Learning to learn

The approach usually taken to learning is to repeat. A good revision timetable will allow lots of time for reviewing material several times; a bad one will allow less time. With each repetition, each review of a list of bullet points or a slide in a lecture, you learn a little bit more, and the memory sticks a little better. A simple linear relationship between the number of repetitions and the amount remembered.

Except that memory works nothing like this. Linearly spaced repetitions are a poor way of memorising facts. Cramming works, for sure, but only to retain information for a brief period of time. And it’s true: the more you repeat material, the better you learn it. But what if you could invest that same amount of time spent cramming—or endlessly repeating—in a different way? What would be the most efficient way to learn?

Learning is best described in terms of forgetting. Plotting the chance of remembering a fact against time shows something interesting: it’s an exponential decay described as a ‘forgetting curve’, and it explains why even a couple of days after poring over a Latin vocabulary list you can piece together very little of it. The curve makes a rapid path downwards before levelling off. At first, memories fade very quickly.

But every time you review material, not only do you reset your forgetting curve by bumping it back up but you move onto a new, shallower curve. Now you can easily recall reviewed facts for five or six days before they fade. Another repetition, a new forgetting curve. The trick is repeating those facts at the right interval—too early and repetition is ineffective; too late and the memories are already gone.

This is the system that Piotr Wozniak spent the last forty years refining. A Polish memory researcher, Wozniak was frustrated as an undergraduate that he and his fellow students seemed unable to improve their English beyond a clumsy pidgin. Starting from the premise that repeating facts aids in their memorisation, he spent hundreds of hours optimising the best spacing for learning English words and tracking the ease with which he could recall them. By the late 1980s his efforts had been met with failure, but he continued to adjust the spacing between repetitions, mostly on paper or with primitive punch-card programs.

Years later, Wozniak eventually cracked the timing and developed a computer-based method for scheduling repetitions. His algorithm was the basis for his own program, SuperMemo, and later repackaged into more elegant open-source equivalents—Anki being one which remains the king of spaced repetition software.

You enter facts as question-answer pairs, whether a foreign word with its translation, an equation with a solution or a sentence with a fact missing. These pairs are the atomic unit of spaced repetition, and the less information they contain, the better your chance of learning them. You do a daily review, not usually longer than ten minutes, during which you are shown a question, try to recall the answer and then hit ‘show’ to see it. You grade your response; easy cards are scheduled for more distant repeats (they lie on shallower forgetting curves) and hard ones sooner (on steeper forgetting curves).

Digesting material into small question-answer pairs is an irritation, but as reviews are scheduled at increasing intervals determined automatically for each fact, you spend very little time actually reviewing each card. More accurately, you review each fact at the optimal time to strengthen its memory: just as you are about to forget it. The software knows which facts you are on the cusp of forgetting and gives you a prod at the right moment to bring them back to the front of your mind.

This phenomenon—that people learn much more efficiently and remember facts much better when they are spaced at increasing intervals rather than at linear intervals—is the spacing effect. By scheduling your learning in this way you do the fewest possible repeats per item to keep it in your long-term memory. Because of the huge efficiency gain over conventional memorisation methods, Wozniak predicts that most people can learn, and remember forever, a few million such items in a lifetime. Naturally, he sticks to his learning regimen religiously.

Of course, spaced repetition is no magic bullet. Unstructured memorisation will never by itself teach you a language, nor will it get you through your degree. It’s not a replacement for teaching. The usual advice still stands: take good notes, review them often, take pains to understand material you didn’t get first time round. And at first, progress can be slower than conventional methods (you really are better off cramming for that test in a week’s time). But Wozniak’s discovery and the software that makes use of it can help with what many people, at all stages of life, find the most tedious part of learning: bulk memorisation of raw facts.

By the sixth repeat, you now sit on a forgetting curve which projects years into the future. A further couple of reviews guarantees life-long remembering of that fact. Some would argue that they simply don’t want to remember everything they are taught, or don’t have time to break down their whole degree into individual facts. But Wozniak’s goal was never to tell people to remember everything they encounter or to digest whole textbooks into flashcards. With dedication, spaced repetition gives you the power to choose which memories to retain and which to allow to naturally atrophy. Those you do focus on you can remember for the rest of your life.

First published in Pi Newspaper, September 2010.


Currently CTO at Mast. Formerly engineering at Thought Machine, Pivotal. Makers Academy alumnus.

I've pledged to give 10% of my income to highly effective charities working to improve animal welfare. If my startup is successful, I hope to give away much more.

Also founded EA Work Club, a job board for effective altruists, and Let's Fund, a crowdfunding site for high-risk, high-reward social impact projects.

Subscribe here to get updated when I post.