Multiagent Systems

Today there is a NYT Times article on prediction markets and an interesting rebuttal by The Big Picture blog. The Big Picture presents the commonsense view that prediction markets can, at best, aggregate the knowledge of their participants. That is why they are good at prediction who will win an election, since the people participating are also voting and reading the polls, while they are bad at predicting events for which they have no information, such as the jury results in the Michael Jackson trial.

Thus, the trick for us multiagent researchers is to build systems that can identify agents with knowledge and give more weight to their opinion. This is what science does. But, how do we implement that?