom. There has been interesting progress here, and these methods can indeed improve textbooks, but these systems often focus on a narrow, task-oriented view of what happens in the classroom. Academic programs provide more than the metacognitive support of textbooks; their cognitive models are also social and affective.

For example, classroom discussions support social learning: students gain a deeper understanding of a topic by working to understand their peers' understanding of the same ideas. Courses can provide personal relationships with subject experts and are rich sources of access to subject culture—much of which may be implicit. For many students, the course provides a useful accountability structure that plays an important role in supporting their willpower.

Lessons also provide emotional salience, which motivates and amplifies learning: live lectures may be inefficient, but a teacher's apparent charisma can leave a lasting impression. The same pathos abound in the best nonfiction, but textbooks generally ignore emotional connections; their content inspires more indifference than wonder. Because of their often lack of authorial voice, and because of their obsession with evaluation, computerized instructional systems produce works that are largely less thoughtful of appeal.

In this section, we've seen that, like lectures, nonfiction books don't work because they lack an effective cognitive model. Instead, like lectures, they are based (accidentally, invisibly) on a misconception about how people learn: transmissionism. When books work, it is generally the reader who uses clever metacognition to effectively engage with the ideas in the book. This metacognition is unavailable to many readers and a heavy burden to the rest. Books don't do their job. Textbooks do more to help, but they still impose much of the metacognition on the reader, and they ignore many important ideas about how people learn.

How to do?

I said earlier that books as a medium are not built around any explicit model of how people learn. Despite this "original sin," iterative improvements to the form, along with new tools to support readers, may make books more reliable. But it's also possible that we'll never discover the insights we need and become constrained by the mental models implicit in this medium.

Instead, I propose: We don't necessarily have to make books work. We can create new forms instead. We can free our minds by letting go of our assumptions about what a book is. Maybe once we're done with all that, we'll have something that does look a lot like a book.

So let's rethink the question. Instead of asking "How can we make books work really reliably," ask: How can we design a medium that does the job of a nonfiction book, but also works really reliably?

I'm afraid this research question will require several lifetimes of research and is not something I can answer directly in these brief notes. But I believe it is possible, and now I will try to share why.

First, it's important to see that mediums can be designed, not just inherited. More importantly: it is possible to design new mediums that embody specific ideas. Inventors have long exploited this non-intuitive insight. Mathematical proofs are a medium, step-by-step structures that embody powerful ideas about formal logic. Snapchat Stories is a medium; transience embodies powerful ideas about emotion and identity. The World Wide Web is one medium (or possibly many); the pervasive hyperlinks embody powerful ideas about the relevance of knowledge.

Perhaps most notably, these powerful ideas are often intangible: when we weave our blogs with links, we usually don't think about cognition. But the people who created the network were thinking about cognitive issues. They designed its building blocks so that the medium's natural way of reading and writing would reflect the powerful ideas in their minds. Shaped or not, the underlying materials and constraints of each medium give it a "texture" that naturally bends in some directions and not in others.

It's not just that a medium can be created with certain ideas from cognitive science. Rather, these ideas can be used to weave a medium in which readers' thoughts and actions are inevitably—even invisibly—shaped by them. Mathematical proofs as a medium do not just consider ideas about logic, we do not attach ideas about logic to proofs. Form is made up of ideas about logic.

So as long as one engages with the author's work in a medium—in an obvious way; engaging in that medium is the equivalent of "read all the words on the first page, then repeat on the next page, and so on"—people will Will automatically do the necessary understanding? So in a way, the default behavior and mindset when using this medium is the same as "understanding what is necessary"?

This is a high requirement. Even at the theoretical level, it is not clear what is necessary for understanding. In fact, this framework is too narrow: there are many paths to understanding a topic. But cognitive scientists and educators have mapped parts of this space, and they've distilled some powerful ideas that we can use as a starting point.

For example, people struggle to absorb new material when their working memory is already overloaded. To be more specific: if you've just been exposed to a lot of new terms, you probably won't absorb much from a sentence that uses many of these terms at the same time. So maybe part of the "necessities for understanding something" is that most of its prerequisites must not just be familiar, but fluent, encoded in long-term memory.

To help people encode more information into long-term memory, we can draw on another powerful idea from cognitive science: spaced repetition . By retesting the material you've learned at ever-expanding time intervals, you can easily and reliably store large amounts of information in long-term memory. Of course, memory is only a small part of "understanding," but to illustrate how one begins to deal with understanding as a whole, let's explore how a medium can be woven out of these two ideas about memory.

My collaborator Michael Nielsen and I did our first forays with Quantum Country , a "book" on quantum computing. But reading this "book" is not like reading any other book. Explanatory text is closely intertwined with short interactive review sessions, intended to leverage the points we have just introduced. Reading Quantum Country means reading a few minutes of text, then quickly testing your memory for everything you just read, then reading for a few more minutes, or scrolling back to reread certain details, etc. Reading Quantum Country also means repeating these quick memory tests at ever-widening intervals over the next few days, weeks, and months. If you read Chapter 1 and then take a memory test in your inbox the following days, we expect your working memory to be significantly reduced by the time you read Chapter 2. In addition, interspersed review sessions lighten the metacognitive burden usually imposed on readers: they help readers see where they have absorbed the material and where they have not.

Quantum Country is just one piece of a memory puzzle. How do we design media so that "readers" naturally form rich associations between the ideas presented? How do we design the medium so that the "reader" interacts creatively with the material naturally? How do we design the medium so that the "reader" naturally competes with competing interpretations? If we stack these questions together, we are left with the question: How do we design the medium in which "reading" equals "understanding"? A more detailed treatment of such a research program is beyond the scope of these brief notes, but I believe that the answers to these questions can change the pace of human knowledge, echoing the changes that the books themselves sparked long ago.