Recall rates are a misleading proxy for more meaningful goals of the mnemonic medium

Since the Mnemonic medium generates tons of data about what and when people remember, it’s tempting to focus on that data to understand how to improve the medium. But the real goal here isn’t to help people memorize things. It’s to augment people’s capacity to do things they find deeply meaningful, whatever those are: Tools for thought should be evaluated in the context of intrinsically meaningful purposes.

I find an analogy to backpropagation in neural networks useful. The “obvious” thing to do is to think of recall error as the target of gradient descent. But what we really care about is practical capacity—what do they do with the memory? Since we don’t yet understand What’s the big-picture impact of the mnemonic medium on readers?, optimizing for memorization alone may or may not produce practical capacity. We don’t know! Conceptually, I think we want to treat recall as a hidden layer, and to think of ourselves as “doing backpropagation” on some measure of practical enablement.

The existing Spaced repetition memory system culture’s fixation on often-meaningless memorization makes this failure mode particularly common: The dominant culture around spaced repetition memory systems is fixated on meaningless goals.