At some point we are going to have a world where computers have become better at solving CAPTCHA images than humans are, on average. Because CAPTCHAs are used to keep spam bots and other malicious software out of places that have value to them, it is therefore a lucrative problem to solve. If you have the spam bot that can solve CAPTCHAs to get into discussion boards and website comment sections where others can't, then your spam bot has more value. Even tiny incremental advantages are big bucks to spammers, so they will keep fighting for every minute bit of progress they can make on this problem. Without entirely meaning to, we have incentivised the world's spammers to solve optical character recognition for us.
Perhaps what we need to do to solve the world's biggest AI problems is to start using them as gateways to spammers. Motivate the world's criminals to start solving things like image recognition, for example. "Which one of these is the elephant?" with four pictures to click on. Whatever difficult computer problems we have, we should use those as the standard CAPTCHA because, while they work, that's their advantage, but when they start to fail, we get better AI software.
Mokalus of Borg
PS - It's kind of a win-win, but in the form of an arms race.
PPS - And if you stop racing, the internet dies.