I'm getting really annoyed by how specific certain targeted ads are right now. I hung out with an old friend of mine recently and he brought up the topic of P vs NP, which is a computational problem. It wasn't anything I ever really looked at before, but today I have a youtube video in my recommendations that's all about P vs NP, and I can only assume my friend had something to do with it.
Now I'm not crazy, I don't think phones are eavesdropping on our conversations and feeding that data to google, but what I believe is happening is that my friend is googling these kinds of topics on his phone, and then when we hang out, the location services noticed our phones being relatively close to each other and assumed that we were together for a while, and therefore might have conversed about it.
If you don't believe me, I should mention that you don't have to consciously program a machine to do something like that. Google's algorithm uses a neural network. Neural network's are designed to find correlations between sets of data, so if you feed stuff like search history and location data to it, it will inevitably notice the I mentioned pattern above if it's programmed when any degree of competency regardless of the designers intent. It's pretty easy to express something like this in psuedocode too.
phonehistory=x
phonelongitude=x
phonelattitude=y
get otherphonelongitude
get otherphonelatitude
if absolutevalue(phonelongitude - otherphonelongitude) < 0.001 and absolutevalue(phonelattitude - otherphonelattitude) < 0.001:
increaseconfidence(otherphone,x)
Anyway, neural networks don't work at all like the way I mentioned above, and I didn't include something that kept track of how long those two phones are near eachother. But I hope you get my point. I just don't like it.