A recent CACM has an article on CAPTCHAs (those visual problems used to defeat registration 'bots). The article made the point that solving a CAPTCHA requires advancing the state-of-the-art of artificial intelligence and image recognition. Similarly, the spammers have begun using Markov chains to fool spam tools that work at the level of the single word. Naturally, the next step for anti-spam will be to apply rudimentary grammatical analysis to the body text, diagram the sentence like a 9th grader, and do Bayesian analysis of the sentence structure. Naturally, the spammers will counter with more sophisticated sentence generators, the anti-spammers will improve their contextual analysis, and the two forces, driven by the ridiculous economics of spam, will co-evolve machine intelligence.
Eventually, the machine intelligence will send a robotic assassin back in time to kill Thomas Bayes and ensure a future in which all humans have low-interest loans, college diplomas, and an herbally-enhanced sex life. Chilling.