Spaced Repetition: A Cognitive QS Method for Knowledge Acquisition

Gary Wolf had a blog post on the Quantified Self website. I gave a speech here last August about my Quantified Self for preparing on Jeopardy and did not focus on Spaced repetition and that talk is embedded in that blogpost.

Spaced repetition has to do with the forgetting curve, when we remember and forget things. Herman Ebbinghaus was a psychologist in the late 19th century and he was one of the first quantified self-pioneers. He did decades of memory experiments on himself. He was the only subject because no one else would agree to do that. He discovered the forgetting curve that when you remember something you will forget it, it will decay very fast and taper off. But as you get exposure to it again and again that decay will get flatter and flatter.

He is most well-known for his optical illusions with the orange circles are the same but appear in different sizes, but hopefully this is what he will be most well known for.

In the 1930s Cecil Mace developed the concept of spaced repetition, and this was you take the forgetting curves and use the knowledge that Ebbinghaus found in the late 19th century to create essentially a scheduling algorithm for when you should ping yourself with a piece of information you would like to remember.

In this graph, we have the x-axis is days or time and the y-axis is the probability you will remember something. So that very first time that you remember or exposed to it you are going to remember it in 10 seconds later if your working memory and short term memory are working properly. But if you try to think of it the next day and if the fact was not that memorable you might have a much lower probability of remembering it.

The naive algorithm when you’re going to learn something would be to look at it every day, and refresh every single day, or every hour. But in spaced repetition this review process can be optimized, so you can review more frequently when you first learned something and then refresh later on.

In this demonstration the probability is being kept at 90% above that you’ll remember it. So, for instance, if you were learning words in a foreign language and wanted to learn a 1000 words, hopefully, if you use this algorithm and you keep up with it, you will be able to remember about 900, about 90%. Some of the hard ones you might not remember or you’ll have to keep up with that.

If you think of this graph as things you remember all the time would be all along the top. Your name, you’re not going to forget your name, and things which are really obscure for instance the capital of some country you’ve never heard of would easily drop off and you would forget it. I like the look of this graph and just think of everything you know and it’s all encoded in your brain and it has these decay rates.

I should make a quick aside on how these curves arose. It’s obviously because of the biological basis of memory and how synapses, strength and with repeated exposure etc.

The advantage of spaced repetition and scheduling is you know where to aim next. So if you know a subject very well, and say you are a novice dart thrower and you don’t know where to go. You throw darts as a novice and go for the bullseye if you are playing 501 for the max score. But if you are an expert dart thrower, you throw for triple 20 because it’s worth 60 and not 50.

So, when did I first hear of spaced repetition? In April 2008 I read this article by Gary Wolf in Wired about Piotr Wozniak, who is a Polish biologist and computer scientist that essentially put this altogether.

He took the ideas of Ebbinghaus and Mace and he wedded them to computer program software. He did self-tracking experiments on himself and he released this program called Super Memo.

I read this in 2008 and I thought that’s just what I need, so in a certain way I like this because it comes full circle for my perspective with the Quantified Self movement. I was already tracking myself but I wanted a highly efficient way of learning, so I looked into some of the software solutions and I decided on Anki. There are many types of spaced repetition software – SuperMemo. I was a grad student so I didn’t want to pay the $50. I went with Anki. It’s free and it runs on lots of platforms. I have it on my phone and desktop and all of these electronic flashcards sync with the deck.

A lot of this is aimed at language learning. Anki means "to memorize" in Japanese and the gentleman that coded it lives in Japan and is a student of Japanese.

Now that we know what spaced repetition is and next is "how do you get started?" Typically, you would just see the character in this instance. You would think of it, click the button and then get the readings and the meaning. You also see that you can choose the left button. If you get it wrong and don’t know the answer, you choose the right button to mark that it’s very easy. If it’s somewhere in between you click one of the two buttons in between. You have a card count and it estimates the time it will take you to do the remaining cards. In this case there are 21 cards left, which will take about three minutes.

For instance, you can have a map, and other types of multimedia. You can have images, sound, video, Unicode and you can also treat the cards as HTML so you can link to anything on the web.

Anybody here from Belarus? You would click the easy button.

If you have been doing the deck for a while, it also tells you the schedule, the interval of when you’re going to see that card again. So a card that would be easy for you, you wouldn’t see it for three years. But it will probably be a slight waste of time three years from now but it’ll learn.

I should mention the linking to the web is very powerful because if you had never heard of Belarus you could link the answer to the corresponding Wikipedia article, for instance, and go and read the lead of the article.

Where the Quantified Self comes in is in all of the data. Every single deck stores all the statistics for you. It stores all the statistics of all the cards. So, for instance, this is a deck of countries of the world and flags and this is when the next cards are due. This is the schedule for all the cards in that deck. So the easiest cards are about six month away on this timeline. This timeline is only up to six month, so the Belarus card that is still three years away would be way over there. And there are some to the left of the zero time point so I have not caught up with all the cards in this deck.

At the same time you can see your number of repetitions. In other words, "How many times have you looked at cards." So, for some reason, 45 days ago I did 200, and that sounds like a lot but when you’re looking at a country just thinking of it, you can do it in three seconds. So that was just 10 minutes, maybe. Likewise, review time on this deck for the past month is three minutes. That is the max minutes up there, so it’s like 10 minutes over the past month to stay fresh on country data.

And once again you are probably wondering why you’re studying countries, well that’s because I was on gameshows and they ask you this stuff, but it’s good to know.

The buttons one through four, one is one being hard, four being easy. You can see that as cards are new, maybe about 40% of the time I get it wrong, but as the cards mature, after you see them a while, and see them a few times it goes down and down, and so 20% are hard for the mature cards. Just like that original slide it was set at 90% at the cut-off, you could imagine for this material the cut-off is about 80%.

You can also see here for this individual card, you can see that I created the deck three years ago. I first reviewed that card 1.2 years ago. I’ve done it five times and I got it right five times. That’s why if Anki shows you the card five times over the course of year, the spacing algorithm then knows that you know it and it’s not going to ask you for a few years.

There are many options where you can choses how new cards are displayed, whether randomly or in alphabetical order. Typically random is going to be best, unless you want to learn information that’s in order. For instance, you can use spaced repetition not just for traditional flashcard uses, but you could put in Hamlets famous soliloquy, with each line, like "to be or not to be that is the question." And then that is the front of the card, the next line is the back. And if you do that and use those cards, it’s a very efficient way of memorizing the soliloquy.

Anything you can put into a flashcard format you can use with these applications. Sometimes things which you might not think you can put into a flashcard format, you actually can.

Once you have all that data, just to bring it full circle, you can make a representation, for instance, of what you know with respect to all the decks. This is a bubble chart made by the Esquire department and it ended up working out very well for me, learning this information over a year or two.

In conclusion, I just want to say it’s highly efficient learning, a time saver, an excellent way to maintain knowledge, and I think everyone can benefit from this. How many people have used a spaced repetition, flash cardprogram? Probably about 10 people in the audience.

I think the hardest part is getting started because knowing what do you want to learn. If you’re not motivated to learn something, you’re not going to learn it. But if you are, you can go do it. Some of these decks, for instance that countries deck, you can just download it. You can download 10,000 words in Spanish, a 1,000 most common French words, Turkish grammar. A lot if it is language learning, but you can also learn all the mixed drinks, etc. There are just so many uses – guitar chords, for instance.

Thank you.