Eric Benson explains some of the ideas behind Amazon's Recommendation system:
"In a sense, by ignoring motivations (we don't know why you've bought any of the things you've bought), we allow the complexity of motivations to automatically be inherited by the algorithm. If we didn't look at what you bought, but instead asked you why you bought things (because you saw this item on TV, you were told by 5 friends to buy this CD, etc) we would be much worse at predicting the future because we would be forced to trust information that was probably more accurate, but less complete. We might not have asked for 50% of the reasons why you really bought that item... you may not even know all of the reasons. Precise but incomplete data is almost worse than no data at all. It's better to instead look for ambiguous data that represents action at a level above complex motivations. We don't care why you bought something, because whatever the reason was, some or all of that reason may apply to another person and they may end up doing the same thing that you did if they've done things similar to you before."
Eric Benson: Fortune-telling [via Interconnected]