The mnemonic medium should give readers control over the prompts they collect

The initial mnemonic medium is implicitly authoritarian in premise. These mechanics:

• discourage/prevent many valuable active modes of reading, e.g. skimming for gems, reading with a specific purpose in mind motivated by authentic use, reading non-linearly, etc: The best way to read is highly contextual
• encourage a passive mental posture—the reader’s along for the ride, contra Do your own thinking, the active choice implied by Spaced repetition memory systems make memory a choice
• create a school-style learning aesthetic, in practical misalignment with Orbit - Values, e.g. “helps you deepen your relationship with whatever you care about most … it’s for ideas and imagery that you could talk about all night … not for things you think you “should” be engaged with … not “educational” in tone”
• create active frustration when a prompt seems poorly-written or non-meaningful; see e.g. Mnemonic medium readers sometimes feel impeded by authors’ wording choices, Cloze deletions often create frustrations in non-technical mnemonic texts
• impede fluid interaction between the medium and the rest of the reader’s knowledge work, unlike e.g. The mnemonic medium can be extended to one’s personal notes
• create the (correct but bad) sense that the prompts aren’t “yours”, material to be deployed as you see fit; instead, you’re visiting a museum created by the author, and everything is under glass
• make it harder to use the medium as a scaffold to learn to write your own prompts (The mnemonic medium may help scaffold prompt-writing through author-provided prompts)
• discourage the essential practice of iterative refactoring (which spaced repetition systems systematically under-value, in general)

Forward directions

I’d like to swing the vibe aggressively more in the reader-centric direction. This will be tough to do while maintaining other important properties of the medium. Expert-authored prompts really do contribute a great deal of value (The mnemonic medium supplies expert-authored prompts to remove the burden of prompt-writing). We want to reap that value (or perhaps even enhance it!) while also providing fluidity, malleability.

There are some relatively simple ways we might begin: readers should be able to disable or skip prompts that don’t resonate; readers should be able to make simple edits to authors’ wording; the embedded interface shouldn’t discourage people from reviewing only part of a page. All this will generate interesting signal for authors: The mnemonic medium can generate interesting author analytics

But I think we’ll have to go deeper than this. A few directions I find promising, which of course need to be fleshed out more thoroughly:

• You should be able to fluidly move between consuming a prompt, to editing its text, to refactoring it (one-to-many) and its adjacent author-provided prompts (many-to-one), to writing new prompts of your own inspired by those prompts.
  • Ideally, this same interaction model should extend to the review experience: as you’re reviewing, if you think of a way to improve a prompt, or you’re reminded of a new prompt you want to write, that should be just as fluid.
• We should explore pick-and-choose modalities for ingesting prompts from the text. Imagine that as you read, some phrases are highlighted in a special way indicating that the author has provided prompts representing that idea. If you find something interesting, you can click on the highlighted text—and boom, now you’ll remember it forever.
  • We’d want the same fluid consuming-to-editing interactions for this type of prompt.
  • This type of interaction points the way to crowdsourced annotations, ML-generated prompts, etc. More generally: imagine if you could point at any idea you found interesting, anywhere, and feel assured that now you’ll remember it (and ideally deeply internalize it) for the long term.
  • Ozzie Kirkby suggests that there may be some interesting middle ground here which would encourage readers to write their own prompts representing the details they find interesting: maybe authors (or ML, or the system) can provide partial/“starter” templates for key text regions, which the user would adapt/complete as they see fit.
• The presentation and interactions should make it very clear that your collection of prompts is yours, even if you’re working with author-provided prompts. Taylor Rogalski points out a helpful analogy to making a copy of a Google Doc to scribble all over it.
• We can start to experiment with fluid interaction between the medium and the rest of the reader’s knowledge work by fleshing out My implementation of a personal mnemonic medium. That direction is really not quite the same as what the rest of this note is talking about: I’m not sure how author-provided prompts interact with this kind of note-taking. So maybe it’s a red herring… but my instinct is that it’ll be instructive, and that all this work is intertwined.
• I suspect this work will inform and intersect with the design of a “library view” in the Orbit app itself. One related priority for me here is to fuzz the “boundaries” between prompts; see discussion in “How to write good prompts” here, grep (“revision is a holistic endeavor”). This also means fuzzing the boundaries between the author’s prompts and the user’s prompts. We should probably avoid paradigms which put those sets in separate conceptual “places”—but I expect we’ll still want some affordances showing provenance.

==TODO much of this section should be fleshed out in more elaborate separate notes==

References

I notice that there’s a connection here with the ideas in Heer, J. (2019). Agency plus automation: Designing artificial intelligence into interactive systems. Proceedings of the National Academy of Sciences, 116(6), 1844–1850: just as a collaborative human/AI interface requires a coextensive data canvas, a collaborative human/author interface requires something similar.

See notes from Yusuf Ahmad on this tension, which he frames as a gulf between “industrial” educational architectures and social constructivist architectures: https://www.notion.so/Placing-Cog-Sci-ideas-within-a-broader-Learning-Architecture-fa69fc15ba9e41b8a2e136f7fbe0362b

Q. Unexpected connection between the Mnemonic medium and Heer, J. (2019). Agency plus automation: Designing artificial intelligence into interactive systems. Proceedings of the National Academy of Sciences, 116(6), 1844–1850?
A. Just as a collaborative human/AI interface requires a coextensive data canvas, a collaborative human/author interface requires something similar.