What I was really trying to get at is the re-harnessing of the PoW compute power to solve an interesting problem. Still with the idea of leaving humans out of the loop -- so not involving humans as screeners, but using the PoW concept to address ongoing problems that are controlled via API calls to/from other computers.
The compute problem could most certainly be distributed, so its not a central/single API server. But, the certification of the blocks would not then be simply a race without a compute meaning, it would imply that each mining event is helping to solve a problem.
I'd suggest the problem be a computation related to science, and we have some candidates in mind, but it doesn't have to be a particular domain of knowledge, just something better than a wasted set of CPU cycles. It was for that reason that I titled it with the reCAPTCHA idea, but maybe it was misleading. I do agree that the current PoW is similar to reCAPTCHA, as a check on validity, but I would argue that its not like reCAPTCHA in that the computation is not harvested to help with some other problem that needs a solution (e.g. book translation or image ML).
Ah, got it. There have been a couple of alts that are (were?) trying to go this route. Two which I remember are Primecoin and FoldingCoin. I'm sure there are other examples.
One of the problems to be aware of is that the PoW that gets put into eg. the Bitcoin blockchain is not wasted, it is used to secure the transactions. Using said computing power for other means always comes at the cost of security, ie. cycles that get used for research can't be used to secure the network and vice versa. However there might be recent research that tells us otherwise. Maybe the aforementioned alts can serve as a starting point.