zlacker

> almost any data point can be predictive of almost any behaviour.

This isn't really true, though, even in theory. In practice it's much worse, as we aren't very good at using even large collections of data points in most cases to predict behavior. We can in some very few, narrow contexts occasionally do better (sometimes much better) than a random guess once in while.

>>sidlls+(OP)
I didn't mean it quite literally, and specified that I'm not suggesting that police had something like this in mind. My uneducated guess is that police were after something specific relating to other information that they already had. That's generally how police investigations work.

That said, the one relatively broad concept where "we" consistently do better than random guesses is ad-tech, which is where the bulk of private efforts to this effect are concentrated currently. The premise here is quite literally "every data point is predictive of behaviour," behaviour being stuff related to the goals of advertisers.

It's not a huge leap to suggest that fb & adwords' system can be used to predict crime, insurance claims, HR-related stuff etc.

>>sidlls+(OP)
We may not be good at this, but the collective wisdom at most corporations states that a best predictor of what someone will do, is what they have done in the past. My point is that it is being used and inferences are drawn regardless.