TheMadScientist - 4:37 am on May 14, 2010 (gmt 0)
If the system 'learns' like e-mail spam filters do, then it has to 'see' the reactions of visitors to spam to 'learn' anything about spam, doesn't it? IMO It's not a stretch at all to need to 'let the system run' for a while when you first fire up an 'AI type' system like tedster was thinking they might be shifting to.
Good e-mail filters get better over time don't they?
If this shift is to the same type of system IMO they might have to just let it run for a while...
If you think about what they could automate by using a system like this then IMO the changes could be a bit more massive than just installing a new infrastructure... Couldn't they theoretically automate the algo doing the scoring as well? Think about personalization, e-mail spam filtering, an incredible amount of computing power and a whole bunch of really smart people and what they might automate WRT search results, much like some e-mail spam filtering systems. That could well be where they're going, IMO, and would require the letting through of some spam, because unlike e-mail you don't have a spam box for people to check and see if anything got incorrectly filtered from the SERPs.
I guess another version of what I'm trying to say is:
It could be necessary to take what seems like a step or two back in order to make a huge leap forward, couldn't it?