It's also subject to an unhealthy feedback loop. Oh, this button isn't commonly used, so let's move it to a slightly less prominent place. Oh, this button's usage dropped, it must be super unimportant, let's move it behind a menu. Oh, nobody ever presses this button, let's get rid of it.
I get this a lot with URL autocomplete. For some reason it decides “news” means “some other URL with news in its <title>”, then out of habit I type news+enter and it takes me to that website, further reinforcing “Yeah he goes to this website a lot, let’s keep that at the top autocomplete priority!”
So now I have to type “news.” before it will fill news.ycombinator.com, because it started showing me the wrong option and I kept inadvertently picking it because it used to give me the right one.
There doesn’t seem to be any mechanism for it to realize “every time this happens he closes that website immediately and goes to a different one that we ranked #2 in the autocomplete search.” So until I retrain my muscle memory I’ll just keep on mistraining the autocomplete even further.
> “every time this happens he closes that website immediately and goes to a different one that we ranked #2 in the autocomplete search.”
This is a disheartening yet frighteningly common problem with UI automation. It doesn't learn enough to actually be aware of the expected result, yet it can't be programmed to behave exactly as you want or need.
Automation in user interfaces should be exactly the opposite. It should try to learn your repeated actions so that you can perform them automatically, yet if the learning goes wrong, the user should be in control at all times and override the automatic deduced actions.
