Project link:  http://www.mozilla.org/projects/ml/autocomplete/

This project was an attempt at applying machine learning to the autocomplete
feature of the Mozilla browser (pre-Firefox).  The project entailed a frequency
module that would watch a user's browsing habits and when an autocomplete
was requested in the address bar, would show the matches that the user had 
visited both most frequently and most recently.  The implementation worked 
rather well from what I can gleam from the project page and Bugzilla entry, but
was never accepted into the mainstream Mozilla development branch.  With the
advent of Firefox 3, a different implementation offering the same results made
it's appearence in the "awesomebar" address bar.

While this project didn't make any waves, it does show how AI is important even
in the field of convenience.  Before the "awesomebar" which watched frequency
and dates of website visits, the autocomplete feature of both Mozilla and 
Firefox were rather useless; if the result you desired was at the end of the
match list, it was often quicker to just type out the address rather than 
scroll.  The idea of AI allowing a program's interface to adapt to the user is
one that I never thought of until I stumbled upon this project, and now that I
think about it, Google's Autocomplete and Instant features make a lot more 
sense to me.