The way of data collecting is basically invalid. Being watched will definitely change the behavior, especially if one knows the study is about usage of auto pilot. More importantly, agressive or reckless drivers who use auto pilot without paying attention to road would not volunteer to such study.

I believe the potential of data mining from social network. 'facebook" doesn't recommend the right thing, because their algorithm is not good. It's almost impossible for the fake users to get much follow from real human users. The persons who likes an application should looks quite even in the big graph, while the robots should be quite isolated from the social network of human users. Essentially, we need some "pagerank" for social network, rather than the simple "1+1=2". With a good algorithm, the spammer can affect their own spammer social network, but not much on the others. We don't really know much about Netflix. They don't really use the algorithm from the 1 minion dollor competition. As they found the information from the new internet business, is different from the traditional renting market.