Watch a lawyer scan through a contract or a copyeditor edit your essay -- same thing. They catch in a second something you'd never catch in hours and hours, if ever.
Machine learning is a model of this thing we observe in ourselves. Don't mistake a picture of a pipe for the pipe itself.
http://en.wikipedia.org/wiki/The_Treachery_of_Images http://en.wikipedia.org/wiki/Reification_(fallacy)
He said that experts doing that is "trained pattern matching gained through experience, i.e. machine learning."
But that's backwards: machine learning in this style is a model of something we observe in ourselves. The thing that's in us is the pipe, the machine learning algorithm is the picture.
What you said in the GP post stands true, it's important that beginners learn the correct patterns (look up documentation for idiomatic examples) and not false ones (blindly copy and pasting code).
You see this all the time in casinos; their goals is to reinforce poor pattern matching behavior. From roulette tables' results signs to bad poker players chasing terrible odds, the human mind has the ability to disproportionally increase the significance of selected memories.