[ben_w]

Incentives can work, but most governments and businesses are still only mediocre at them even with enough money to throw at the problem they get to do do-overs when they get it wrong.

Trying to do this with humans on a big scale combines the worst of software development in the days of punched cards, working without anyone having given you a formal language spec, and black-hat hackers on the modern internet.

It is very very easy to pick your incentives badly; you only get feedback on a very slow cycle (in the punched card days you might run the program overnight and only find it crashed on line 32 from a typo the next morning, but it's much slower than that in meatspace); and you also need constant fine-tuning as people interested in gaming the system share their methods for doing so.