In addition, it's not like leaving a PhD funded by an NSF GRFP leaves zero net positive impact. Funding research leads to papers (hopefully with valuable insights) and to knowledge and training passed down to other researchers. I don't think it's a very good analogy to startup funding in that sense. (Although I suppose one could make the argument that alums of failed companies have hard-earned experience so some of the value is still there, diffused out to wherever they go next.)
In the best case I think that's possible, but in the usual case someone who leaves a PhD program after a year does very little research. US-style PhDs don't typically require a master's degree first, so first-year PhD students are doing mainly course work and "figure out what I want to do for my PhD" project-shopping. If you do that for a year and bail, it's pretty common that you wrote zero papers and did very little research. If you leave after two or three years, it's more likely you will leave some kind of research contribution, although it's also pretty common for people who realize they are going to leave to just "pivot" to full-time MSc coursework, so they can leave the program with a master's degree. In that case, the NSF basically funded someone to take MSc courses, which is arguably not useless, but isn't what the NSF intends to fund.
I think with this amount of money it's just a risk of funding people. Putting in strings requiring repayment if the person changes areas, when you're giving someone <$100k, is pretty heavyweight for either the NSF or Y Combinator, so they just try to pre-screen for people they think are going to use the money as intended and stick with it, and accept that some percentage of people will instead use the money to flail around for a year or two and leave. If the percentage gets too high, then I guess reconsider the screening process.
Yes, that's true. For full disclosure, I left a PhD after 4 years to work in industry (I was frustrated by some aspects of academia), having spent three years on an NSF fellowship, so I may be biased here :-) I absolutely didn't go into it thinking I would do that, and I try to bring a researchy perspective everywhere I go now. I also think it depends on the school and the field. Where I was, it was common for first-years to get on a project right away and help churn out results. Probably guided closely by older students or advisors, but still doing productive work.
I think you're right re: clawbacks, though. I can't imagine the mental pressure were the scenario to be "you MUST stay in academia or you owe the government $XXXk". I actually know a PhD student on a (non-US) government scholarship who has a clause like that. He calls it his "slave contract".
Getting back to the original topic, I think that for any venture with a high rate of failure (startups or grad school or...), there has to be some room to allow the funded person/people to say "this isn't working" in good faith and pivot or leave. That's totally different from someone taking funding without the intention to use it properly, or just deciding they're bored or whatnot.
