TheMadScientist - 12:31 am on Jan 7, 2013 (gmt 0)

To 'make a machine learn' you have to show both the good and the bad results to individual users (or segments of users) so it can tell the difference. Then, visitor behavior would dictate what 'stays' for one user (or group of users) and what doesn't for another user (or group of users) ... The algo itself would have to be 'flexible' based on the behavior of individual users (or groups of users).

IOW: It might not look 'promising' to webmasters for specific queries on specific days, but the algo would have to go through 'those phases' to get where it's going and 'drill down' on 'right answers' for individual users (or user groups) ... There's no way it would be 'right' from the start ... But, I also wouldn't be surprised if Panda is the base of the 'new algo' they're moving to and adjusts it itself based on individual user (or user group) behavior in the not-too-distant future.