Bring meta-rationality into your Orbit

Example of Orbit SRS in use

Mnemonic text is a new medium for thought. I wrote my most recent essay “Maps, the territory, and meta-rationality” with Orbit, a technology for mnemonic text, created by Andy Matuschak.

It’s common to read an exciting book or blog post that made perfect sense as you went along, but then to find that you can’t explain it to someone else over lunch the next day. And a few months later, all you have is a dim memory that it was cool, and maybe a one-sentence takeaway summary. Nearly all the substance got lost.

Orbit aims to deepen your understanding, and to make your memory permanent.

We’d love to hear about your experiences with Orbit, in general and in this essay in particular. More about that below! But you can just go try it out now; it’s meant to be self-explanatory.

Orbit combines two innovations developed by Matuschak in collaboration with Michael Nielsen:

Spaced repetition for boring stuff

Orbit is about using spaced repetition for exciting stuff. But traditionally it’s used for boring stuff, and I have to explain that first. (If you already know about this, you can skip to the next section!)

Our memory for details fades rapidly over time unless we refresh it. There’s a ton of cognitive science research on this. In short, taking a tiny effort to retrieve a fact from memory, frequently at first and then only occasionally, reinforces it. After a few repetitions—totalling less than a minute of work—the knowledge sticks permanently.

Spaced repetition systems (SRSs) automate this by quizzing you periodically on material you have chosen to learn. As you learn a “prompt”—a question about the topic—the program spaces it further out in time: it asks you about it less and less often.

Most people use spaced repetition for learning tiny facts. The two classic examples are word meanings (when learning a second language) and anatomical trivia (when studying for medical school exams). That sort of spaced repetition is highly effective, but not fun or interesting, and it’s irrelevant unless you need to learn something like that.

Spaced repetition for understanding

Nielsen and Matuschak (and others) have discovered that SRSs are also effective for learning abstract conceptual material, not just atomic factoids. More excitingly, this “learning” involves deepening understanding, and not just the ability to parrot definitions or procedures. Michael and Andy have written several essays about this. Some are linked from the Orbit home page, and others from their personal sites. This material is dense and extensive and I won’t try to summarize it.

Exactly how and why concept learning with SRSs works is not yet clear. Psychological research on spaced repetition has only looked at fact learning so far.

Nielsen, a former physics professor, has observed that detail knowledge is a prerequisite for concept understanding. Often when his students were struggling to understand an aspect of quantum mechanics, the problem was just that they had forgotten specifics of earlier material on which it was based.

We understand concepts largely through their connections with each other. One way spaced repetition prompts may help is by reinforcing that web of relationships.

The mnemonic medium

Despite solid evidence for the dramatic effectiveness of SRSs, not many people use them. One reason is that creating a set of questions and typing them into the program is tiresome. Another is that figuring out good questions to ask is itself a skill. It’s not obvious, when you are starting out, which points are important enough to be worth the effort of remembering them. Also, it’s often unclear, as you read a text, how to phrase prompts for maximum effectiveness.

In the Orbit mnemonic medium, the author of a web page intersperses prompts within the text itself. The reader gets nearly-immediate reinforcement of memory for what they’ve just read. If the questions are written skilfully, they also get a nearly-immediate check on whether they’ve understood what they’ve just read. And relevant questions are entered into the reader’s personal SRS with nearly zero effort. And, because they are written by the text’s author, who is supposed to know how to write good SRS prompts, they may be more effective at extracting and reinforcing its key points.

Spaced repetition for self-transformation

Due to an unfortunate error in gender assignment at birth, and archaic cultural norms concerning proper occupations for young ladies, my spouse Charlie Awbery was tracked into humanities courses, despite excelling at science and math as a teenager.

Early in our relationship, Charlie recognized that I think qualitatively differently than they did, or than anyone else they knew. (They had never talked much with anyone who’d done advanced work in STEM subjects.) This was not a matter of specific knowledge or abilities, but cognitive style. They found this compelling, and very methodically went about learning the alternate way of thought.

One difference they identified early was that our memories worked quite differently. Charlie had near-perfect recall for events, like what we had had for breakfast on any given day in the past year, but was completely useless at facts. I can’t tell you what I had for breakfast ten minutes later, but if you want to know the melting point of bismuth, I’m your man.

I explained the cognitive science of episodic vs. semantic memory, and Charlie decided to shift the balance. Using an SRS was part of that, and it was dramatic. They’re better at facts than I am now, and often I have to remember events for both of us.1

Through this and other non-traditional means of learning, Charlie came to think like someone with a STEM degree. That went well beyond factual knowledge, conceptual understanding, or skill development. It was fundamental reorganization of their worldview and way of thinking and acting. Using an SRS was only one part of that, but a significant one.

This was fascinating to watch, and gave me a visceral (if vicarious) sense of the power of spaced repetition. It prepared me to find Nielsen and Matuschak’s explanations convincing, in the absence of rigorous evidence.

Spaced repetition for learning meta-rationality

The new “Maps” essay, the first I’ve written with Orbit, is about meta-rationality. Meta-rationality is another way of thinking, which is “meta to” STEM-style conceptuality. It is about how and when to apply rational knowledge and methods.

That may make the Maps essay less than ideal as a first text to Orbitize. There’s solid evidence for the efficacy of SRSs for learning facts, and persuasive anecdotal evidence of their efficacy for learning concepts. But meta-rationality is not primarily conceptual; it is largely tacit.

Is Orbit efficacious for learning meta-rationality? One reason to suspect it may be is that meta-rationality depends on mastery of masses of concepts, in the same way that rationality depends on mastery of masses of facts. If SRSs help you learn concepts partly by stabilizing your access to facts, perhaps it can also help learn meta-rationality by stabilizing your grasp of concepts.

The Maps essay explains that meta-rationality takes purpose and context much more into account than rationality does. Those determine the “how and when.” Just noticing opportunities to apply it—times when mere rationality is inadequate—are a substantial aspect of meta-rational skill. It is possible that further software tools could assist that, and we’ve begun brainstorming them. But also, just recurring contact with the themes of meta-rationality, via spaced repetition, may help connect the ideas to specific applications as they occur.

My experience of Orbit

I wrote above that “the text’s author is supposed to know how to write good SRS prompts.” This generally comes only through extensive experience using the system. So it is super embarrassing to admit, this far into the post, that I have nearly none!

Orbit is the first SRS I’ve used, and I began only a couple weeks ago. I had previously tried several times to to use Anki, on Charlie’s recommendation among others, but found the interface so horrible that I immediately gave up on each attempt. Orbit, by contrast, has been genuinely enjoyable; I recommend it even just for its excellence at the user experience level.

For the benefit of Orbit text authors, Andy has written a guide to how to write good prompts, and has also run workshops which I’ve participated in, and I’ve tried to take his advice on board. If you find my prompts obtuse, though, it reflects my own inexperience.

I’ve found the experience of writing prompts unexpectedly interesting. It forced me to revise the Maps essay substantially, because I realized I had left key points for the reader to infer, rather than making them explicit. It may be significantly clearer as a result.

I added the prompts at the end of a six-year writing process. As the last bit of the essay explains, it was an unusually slow and frustrating bit of work, partly because I didn’t know where I was going with it.

I suspect now that writing towards a mnemonic text from the outset might be a whole ’nother thing. It might help me, and other authors, get clearer about what an essay needs to say.

(Software folks may recognize an analogy with test-driven development here!)

Please tell us about your experience with Orbit

Orbit is an experiment; my Maps essay is one of its first applications. You can help us learn more about how to write prompts that help you understand and remember reliably and with least effort.

We’d love to hear how it goes for you. What works well, and what doesn’t work as well as it should?

We’d like to know:

We want feedback about:

We’d love to hear from you:

You could leave feedback via a comment on this post, or on the essay. That’s ideal, so other readers can reflect on and riff off of your observations. Alternatively, you can contact either of us privately by whatever channel, and we’ll share what seems appropriate just between the two of us.

Since Orbit is brand new tech, there may be bugs, although it seems solid to me now. If you encounter any problems, please email andy@andymatuschak.org.

Support Andy’s research on tools for thinking

Orbit is one product of Matuschak’s broader research program: building innovative software tools for thinking. He’s chosen to do this work outside both academia and the tech industry. He certainly could find positions in either, but both make genuine innovation preposterously difficult—despite their stated purposes.

Instead, he’s pioneering a new model of crowd-funded research, via Patreon. Because the university system is gradually imploding, new research funding models may become critical for progress. He writes:

If you enjoy my work, I invite you to become a member of my lab community. You’ll get behind-the-scenes articles and early access to my projects. Think of your contribution not as a “tip”—buying me a coffee in appreciation—but as grant-making.

If you find Orbit interesting or useful, please support Andy’s work via his Patreon.

  1. 1.Charlie didn’t intend the decrease in episodic memory capacity. I don’t know if that is a natural consequence of training semantic memory, whether it’s inevitable, and if there’s any research on this. Anecdotally, there seems to be a trade-off. It certainly seemed worth it to Charlie, but maybe SRSs should come with this caution label?