It amuses me that people's imaginations always run towards retail discrimination when the technology clearly would be so much more cost effective for wholesale discrimination.

Think "credit scores". Except based off of much, much deeper data sets, and applicable to more fields than credit/insurance/employment.

I mean, I can tell with 80% accuracy what gender somebody is with 400 words they've written. Think of what could hypothetically be done with tying as much of their online persona I could possibly get to and then judging them on it.

Of course, you could get a new free email account and start fresh -- but then, that tells me something too, right? It will be just like a credit score: an absence of a credit score makes you look incredibly risky. Similarly, a few years from now, only reason someone in their early thirties won't have an online history is if they have something to hide. At the very least, that will be a good enough approximation to work with, given how stupidly effective the technology will be.

I mean, say I'm pricing car insurance for you. (Not for you, really. For a pool of a million people. But in the instant case, for you.) If I can make a determination of your risk algorithmically, even if I botch it some of the time, all I have to do is shave a fraction of a percent of accidents off my total and I save the company millions of dollars. How could I do that? Well, lets see whose Facebook accounts suggest that they routinely enjoy hard partying and alcohol. That might be worth a percent or two. And I won't be firing blind -- I'll have a hundred years of insurance claims, scores for people involved in 80% of the accidents last year, and a Hadoop cluster at my disposal to torture the data until it gives up its secrets.