A few years ago, I'd have laughed at this. But AI may very well be good enough to make this a serious problem. The question will be, how certain is certain enough? Thirty five years ago I worked for an insurance company and we had some guys in who had a neural net program that identified accounts that were a bad bet with striking accuracy. Management said, thanks for showing us. Good luck, but we're not interested. Nobody questioned that it worked, but as a neural net, it couldn't give an audit trail explaining why, so you couldn't simply cancel policies without exposing the company to lawsuits for redlining and such. AI is infinitely more powerful now. I worked for a company a few years ago, and we noticed that people who purchased liquor in bottles with a D handle on a Friday were likely to default on their mortgages. Purchases on other days had no meaning. Regular bottle types, no meaning. It was D handle, on Friday. To me, the idea that AI could identify people who are highly likely to commit certain crimes, especially things like insurance fraud, is highly believable to me. It's not just the AI. It's the combination of AI and massive amounts of trivial data from purchases, surfing the web, how you handle the mouse and keyboard, how long you linger on certain images, etc.