How Should I Learn Cognitive Science?

My usual approach on this blog is to write about a strategy once I’ve already got it working. This time, I thought it would be interesting to instead focus on a current learning challenge I have, and my thinking process about resolving it.

One of the big learning challenges I’m working on right now is to learn cognitive science. Here’s a quick summary:

  1. Cognitive science is a multidisciplinary field crossing psychology, neuroscience, computer science, linguistics and philosophy, all using different tools to answer the question of how the mind works.
  2. My current benchmark for the curriculum is UC San Diego’s reading list for their incoming doctoral students with backgrounds in other fields. This consists of about 40 textbooks.
  3. My first goal is to read these textbooks, but I’ve also studied from outside this list when appropriate. For instance, when struggling with a Neurobiology textbook, I did Duke’s Medical Neuroscience class via Coursera first.
  4. So far I’ve completed about eight of the textbooks, with another few partially completed.

The challenge is still early on, and since I’ve opted to not make this one heavily time-constrained, I’m still figuring out what’s the best way to learn as I go.

How Do You Practice a Theoretical Subject?

One of the key ingredients of all of my other ultralearning projects was practice. The MIT Challenge had tough final exams. The Year Without English had immersive conversation practice. My Portrait Drawing Challenge was almost entirely practicing drawing.

Part of the reason for practice is simply as a way to test yourself. If I hadn’t done the final exams for MIT, my confidence as to how well I had learned the material would be pretty low. After all, it’s certainly possible to watch a lot of lectures or read a book and not be able to use the knowledge.

However, verifying your ability is only a small part of the benefit of practice. Doing practice is actually a much better way to learn the material. Exams didn’t just confirm I knew the MIT material, it was how I actually learned it.

Practice, therefore, is a crucial ingredient to learning well. But how do I practice cognitive science?

This has been a tricky problem because my curriculum is fundamentally a reading list, not classes. So there are no exams, assignments or projects. The nature of the material also makes practice difficult. While reading a programming textbook might not require practice, it will suggest some possible projects after. But what practice does a neuroscience textbook suggest?

I’d like to share some of what I’ve been trying to cope with this problem, discuss the strengths and weaknesses, and share some of my future attempts to resolve this problem going forward.

#1. Writing Book Summaries

My first attempt was an effort of least resistance—writing book summaries. After I finished a book, I would write a 1000-2000 word article, summarizing the contents of the book. I’ve done this for most of the books I’ve completed so far. I usually go chapter-by-chapter to prevent skipping.

The advantage of this approach is that it is quite easy to do, maybe an hour of work for a book that took 10+ hours to read. It also provides notes for the book, so you can later review a condensed version. Finally, summarizing forces me to think about what the big ideas were and organize my thinking.

Despite this, there’s some clear disadvantages to this approach. Importantly, a lot of depth is sacrificed in a review. Taking 600+ pages of tersely written description into 2-3 pages of summary means I’m simply dumping 99% of the information provided.
Because there’s no exact criteria of what I need to include, this compression also allows me to be selective in what I write, meaning I can write about the things I understood well and avoid those I didn’t.

The book summary approach also doesn’t practice any technical skill. So if I’m reading a book on natural language processing, I have zero experience working with any of those systems. If I’m reading a book about Boltzmann machines, I’ve never used any of the equations. I know from doing the MIT Challenge, that this kind of practice was more than half of the actual learning experience, and it’s being entirely omitted by just summarizing.

#2. Taking Question & Answer Notes While Reading

Something I attempted for one book was to take notes which were in question-form, with a note to the page number in the book. My idea was that this would mean reading created its own practice questions later.

The advantage of this approach, versus just taking notes, was that when reviewing them, I could do active recall instead of passive review.

The problem, which I hadn’t anticipated using this technique, was that I tended to write very difficult questions that referenced an obscure detail of the chapter. Rather than focus on what were the big ideas, I got caught up quizzing myself on some obscure facts or opinions of the authors.

This led to the questions being almost entirely unanswerable, and reviewing them again, I felt they didn’t capture the spirit of what I was trying to learn from the book.

#3. Teaching Selected Lessons and Recording Them

Later, I tried flipping the problem around. Maybe, if my main goal was to understand cognitive science well enough to write about it intelligently, then I shouldn’t be focused on practice questions but on teaching. I could teach selected lessons from each book and that would ensure I understood it.

The advantage of this approach is it does foster a deep understanding of what you’ve covered. The downside is that it is incredibly time consuming and covers far, far less than the looser book summary approach.

I still think this method could work well if the book were about arguing a single thesis. However, most of the books I’m reading are surveying large literatures, so teaching specific lessons is going to omit almost everything else.

For my first attempt with this method, I took the book Human Memory by Alan Baddeley, and figured I would do a 5-minute lesson on some of the chapters (there were over a dozen). By the end, my recorded lesson was 20+ minutes and covered only briefly the first part of just one chapter.

What Should I Do Instead?

As I mentioned in the beginning, I still haven’t resolved this challenge. But here are some further experiments I’ve been thinking about to get at solving this issue of practice:

#1. Post-reading Question Book

I’m beginning to suspect my failure with writing my notes as questions came because it’s difficult to think of intelligent questions while also managing the cognitive demands of reading a book. Given more time and space, I might be able to create a list of questions after reading that capture what stood out to me as the important points.

The idea here would be that, after reading, I would create 25-50 questions for each book, as well as my own answer at the time. I could then keep two electronic files (one with just the questions, one with questions and answers) and randomly quiz myself on these later.

#2. Supplementary Classes

It may just be that books aren’t going to cut it for some topics. So I may need to actually find classes (either MOOCs, OCW, or just begged/borrowed real class materials) that will have included practice components.

I’ve already done this with Neurobiology, taking Medical Neuroscience and passing the exams, made a huge difference in how much I was able to understand and retain from that class. Perhaps I’ll just do this for all of the topics I want to explore more deeply.

Although this is a completely valid solution, it also strikes me as a bit of a cop-out. Why read the books at all if you’re just going to take classes on them later?

#3. Personal Project

Another alternative might be to read the materials and accept, at least for the moment, the insufficiency of my understanding. But, once the reading is done (or nearing completion) switch to working on an implementation project. The project would force me to go back and assemble knowledge I’d previously learned in a richer way.

Examples could include: attempting to write/publish a journal article on the topic, designing some kind of computer model to simulate some of the work or recording a set of lectures around key themes of cognitive science.

This approach appeals to me as well, but I don’t like the idea of delaying practice so far into the project.

What Would You Do?

I didn’t bring this up simply to show that I, too, suffer from learning challenges. Instead, I wanted to showcase my thinking about these difficulties. Having done big, mostly successful self-education projects, I’m acutely aware of what needs to be done to make them work.

I suspect that many people would look at my reading list and not see a problem at all. Heck, just reading all those books is going to be challenging enough, why worry about additional practice on top of it?

But I suppose that’s also what motivates me to improve my understanding of how learning works. I’m always eager to try exploring what might be the deficits of my current approach and see if there are ways they can be overcome.

Learning What Can’t Be Taught

The example most people think of for learning is school. However, I’d like to argue that many of the features of learning in school are the exception, not the rule.

One example of this is that learning in school is almost always based on a standardized curriculum that teaches a heavily codified set of facts, concepts and skills. This is a necessity for the formal education process— it’s very difficult to train teachers and almost impossible to create uniform outcomes without articulating exactly what they need to teach.

However, the ubiquity of this style of education can create the illusion that most learning we want to do is of this type. Looking more deeply, it’s obvious to me that many of the things we’d really like to learn are not easily codified and so they’re not amenable to mass education.

How Did People Learn Before Schooling?

Interestingly, our stereotype of how learning works—standardized lectures, curricula and readings—is a much more recent invention. For the vast majority of human history there was no writing, reading and probably no dedicated teachers.

Instead people learned by copying. They found prestigious people and emulated their behavior.

Often this emulation was beneficial, even when both the model and the learner had no idea what the underlying mechanisms were for its usefulness. As Joseph Henrich documents, South American tribes avoided the scourge of niacin deficiency, common in Europe at the time, by cooking their corn with a bit of ash. This was a tradition passed down over time, probably the result of imitating healthier tribe’s eating rituals, but nobody knew the reason why it worked.

It seems this kind of learning was the dominant form of learning in premodern societies, which can help to make sense of how effectively people were able to live and survive in environments, even when their individual beliefs about the nature of the world seem almost absurd to us today.

Formal education came much more recently, but its prevalence now can create the impression that learning is mainly an activity of remembering things from books or understanding key, predefined concepts. However this wasn’t true for the majority of our history, and it probably isn’t even true today—we still learn most of what matters through some variation of the cultural learning protocol used in hunter-gatherer tribes.

What Can’t Be Taught?

Instead of equating learning with what is commonly taught in schools, I prefer to think of formal education as being the tip of an iceberg sitting on a far larger amount of tacit learning. This is knowledge and skills that we can’t articulate easily, but still make up useful knowledge for doing things.

I also think an inability to notice how large and predominant this tacit knowledge is can lead to a lot of frustration from students. They learn a lot in schools, but then go to work and find that they can barely do anything, and need to learn a completely new set of skills and knowledge to be effective.

Many of these same students then look back and think their education was a waste of time. In reality, it probably was as good as it could have been. The problem was simply that a lot of the knowledge and skills they needed to learn couldn’t be taught directly.

How Do You Learn What Can’t Be Taught?

I’ve thought a lot about how you learn this kind of tacit knowledge. Unlike learning from a formal curriculum, or a well-defined skill, there’s a number of problems with learning in this way:

  1. You might not recognize the skill or knowledge others possess. The environment is noisy so it can be hard to distinguish those with skill from those with luck, innate talent or other advantages.
  2. You might not even know how to describe what it is you need to learn, even if you can recognize it in others. Much ink has been spilled trying to explain exactly what makes Steve Jobs an excellent entrepreneur or Warren Buffet a great investor, and they’re far from isolated examples.
  3. Because there’s no books you can read or classes you can take, even if you correctly identify which skills you need to improve, it’s not clear how you do that.

These barriers mean that, unlike learning history, calculus or accounting, much of what makes learning tacit knowledge difficult isn’t even a problem for most learning tasks.

How do you get around this? I think the answer is to see what has already worked in our species for millions of years—finding prestigious examples and copying them. This may seem unglamorous, but it’s the mechanism by which you’ve already learned most of your tacit knowledge and how most of humanity has learned since the dawn of history.

Find people who have the skills you want to learn, observe them closely and copy their approach.

One of the major bottlenecks in this process is access to good models to emulate. Therefore, anything you can do to expose yourself to skilled individuals will accelerate your learning. In this sense, networking and mentor-relationship building can be seen as a core learning skill.

Another challenge is that much of the tacit knowledge that these models possess, even they won’t be able to articulate. Just as the tribespeople who mixed the ash and corn together didn’t know about niacin, many of the successful behaviors of models are learned unconsciously and can’t be easily justified by giving reasons.

This means that asking successful people what to do is less effective than watching what successful people do. Ask, and you’ll get a confabulated rationale for their behavior. Watch, and you’ll see their knowledge in practice.

Ultimately, I think most of what we need to learn to be effective in the world, even in highly technical, codified domains such as law or engineering, is in the realm of this kind of tacit knowledge that can’t be taught. Learning well in the world must deal with this fact. It doesn’t matter how good you are at reading books if the knowledge and skills you need can’t be found there.