Way back in '95, those wacky and ever-inventive Extropians created a web-based implementation of an idea that was apparently first presented in a 1991 paper by Robin Hanson, something they called "idea futures." The general concept is that people bet on the likelihood of a given predicted event happening, and pure market dynamics are supposed to result in a reasonably good "expert consensus" of that likelihood.
The first web implementation was known simply as "Idea Futures"; it's since transformed into the Foresight Exchange, and appears to still be going strong after about eight years. I played the original version briefly, but got distracted before I could become too immersed in it. One thing I recall about it was that the exact phrasing and details of a given claim/prediction made a big difference; a claim titled Human Clone before 2005, for example, specifies that the clone must live to one year of age, which might have a big effect on how likely one thinks the claim is. I rather liked IF for the brief time I played it, partly because it gave me a slightly better idea of the basics of how a stock market worked, which I'd been previously entirely ignorant of.
I'm not sure whether Jay is still an active FX player (he was the one who introduced me to the concept, iIrc), but it appears that his autonomous robot player, Markybot, is still playing.
Anyway, the reason all this comes to mind is that Fred points out that the Pentagon has become interested in the concept; at a cost of only $3 million, they intend to create a global idea futures exchange called the Policy Analysis Market to get advance warning on acts of terrorism and such. In the Bloomberg article, DARPA is quoted as saying: "Prices and spreads signal probabilities and confidence.... Since markets provide incentives for good judgment and self-selection, the market will effectively aggregate information among knowledgeable participants."
I'm still a little unclear about how to get from a market-based consensus opinion on the likelihood of an event to "market-based techniques for avoiding surprise and predicting future events"; it seems unlikely to me that the people with any real knowledge of details will engage in trading in this market, which suggests to me that this is basically a high-tech way of taking a poll. But there may well be lots of details I'm unaware of. DARPA says that this approach has been successful at predicting things like movie box office success and the outcomes of elections. (Though again I would note that in those situations, people who participate in the market may also be involved in determining the outcome, such as by voting in an election or attending (or staying home from) a movie. So this again starts to look like a sophisticated polling technique, which is a lot less useful when the people who are going to take the relevant actions aren't in your polling sample.)