That is one of our main long-term goals. It's too early to say exactly how we would do that, but odds are there would be some sort of helper service in the cloud which relays and handles DNS and such. We need to get the client built first though.

And what would be your approach to spam-filtering? One of Google Mail's strengths is that it can use data from millions of accounts' emails to train its spam-filter.

Do you see Mailpile offering some way to do the same thing in the future, with clients flagging spam and the result being used to train a 'community' filter?

This is still largely an open question. But there are some interesting possibilities for those who actually make their Mailpile visible on the public web - Mailpiles could talk to each other and exchange data of this sort. But ideally things like this would be a community effort shared between projects and we would join, not necessarily lead.

If encrypted & signed mail becomes common enough, then I suspect the spam problem goes away. I suspect spammers won't want to sign their mail, and if they do, it would be trivial to distrust new public keys and maintain a global blacklist of bad established ones. Then almost all encrypted mail can be simply trusted, and spam filtering only has to pick out the occasional real message from the unencrypted dross.