Thanks to the pioneering efforts of among others Paul Graham (with the now classic A Plan for Spam) bayesian filters have reached the mainstream, to the point of AOL implementing bayesian filters (annoyingly Hotmail hasn't yet and there are known holes in their filtering approach - so Hotmail should be avoided even as an MSN messenger identifier).
Graham has identified what could be the next frontier in anti-spam, namely raising the cost of spam by fighting back: Hitting the advertised websites with traffic proportional to the number of emails they send out. When spam success rates are low enough this will raise the cost per sale significantly and could be an efficient way to stop the sending of spam, not just filtering it.

Details in "Filters that Fight Back".

The possible problem if the approach becomes standard is of course that rabid anti-spam activists will be able to abuse this notion in the way they are abusing the open relay databases today. The idea amounts to making DoS attacks common and accepted practice. In contrast to spam filtering, this is close to vigilante justice, and I am never in favour of that - the annoyance of spam notwithstanding.