In python we have a saying that we follow most of the time: if you don't know, refuse the temptation to guess.  So currently this is all working as designed: you have to know the encoding of the file you are trying to read as unicode.

Adding a 'guess' function that could be called explicitly is a possibility, but if we were to go that route we'd probably really want something general to guess the encoding of strings, such as (I think) ICU has.  This larger topic is a topic more suited to python-ideas, probably followed, if response is positive, by a PEP.

So I'm closing this issue as rejected, but feel free to bring it up on python-ideas.  (Search for existing threads about it first, please.)