Programming Herd Behavior

A topic that i expected to be big at the conference but didn't see at all was the problem of getting benefit out of data-mining vs. pushing herd behavior.

Will the users behavior be negatively influenced by early published results from data-minined peer-data?

Tim O'Reilly blogged about this in context of the inbreeding techrcruch-top-bloggers.

What are techniques to avoid this?

Where was the discussion about this?