I think there might be some confusion over terms here. I don’t think “pattern matching” is the best way to phrase this.
Musk seems to be arguing for “rule learning” (figuring out the underlying rule) as opposed to “example learning” (interpolating to the nearest example in your collection). In the book Make it Stick, the authors mention that rule learners tend to be better learners. (These terms come from the psychological literature.)
I don’t think this observation is incompatible with the importance of recognizing patterns. You need to “pattern match” which rule to invoke. You also need to recognize the pattern that is the rule in the first place. Recognizing which examples to use also could be pattern matching, too, so this is why I don’t think the term is right.
In the same book mentioned previously, the authors write about Kahneman’s systems 1 and 2, and I got the impression that mastery often is moving things from system 2 (more careful reasoning) to system 1 (automatic pattern matching, which might simply be precomputed). Here’s an example: Vaniver suggested to me before that (if I recall correctly) when playing chess, someone might not explicitly consider a certain number of moves; their brain just has a map that goes from the current state of the board and other information to their next move. Developing this ability requires recognizing the right patterns in the game, which could come from simply having a large library of examples to interpolate from, or whatnot. This is precisely what I thought of when I read that it took (the famous) 10,000 hours for JonahSinick to see the patterns.
(To be fair, you do need both, but it seems that if you can develop good rules, you should use them. Also, developing accurate intuition is useful, whether it uses explicit rules or not.)
I think there might be some confusion over terms here. I don’t think “pattern matching” is the best way to phrase this.
Musk seems to be arguing for “rule learning” (figuring out the underlying rule) as opposed to “example learning” (interpolating to the nearest example in your collection). In the book Make it Stick, the authors mention that rule learners tend to be better learners. (These terms come from the psychological literature.)
I don’t think this observation is incompatible with the importance of recognizing patterns. You need to “pattern match” which rule to invoke. You also need to recognize the pattern that is the rule in the first place. Recognizing which examples to use also could be pattern matching, too, so this is why I don’t think the term is right.
In the same book mentioned previously, the authors write about Kahneman’s systems 1 and 2, and I got the impression that mastery often is moving things from system 2 (more careful reasoning) to system 1 (automatic pattern matching, which might simply be precomputed). Here’s an example: Vaniver suggested to me before that (if I recall correctly) when playing chess, someone might not explicitly consider a certain number of moves; their brain just has a map that goes from the current state of the board and other information to their next move. Developing this ability requires recognizing the right patterns in the game, which could come from simply having a large library of examples to interpolate from, or whatnot. This is precisely what I thought of when I read that it took (the famous) 10,000 hours for JonahSinick to see the patterns.
(To be fair, you do need both, but it seems that if you can develop good rules, you should use them. Also, developing accurate intuition is useful, whether it uses explicit rules or not.)