"I Know Kung Fu"

First, that scene is a huge part of what makes the first Matrix fun. Thanks for reminding me of it.

Reading this reminded me of something that I've thought of a number of times but don't have fully worked out. (I.e., bear with me for a minute if I ramble.) We need a better way to describe the parts of (many) skills that lie between 'thought' and 'muscle memory'. (Maybe some researchers already have better ways?) The only sport that I was ever expert at was pool. For many years in my 20s, I played at least twice a day for anywhere from 1 hour to 3 hours each time. (I was in grad school. First stretch was post lunch, usually shorter. Second stretch was out at bars post evening work. How long depended on how many people I beat.) Eventually I reached the point where I would shoot, pick up the chalk, rechalk the cue slightly and already be walking to my next shot. I didn't really think about all this and it wasn't exactly muscle memory (as the grip, stance or swing is muscle memory). But I knew where the cue ball would end up, and I knew what my next shot (and ideally, two, three or more forward) needed to be and I would move towards that spot intuitively.

Wayne Gretzky has the famous quotation: "A good hockey player plays where the puck is. A great hockey player plays where the puck is going to be." That's the thing I wish I could describe better. It's not pure thought (it's barely conscious thought at all, usually), but it's also not physical training only. It's both - combined with a very clear mental picture of the domain you're dealing with. (Cf. what people always say about Larry Bird "seeing the whole court" at all times.)

Ok, I'm guessing the purpose of the post is exploring efficient methods of learning quicker.

Personally, I say tons of examples and exercises to test that you understood said examples.

For instance, someone gives you an api, but only gives you an explanation of the parameters. It is left to you now to figure out how exactly this api works(Does it take the full file path?, Is it the printer ip or the printer name? Is the array associative?) Now if I wanted to figure it out, I would draw from my experience how things like this normally works. Normally, it is two or three choices for every parameter and you fiddle with it until you figure out the correct format for what you are doing. So, with 4 parameters that is like 16 combinations, which aint so bad.

Now, if the creator of the api, had given an example of each unique way of using his api, learning the api would be much easier. Since now instead of debating what each parameter means and how to supply the given format, he can simply copy the example format[O(n)vs O(2^n)]. Learning without examples is an exponential function. Learning with examples is a linear function and learning matrix style is constant time.

[Did I just reduce all learning activities to be a function that takes parameters and to efficiently learn is to find an efficient algorithm to figuring out the best parameters?]