In 1970, the harrowing almost-fiasco that was Apollo 13 promoted the astronaut’s status from being a passive payload to the active role of an expert again ₀. Four days of Type III fun ₁ in which the crew hot-wired the landing module to provide primary life support made it impossible for anyone to think of them as just spam in a can ₂ anymore. The idea of the perfect software design died, and human-in-the-loop ₃ became the default.

There is a phenomenon to be found in almost any spaced repetition software’s community: The drive to empower and encourage users to fiddle with all kinds of input parameters of the respective algorithms. Examples truly abound. The Anki Subreddit ₄ features extensive discussions about the settings in Anki’s similarly extensive Deck Options menu. Supermemo has a whole FAQ just about the intricate balancing act of setting the forgetting index ₅. Ebisu even requires you to set some fairly abstract initial model parameters (α, β, t) ₆.

I used to view all of these as flaws — after all, the second thing SR communities are obsessed with is finding the optimal SR algo ₇ ₈, and if humans are truly bad at anything, it’s numerical intuition. From this angle, every degree of freedom set by a human instead of optimized by a machine is a waste, a sign of hubris — or just technology being not quite there.

However, just maybe, some of these manual inputs fulfill a purpose other than filling the gaps the algorithm has left. From a cynical marketing perspective, it’s just the IKEA Effect ₉, but perhaps it truly is a step towards truly personal software ₁₀, an acknowledge of the Substitution Myth ₁₁, an embodiment of the idea that humans are more than spam in a can, even if they are just practicing for an exam.

I still want to have a better tool for calculating the ideal introduction rate for new learning items than friendly Redditor’s guesstimates, though.

This is a Mini-Essay in the spirit of learning in public. Feedback is truly welcome. Until next time!