Today is my birthday. For the last seventeen years, I’ve been sharing a birthday post with personal reflections, and I’ll continue the tradition today.
My wife and I did our part in contributing to inflation by buying a house this year. Timing-wise it could have been a little better. We made our offer right before the market fell—yet before we cashed out our investments for the downpayment. That, on top of the general ridiculousness of our local real-estate market, made the experience a little stressful.
That said, it feels nice to move into a place that we own. I’ve lived in small apartments in downtown Vancouver for over a decade. While it was fun while I was in my twenties, it started to feel a bit cramped with a growing family. On top of the extra space, it’s nice not to worry that we might need to move again in a few years.
The past year was a difficult one for me creatively. Almost a year ago, I pitched a follow-up book to Ultralearning to my publisher. Shortly after, I found out the research I had wanted to use to support the book looked unreliable. This put me in an awkward position of having a book deal but not being entirely sure what to write.
Fortunately, my publisher graciously gave me an extra six months to organize an alternative. Still, the overall experience was frustrating as creating a book that’s true, useful, interesting and which I’m capable of writing can be quite tricky.
Nearly a year later, I have the beginnings of a book I’m happy with, but the writing is still challenging. I’m cautiously optimistic that I will be able to write a good follow-on book to Ultralearning, but it will still be a lot of work.
Reading Way Too Many Books
A positive side-effect of my writing headaches was that I ended up doing way more research. This is the fun part of writing a book.
I dug much deeper into foundational cognitive science than I would normally for an advice book. Foundational stuff tends to be both dense and somewhat removed from practice. At the same time, I’ve come to appreciate it as a tool for evaluating advice.
It’s hard to judge if an approach makes sense without having some theoretical underpinnings. This is especially true in an area like learning, where a lot of the empirical research is fairly poor, so you get mushy “everything works” research as well as biting “nothing works” rejoinders.
I did write three longer articles surveying ACT-R theory, cognitive load theory, and Construction-Integration. I also read a lot on situated learning, Direct Instruction, connectionism, constructivism, apprenticeship, and educational theory. I’m still far from an expert, but I feel like I understand the basic positions of the different camps, as well as the strengths and weaknesses of their arguments.
I always enjoy learning, so this part was fun, although it would have been a little less stressful if I hadn’t been doing most of it while also under a book deadline. In the future, I’ll be sure to separate exploratory and foundational learning from the targeted research I need to do to produce tangible output.
Evolving My Thinking About Learning
Going deeper into these foundational topics also shifted my beliefs about learning in a few different ways.
One major shift was recognizing the importance of having examples to learn from. I think I downplayed this a little too much in Ultralearning, as I mostly took it for granted that you’d be able to find examples. I had no trouble finding examples for the classes I took during the MIT Challenge. And finding word lists and grammar explanations for languages is easy. In both cases, it’s doing the actual practice that’s hard.
However, my deep dive into the research on cognitive load theory and direct instruction suggests that a lack of examples is actually a significant bottleneck to proficiency. People fail to learn complex skills like programming, painting or calculus because nobody breaks it down enough for them to get a foothold. Instead, they flop around trying to figure things out in a way that’s often inefficient and frustrating.
I’ve also updated my views on cognitive science. I started studying the subject when the deep learning craze was in full swing, so I tended to see older, symbol-processing approaches as an anachronism. Didn’t neuroscientists show that the people who thought the mind was like a serial computer were simply wrong? Isn’t learning just the result of complex neural networks, trained via backpropagation and reinforcement learning?
Going through the research, I now see that the two views are more complementary than I had previously realized. While at the low level, many computational neural processes resemble the kinds of machine learning algorithms used by Google and DeepMind, conscious thought is probably both serial and symbolic.
By symbolic, I don’t mean that the brain is literally moving little tokens that represent things around in the brain, but simply that we can learn patterns with variables in them. If I understand the pattern, “1 -> 1”, “2 -> 2”, and “3 -> 3”, I easily generalize that to “4 -> 4” or “2000 -> 2000”. This generalization requires the representation of knowledge to be more like “x -> x” than a simple memorization of fixed patterns.
So, the prior course of my thinking was closer to constructivism for education (skill learning is mostly doing, not reading or watching) and eliminative connectionism for cognition (learning is just neural weights, there’s nothing like rules or variables in the head). Now my views are closer to direct instruction (looping between examples and practice) and the vision of cognition espoused in theories like ACT-R.
The next several months will be a real push to finish my book on deadline. I could probably spend my whole life just reading books and thinking about them. Still, I’ve read enough in this area that I’m (hopefully) confident it’s time to assemble it into a book.
I’ll probably share more about the book as it gets closer. I haven’t yet landed on a title, but the core idea is related to the “see, do, feedback” posts I’ve written. I generally prefer books that are dense with ideas, and I expect the finished edition will try to explain what I’ve learned about how people get good at things in a way that complements my previous book, Ultralearning.
While I’m optimistic about the writing, I’m also aware it will probably be a bit stressful. Before becoming a father, I used to handle intensive projects by pouring on extra time. That’s now a lot costlier, so I need to be efficient—and clear-eyed about what things I can’t do—if I’m going to reach the finish line on time.
The past year definitely had more stresses than I would like, and it seems likely to repeat in the coming year. I’m not sure there’s a profound insight here; most of the stress is simply due to projects evolving in ways I didn’t expect when I committed to them. However, it may also be that I’m sticking to a success script that worked for me in my twenties when it needs some editing for the next phase of my life.
Despite these difficulties, I’m incredibly grateful. Being able to spend my time learning things and (hopefully) sharing them in a way that other people find useful and interesting is a profound privilege. At times, I feel a bit guilty for being stressed out, recognizing that I’ve already got my dream job. Of course, this is only possible because of readers like you. So thanks for listening to me for another year. Hopefully you’ve enjoyed the writing!