If you're really interested, I have a vague recollection of a paper by David Wolpert that feels similar. See his vitae below. The physics and computation section should be of particular interest.
http://www.santafe.edu/media/staff_cvs/3cv.complete.fall.2010.pdf
Interesting fact in his vitae:
Top two winners of 2009 Netflix competition made extensive use of my patented Stacked Generalization technique.
What do you think? There might be a theoritical limitation to how much data an AI could collect without influencing the data itself and making its prediction redundant. Would this negate the idea of a 'God' AIand cause it to make suboptimal choices even with near limitless processing power?