It seems to me to be impossible to document why you do something if you do not say what you do first. what is especially important if you use a non-trivial algorithm. Of course you do need to explain why you use that specific algorithm (speed? accuracy? etc).
Even if the why is not important, the what can be. In my experience it is much faster to read 1 line of English comment than 20 lines of code, even if the code is very understandable.
Besides, if there is no comment explaining what a piece of code should do, how can you determine whether it is doing it correctly? Indeed you might have the problem of outdated comments, but then you can at least make sure that at one of them (comment or code) is incorrect (or both). W/o the comment, you have no clue to what your predecessor meant to do.