Mod this guy up. As someone from a field who tries to separate signal from noise and develop predictive models on a daily basis (in supposedly well-controlled conditions) I can say that they have their work cut out for them.
I mean, I use proven methodologies in clear-cut and well-designed experiments and end up with data that is extremely difficult to manipulate (genome wide association studies). These guys/gals are trying to observe millions of humans interacting in indiosyncratic and complex ways with millions of input-points, and they think they can use that data? Talk about multiple-testing correction. Bonferroni is the tip of the iceberg in such a data-set.
The scary thing is, if you set something like this up, you will get "answers". It might be the result of a random walk, but who in the "jury of your peers" is going to understand that defense?
"But your honor, they didn't even define an acceptable false discovery rate!"