Mnemonic search

[gusl]

What is the best way of memorizing something some information x?

For computers, this is the compression problem. We have the usual trade-off between space and time. At one extreme (ignoring time), the space occupied is the Kolmogorov Complexity of x (relative to the computer, and the memories it already has), written KC(x).

If humans are just a different kind of machine, the optimal encoding's length will be the KC relative to this machine. Can we make such a machine from ACT-R?

Since KC is incomputable, computable approximations are found through a universal program search: what is the shortest program we can find whose output is x? The analogous question for humans is: what is the simplest way of presenting this information?

If we do a "cognitive TM search", won't the output be a near-optimal tutoring program? Maybe not: unlike the case with computers, tutors need to worry about retention.


I have a more ambitious dream of automatically creating a tutor from a formal theory. One way is to encode these formal theories as ordinary memories.


formalizing my analogy (this needs to be refined):
human memory <-> computer memory
compressed data <-> axioms of the theory

Of course, people aren't meant to learn theories only as formal systems. They need to use them concretely too.


See Phil Pavlik's work on optimizing of memorization schedules.

Comments

[shaktool.livejournal.com]

An additional consideration for humans is that there are completely different ways to remember information, and they are useful for different reasons. For example, I had a locker combination last year. I can't remember what the numbers are, or which direction to turn the knob or how many times to go around before stopping at a number. But when I put my fingers on the lock, they do all the work for me, and they get it right most of the time. In this case, I used "muscle memory" to remember my locker combination, because the numbers weren't useful to me except for the physical action of opening a door.

The whole point of grade school mathematics classes isn't to teach you theories for you to recite. It's to teach you how to do math without needing to remember the theories. Being able to remember the theories may be a useful skill, but it's a separate skill. Any "theory tutors" you create would need to be specialized according to the purpose of the information you are tutoring.