Machine learning for better default prefs?

Bug 182366 is for using machine learning to order the autocomplete
results. The same concept can be applied to prefs.

A lot of Mozilla prefs use what I consider to be the wrong defaults.
The reasons could be many, but with this technology we can get real
evidence as to what users actually use/prefer.

Is there an open RFE? has this idea been discussed in the past?

Prog.