That's not the near-term scary scenario. The near-term scary scenario is a machine learning system that's better than most CEOs.

What happens when someone develops a system that measurably makes more profitable decisions than 70% of CEOs? That isn't totally out of reach. Capital would flow to companies run by such systems. We could end up with machine learning systems running the corporate world, based purely on better ROI.

A good start on the problem would be to train on a collection of funding proposals, such as old Kickstarter announcements. Look deeper than the hype. Go out and suck up all the information you can on the founders, using credit databases. (They're asking for money, so it's a legit credit query). Do an automated background check. Then check on how the project did, five years later. That's your training set.