tft
memory

How to make memory systems widespread?

The music isn't in the piano. – Alan Kay

"If memory systems are so great, why don't far more people use them?" As someone who helped develop a memory system1, and who has written extensively about such systems, I hear this question a lot. It's a great question. I believe memory systems are shockingly powerful: when used well, they make memory a choice, and substantially enlarge the type of creative work a person can do. But this power is obfuscated by several obstacles. If those obstacles can be overcome, then memory systems ought eventually be used seriously and routinely by hundreds of millions or billions of people2, for whom the benefit will greatly exceed the effort required. In these rough notes, I sketch out a few things I think ought to happen in order for memory systems to be far more widely used.

By "memory system" I am referring to the penumbra of systems and media forms that are sometimes called spaced-repetition memory systems. Anki, SuperMemo, Orbit, the mnemonic medium, incremental reading, incremental writing, and many more. I prefer the name "memory system" over "spaced repetition system", because while spaced repetition is important, so too are many other ideas. I don't believe we've found the final form of such memory systems; although existing systems are powerful, I believe future systems will contain important additional ideas; indeed, there's a sense in which we don't yet know what memory systems are. I'm also using the term to include mnemonic systems such as memory palaces, though as Andy Matuschak and I have pointed out3, there are important ways they are substantially less useful than systems more oriented to spaced repetition. But I won't be surprised if future memory systems draw on all these ideas and more.

Consider skillful practices such as yoga, basketball, meditation, and dance. These have some notable properties:

(I've leaned in this introduction on a few examples: yoga, basketball, meditation, and dance; and snuck chess in at the end. But similar observations apply to many types of skillful practice. I'll occasionally use other examples below – surfing, violin, tennis, and so on, whatever seems apposite. In all cases my comments are based either on direct experience – I've played basketball and tennis, done yoga etc – or else on conversations with or direct observation of people who have done the skill seriously.)

In a similar way, the use of memory systems is a skillful practice, and has much in common with these other skillful practices. People sometimes think about memory systems in Silicon Valley terms: what's required to build a company and tool that will make memory systems widespread? What user interface, what features, what marketing? This isn't entirely the wrong way to think about the problem. But it's more helpful to think about questions like: how did meditation and dance and yoga and basketball become popular? What would high virtuosity in the use of memory systems look like? How do we get there? And how do we rapidly transfer those skills, so people can quickly get enormous benefit out of the use of memory systems?

One point easy to miss is that virtuoso skill in using memory systems is possible. It's a point usually not understood by people who've only used such systems briefly; there's no obviously ostentatious skill on display, like an amazing yoga pose or basketball slam dunk. Indeed, this point is often not understood by people who've used memory systems extensively, but don't realize they're using them poorly. I certainly wouldn't claim I'm a virtuoso – I often feel I've barely begun to understand them – but I instinctively feel they can be used extraordinarily well. My use has improved through hundreds of conversations with other people, so much so that I'm amazed at how poorly I used them early on, and how much I've improved. It's natural to suspect much higher skill levels are possible.

When I talk with skeptics of memory systems, there are a few common reasons they're skeptical. Let me just list a few here, without attempting much retort. One common set of skeptical comments is based on misleading beliefs about memory or memory systems: "I don't need memory systems, I can just look things up on the internet"; "memory isn't important for, or interferes with, creative work"; "you should let your 'natural' memory decide what is and isn't important"; "memory isn't the same as understanding" (a false dichotomy); "rote memory is bad and doesn't matter" (this one is beloved of progressive educators); or just people who are generally defensive about memory (this occurs often with ambitious people who believe their memory is bad: they need to believe memory doesn't much matter, to protect their self image). The net conclusion from each of these beliefs is that memory systems are not important.

A second issue is that people use memory systems very poorly. There are three broad reasons for this. The first is that often people have barely used memory systems at all, but nonetheless have strong opinions. They'll use a system a few minutes a day for a few weeks, and conclude that "memory systems don't work". This is like practicing violin for a few hours, and then saying at the end "that sounds terrible, this instrument is no good"4. In fact, the trouble is that they aren't any good with the tool. A second reason people use memory systems poorly is that, as I noted earlier, even if someone has used memory systems extensively, that doesn't mean they've learned to use them well. You can practice dance for thousands of hours on your own, but you may not end up very good. I sometimes chat with devoted users of memory systems, and find them making what seem to me elementary mistakes5. I suspect they're getting some benefit from the system (or they would cease using it), but they could get vastly more benefit with some improvements in use. Indeed, many common challenges with using memory systems may be addressed by more skillful use: "I got bored"; "I started to confuse memory with my creative work"; and so on, a plethora of problems which can be addressed by more skillful use. The third reason is that humanity-as-a-whole is still in the early days of using memory systems: we haven't (collectively) figured out how best to use them, or what their best form is. And so part of the challenge is to increase our collective understanding, just as we've gradually increased humanity's collective understanding of yoga, meditation, basketball, etc.

These are all reasonable critiques; but they're also for the most part beginner's critiques. Consider meditation. People have many reasons not to meditate. But meditation practitioners typically don't engage with those arguments head on. Instead, they more often say: it's impossible to explain the experience of meditation, it's something you must do, and which you very slowly get better at. And then rely on the testimony of reliable people saying, essentially, "look, it's a lot of time, but I get far more out of it than I put in". Put another way: the key thing is to make the core experience of memory systems extraordinarily good and valuable. It needs to be a very good use of time. If that's very obviously true, then people will be willing to work through the beginner stages, in order to get the benefit. If it's not true, then people shouldn't be using memory systems: they should go the gym, or spend time with their family, etc. To achieve this state of affairs will require us to develop deeper and better ways of using memory systems, and to develop better ways of transferring those skills. Put another way: modern yoga, basketball, meditation, and dance all arose out of strong cultures: a yoga culture, a basketball culture, and so on. And what's needed, in my opinion, is to develop a strong memory culture, one that develops our collective understanding of the skills, and which develops techniques for better transferring those skills.

In consequence, I mostly won't engage with the skeptical arguments above directly. They are important, and I've written about many elsewhere. But they're also all downstream of "develop a strong memory culture". What's required for that to happen? In the notes I sketch out a few elements of such a culture. Ultimately, I believe that's what will lead to memory systems being widely and seriously used, perhaps even by hundreds of millions of people.

My assumed audience here is people who use memory systems seriously: beyond basic rote memory (vocabulary, basic facts), and in support of major life tasks. The notes are not for people who haven't done such serious use, nor are they written to convince skeptics. If you're such a person, and for some reason decide to continue reading, you're going to be frustrated: I've deliberately avoided addressing beginner concerns, except incidentally. It's just too much like arguing the merits of meditation with someone who doesn't know how to meditate. Of course, figuring out those arguments is useful, but it's a separate activity! What I'm doing here is attempting to figure out what memory culture is needed to make memory systems as powerful as possible, and to make them as transferable as possible.

Developing a strong memory culture

I have a small group of friends with whom I talk over memory system skills. We don't get together systematically or as a group (perhaps we should). But we just occasionally talk over how we use the systems. I can't convey this experience easily, any more than I expect a surfer could easily convey surfer culture to non-surfers. It would take an entire book, an ethnography of nascent memory culture. But the conversations will sound familiar in tone if not in detail to anyone who has ever developed a skill in an esoteric area, where they find themselves on the leading edge. Here's a relatively non-esoteric example to give you the flavor:

"Oh, at conferences I've started trying to write a minimum of 5 questions in each session. A surprising side effect is that I'm finding I pay more attention during the talks."

"I've tried a similar idea! But I found I got too focused on minutiae, trying to meet my quota of questions. I thought maybe it was causing me to take my eye off the big picture. So I've been experimenting with writing 3+ 'big picture questions' at the end of the sessions. Just the anticipation makes me more aware of the big picture. Of course, some talks are just hopeless, and there's no real big picture."

"Yeah, I worried about that too. Some talks you should forget as fast as possible. How do you avoid cluttering up? Or deal with talks on subjects that really aren't supporting your core creative interests? It's a mistake to add questions you don't care about. So I wonder if the idea of quotas is a mistake from the get-go."

"Yeah, a different approach I've been experimenting with is [/etc/]".

This is an evocation, intended to remind you of what such conversation sounds like. I've had hundreds of hours of such conversation. It is, I suspect, a little like the early days of surfing: surfers trading and trying a thousand tips, sorting wheat from chaff, gradually iterating on ideas, discarding many, just trying to get a little better. Or it's like programmers, talking over tricks they've discovered for using their text editors. Or dancers, talking over detailed ideas about movement and music and emotion. In all cases, there's a little coterie of power users, developing arcana; a loose network of edge-of-art nerds nerding out over their esoteric skill. It's fun, it's bonding, and very satisfying.

I am certainly not suggesting there will ever be a widespread nerding-out over memory systems(!) But: someone learning to surf today benefits immensely from millions of conversations like this from the early surfers. The very best parts of those conversations are reified in today's tools; the most powerful phrases and explanations and physical modeling have been distilled down, and are shown to new surfers, who in turn pass them on to future surfers. Today's beginner surfers benefit enormously from that strong surfing culture.

When I talk about this experience – trading ideas about memory systems with friends – I'm describing the beginning of a little local memory culture. Another part of memory culture is writeups like Piotr Wozniak's wonderful wiki6, Gwern Branwen's notes7, Andy Matuschak's notes8, and many, many more. My own memory experience owes a lot to several such pieces – I'd tried memory systems several times before being convinced to give them a serious shot by Sasha Laundy's notes on using memory systems to become a better programmer9. And I've written or co-written many pieces myself10. All these things are pieces of a burgeoning memory culture.

I want to be more explicit about the distinction between the memory system, and the skill-in-use held collectively by the community of practitioners. In surfing, surfboards have changed a lot over the past 70 years. But surfers have changed far more: most of the advance in surfing is due to advance in the skills collectively held by the community. I sometimes find this distinction difficult to keep clearly in mind, in part due to a lack of widely-used language. I'll find myself falling back (as many do) to talking about improving memory systems, when really I mean improving memory culture, or the skill in using such systems that is collectively held by our memory culture. That skill can change and improve, even if the tools remain exactly the same. And, just as with surfing, I expect it's the memory skill held by the memory culture which is most improvable, far more so than the tooling. Better and more widespread tooling will be a consequence of improved culture, moreso than the culture will be a consequence of better tooling, just as modern surfboards are downstream of early surfer culture. Of course, it really goes both ways. But I meet a lot of people who assume tools mostly drive culture. And here I think it's mostly the other way around11. The question is less "how can we make a better system?" and more "how can we get vastly better at using the systems?"

As I said above, for memory systems to take off, their use must become both an extremely powerful skill, and a relatively transferable skill. If power-and-transferability gets high enough, a lot people will want to use them. At the moment, I expect they're mainly being used by bored people to badly memorize state capitals (and similarly dull things); to poorly internalize Chinese or English vocabulary; and by students of medicine and a few other memory-intensive disciplines. The use-by-students is interesting, but the other uses less so. But suppose we figure out how to routinely use memory systems so people can rapidly master and contribute to complex conceptual subjects. Then I expect they'll become very popular. But we need to improve the skills and the transferability quite a bit.

What will it look like to grow a memory culture?

It'll look like lots of little informal communities of practice, like those I described above, people nerding out, trying to push the edge of their art. Much will be purely private. Some will be shared in small groups. And some people will share their memory practice much more publicly. Better and deeper communities will form, both online and offline. At present, I've tried to find online communities, and most are at a rather disappointing skill level (though some of the medical communities are interesting). But if the techniques get powerful enough, people will run memory conferences to develop the practice. They'll offer memory classes and memory coaching and memory TikTok. They'll write books and run YouTube channels. More and more people will develop the use of memory systems in a virtuoso way. Some will just be fascinated by the systems, and perhaps use the tool far beyond what makes instrumental sense for them. But some will get tremendous benefits in other parts of their lives, far outweighing the cost. The best ideas and patterns of use will gradually feed back to modify the tools. And so on, a gradual blossoming of memory culture12. Most of these things are already happening on a small scale. But the culture has a huge amount of room to deepen and broaden.

A crucial part is for the culture to be cumulative. If it's just people reinventing the wheel over and over and over, then no collective progress will be made. This is part of why it helps to have shared resources like Piotr Wozniak's SuperMemo Guru wiki. It's why it helps to have a culture of citation and acknowledgment of prior art, and for people to do things like compile bibliographies or survey documents or books or FAQs or Subreddit wikis. Those acts of compilation and distillation and acknowledgment are how a culture makes progress, as opposed to merely covering the same ground over and over13.

Incidentally, this point of view throws some light on one of the best-known articles about memory systems. It's a 1988 article by Frank Dempster14, bemoaning the fact spaced repetition isn't widely used in schools. He lists nine impediments to their use. Now, of course, school systems can dictate by fiat, imposing whatever systems they want on students, including systems that work poorly or where teachers don't themselves understand very well how to use the systems15. So schools can impose memory systems, and I expect this will occasionally happen. But I expect a much more felicitous situation to obtain if this happens after memory culture gets much stronger. That way we'll have a much better collective understanding of how to use memory systems, of what they really are, and much better shared resources explaining how best to use them. This will put both teachers and students in a much better situation.

One way a memory practice is different from many of the other skills I mentioned is that it's largely internal. You can watch a surfer or a dancer or basketball player and gradually understand what they're doing, learn to imitate and even improve on it. The great snooker player Ronnie O'Sullivan has reportedly said that he taught himself to play as a boy by watching tapes of the previous generation's greatest player, Stephen Hendry, over and over. It's less easy to do that with a memory practice: more of what's happening is internal. Of course, people can talk about their practice, and they can share artifacts (like Anki cards). But on balance a memory practice is perhaps more like meditation in this respect than it is like dancing or basketball or many other skillful practices. I suspect that makes it harder to build a strong memory culture, since the skill itself is less ostentatiously evident, and so less easy to learn from.

A second way a memory practice is different from many of the other skills is that it's largely instrumental. Most people learn to surf or dance or play basketball as an end in itself, for the intrinsic enjoyment they get out of the activity. Certainly, there is enjoyment in learning, and I expect that is enough for some users of memory systems. But many users of memory systems use them because they are instrumentally useful for other goals in their lives. Indeed, my own personal experience is that they are most helpful where I'm doing a creative project in some area where I'm not especially expert, and trying to rapidly come up to speed. (In areas where I do have deep expertise, I am able to learn rapidly, and it's better to rely primarily on my natural memory.)

How widely should memory systems be used? How deeply should they be used? Most people don't play basketball or surf. And for most who do it's just a casual hobby. Only a tiny fraction of people will ever do those things seriously. So how many people should be using memory systems? Over the long term it's very difficult to know: other changes in our society may have a significant impact16. But, at this time in history, knowledge work is at a premium. I suspect at least hundreds of millions (and perhaps, eventually, billions) of people's lives would benefit a great deal from skillful use of memory systems. I don't think it's like surfing: I think it's perhaps closer to reading and writing; memory literacy may be a desirable thing to have as a near-universal goal. This is, of course, speculation on my part. But it's my instinctive belief. And, as with all my other speculations on the future here, I may be wrong!

I've mostly ignored the role of startup companies in these notes. The role is complicated, so rather than write another essay, let me just write one summary paragraph. I believe these two processes – the development of memory culture and startup companies – are to some extent naturally separate endeavours. I expect people will keep starting companies in this space; some will be modest successes. But until and unless memory culture gets a lot stronger, I don't think there will be a huge success. Now, it may be that the hugely successful company will drive the development of the culture; but I expect it'll be the other way round. In particular, I expect many hard problems about memory systems need to first be solved by the informal culture of practitioners17. This is similar to the path taken by modern yoga, which after extensive development through multiple traditions, has been appropriated by various modern yoga "brands". Those brands contributed modestly to our understanding of yoga, but added marketing and distribution. Put another way: I don't think a tool-centric and company-centric approach to memory systems is quite right; they will ultimately be necessary to achieve scale, and they're attractive to some people as a means of control and ownership. But they're not where the core problems are, any more than the main problem in surfing is making surfboards. Unsurprisingly, in such a situation there can be friction between aficionados of the skill, who may resent the appropriation and the rewriting of history ("we actually did it, those so-called pioneers just talked and built bad systems"); and the companies, who are proud of their marketing and distribution and working to make the skill widespread. Both things need to happen, but friction is natural, perhaps inevitable. In any case, I plan to keep contributing to memory culture.

Acknowledgments

Thanks to Sebastian Bensusan for encouraging me to write this, and to Sebastian and Andy Matuschak for feedback on a draft. Thanks also to Sebastian, Andy, Kevin Simler, and many others for conversations about memory systems.

Footnotes


  1. See: Andy Matuschak and Michael A. Nielsen, "Quantum Country", https://quantum.country/, San Francisco (2019). And: Andy Matuschak and Michael Nielsen, "How can we develop transformative tools for thought?", https://numinous.productions/ttft, San Francisco (2019).↩︎

  2. I don't know how many people routinely use them seriously and voluntarily today. I expect it's probably in the hundreds of thousand to millions, but could easily be wrong. It also depends on what precisely one means by "seriously" and "voluntarily". For instance, I'd be skeptical of including Duolingo in this – it's far from clear to me that Duolingo is a serious memory system.↩︎

  3. See https://numinous.productions/ttft/#two-cheers-for-mnemonic-techniques, in: Andy Matuschak and Michael Nielsen, "How can we develop transformative tools for thought?", https://numinous.productions/ttft, San Francisco (2019).↩︎

  4. This is common, though far from universal, on Hacker News, Twitter, and many other online forums. People release new tools, and there will be many very confident critical remarks from people who have used the tools briefly, and may not even be in the target user base. Strangely, this behavior is so widespread that it doesn't even seem all that notable. It doesn't mean there is nothing to learn from such remarks, but obviously they're of limited utility. It's just too easy to imagine the Hacker News comments upon release of a new tool called a "violin". "Sounds like cats fighting". "My arm got sore". "I couldn't play even very simple tunes on it, will never be used by serious musicians". And so on.↩︎

  5. I also wonder what elementary mistakes I'm making.↩︎

  6. SuperMemo Guru: https://supermemo.guru/wiki/SuperMemo_Guru↩︎

  7. Gwern Branwen, "Spaced Repetition for Efficient Learning", https://gwern.net/spaced-repetition.↩︎

  8. Andy Matuschak, "How to write good prompts: using spaced repetition to create understanding", https://andymatuschak.org/prompts/ (2020).↩︎

  9. Sasha Laundy, "Using flash cards to become a better programmer, Part 1", https://sasha.wtf/writing/anki-post-1/ (2015).↩︎

  10. Instead, see: Michael A. Nielsen, "Augmenting Long-Term Memory", https://augmentingcognition.com/ltm.html, San Francisco (2018); Michael A. Nielsen, "Using spaced repetition systems to see through a piece of mathematics" http://cognitivemedium.com/srs-mathematics, San Francisco (2019); and: Andy Matuschak and Michael Nielsen, "How can we develop transformative tools for thought?", https://numinous.productions/ttft, San Francisco (2019).↩︎

  11. Part of the reason is that tool-building companies have a strong incentive to claim afterward that they're the ones who invented the magic secret sauce. Often, their actual contribution is to capture and distill and market the output of some community. Those are valuable things, but it also leads to considerable confusion about the actual origins.↩︎

  12. A downside: if it's successful enough, there will be a certain amount of grifting. One great thing about the early days of a culture: there are usually relatively few grifters: they'd rather go where there is money or status. And at present there's far easier ways to make money or status than in memory systems. But it's quite sobering to read up about yoga cults and meditation gurus and controlling dance instructors and basketball coaches, and so on. It'll be a (deeply unpleasant) sign of success when there's a memory scandal.↩︎

  13. It's a problem with cultures based primarily in social media, which tend to be ephemeral and cover the same ground again and again, while making collective progress very slowly. Such collective memory and progress is, incidentally, something that academia is remarkably good at: norms about citation and acknowledgment of prior art are strong.↩︎

  14. Frank Dempster, "The spacing effect: A case study in the failure to apply the results of psychological research", American Psychologist, at http://augmentingcognition.com/assets/Dempster1988.pdf (1988).↩︎

  15. Like many, I have innumerable instances of this from my own youth. My favorite was the calculus teacher who didn't understand why integration and differentiation are inverse processes. I certainly don't expect everyone in the world to understand it, but for a calculus teacher not to understand it is unfathomable (but, perhaps, not all that rare).↩︎

  16. The one I hear most commonly: enthusiasts of brain-computer interfaces (BCI) who think memory systems are a waste of time. It's BCI-as-thought-terminating-cliche. Over a long enough timescale, it may even be true; on a long enough timescale, perhaps the only thing "worth working on" is slowing down the heat death of the universe. It's not a very good argument: memory systems work remarkably well, today, and the memory culture and memory systems can easily be improved a great deal. BCIs are interesting toys today, with any large impact on memory a minimum of decades away. On a 20-to-30 year timescale, I'm also more optimistic about pharmaceutical and nutritional and exercise interventions than BCIs. But I expect memory systems to easily beat all of those, over the foreseeable future. Incidentally, I'm not arguing that BCIs aren't worth doing, merely against the (silly, but I've heard it repeatedly) argument that other things are not. "Do what we can actually make work today, not just what makes you feel like the protagonist in a badly written sci-fi novel" is a position that some people find quite indefensible.↩︎

  17. A nice up-close depiction of this is given by: Eric von Hippel, "Democratizing Innovation" (2005).↩︎