The perfect email


While I was reading Jon Udell's column "Working with Bayesian Categorizers", I had a thought. Once you've trained your spam filters and blog filters to recognise the items that are most important to you, while throwing out the junk, you could then write a counterpart program that would attempt to generate (probably using some kind of evolutionary algorithm) your perfect email. The email that you'd most want to receive, which would be the perfect synthesis of the items you've already collected, and which would achieve the highest possible score when run against the filters.
One problem could be that these Bayesian filters (I think) only analyse word, rather than phrase, frequency, so a block of text with nonsensical grammar could still score too highly.