A Kaleberg
Aug 25, 2017 · 3 min read

There is a huge difference between being able to answer a bunch of questions by rote and actually understanding something. Rote human learning is like machine learning. If the computer has seen or heard something similar before, it does great. If it is something a bit off, there’s a good chance of getting total nonsense. It get be worse too. Google “adversarial machine learning”. Neural networks with huge training sets are great, except when they are not, and there is no way they can fix themselves. Like a student trained only by rote, they have no model of what is going on.

On the other hand, I have to stick up for rote learning. I know a lot of creative people and over the years, I’ve even met a few geniuses, and one they have that other people don’t is a command of their area of expertise. They are full of stuff that they learned by sheer repetition or brute force memorization. This includes musicians, computer scientists, auto mechanics, cooks, novelists, physicists and a host of others.

No one gets really good at music without countless hours, and a lot of it is scales, chords and the same piece over and over again. Then, when they improvise or compose, they have all the scales, the chords, the sequences right there at their fingertips. Physicists cram for their orals where they may be asked questions about everything from freshman mechanics to black hole tensors, but when they actually have to do something creative, they have a big box of Legos to use for analogies and insight.

Feynman wasn’t all that fond of rote learning, but he crammed for his orals like everyone else in his class. Maybe it was a bit easier for him than for others, but that’s why he was so good at coming up with analogies. Those Feynman diagrams represent summations, and they didn’t add up themselves. Fermi, one of Feynman’s peers, was famous for being able to derive just about anything from basic principles, but he did that using his gigantic barrel full of basic principles and mathematical techniques.

I think that people who forget everything right after the test are cheating themselves. The only way to actually develop a model of something is to study it again and again from different viewpoints. Fractions might be slices of pie or deferred divisions or members of the set of rationals. They might be physical, numeric, algebraic or purely symbolic. If one forgets what one learns right after the test, there is no way to integrate the various views. Even the blind man with an elephant could compare notes with the other guys. That’s because they didn’t forget everything right after the test.

Our civilization relies on our expertise. Being able to survive in the world requires knowing something about it. Knowing more makes it is easier to survive. Our planet would have less than half its human population if some guy hadn’t developed the expertise to react hydrogen and nitrogen. We could have a lot of funerals just by dropping one piece of knowledge, and our civilization is based on a library. Maybe the argument is that our species would be better off ignorant and with a much smaller population struggling against much greater odds for survival. It sure beats being reduced to a single pale dot of productivity, except when one gets hungry or thirsty or sick or attacked by an angry dog.

)