This ! And I really like your analogy to computability theory. I wonder if it applies to all economic activities. My experience so far is that verification is nice for simple knowledge tasks. But for the complex ones, doing it from scratch is often better. But I guess this may change as AI systems become better and better.
Of course, brute forcing a solution by verifying a massive amount of garbage is much less efficient than doing it from scratch.
Verification becomes more productive than sythesis only when the probability of generated content being correct is significantly high. So I suppose, as AI gets more accurate, verification will become more productive.
I'm in a peculiar position with Bing AI. My work provided me an enterprise account, so I can use it for work-related tasks. Often I find myself asking it to write a program for something I don't wanna do myself from scratch, then I just analyze it, see why it doesn't work and fix it. It has got to a point where it's much faster for me to do that, than to read the relevant documentation and figure out how to do it from scratch.
