Yes. I think having a good guess on capabilities vs. alignment requires a complete theory of how alignment happens so it's alignment-complete. But that doesn't mean you can't take guesses. But they're guesses so it would be quite hard to quantify them. But it might be worth the effort. Estimates would have to take in multiple theories of AGI and alignment and somehow sum across them; just using one theory of how we get to and align the first takeover-capable AI isn't realistic. Nobody knows for sure. This illustrates how hard it would be to even make a reasonable guess at dual-use fractions. This also illustrates how useful it would be to have a better sense of the likeliest routes to first TCAI.
Of course each such estimate would include a term for how fast those capabilities are pushed forward if you don't do this work. This is the biggest factor in people deciding to do dual-use work. They believe (and I agree in many cases) that capabilities work is likely to proceed at a rapid pace whether or not people are doing alignment work. I certainly think there's little virtue in claiming moral purity where the correct estimate is that doing dual-use work would probably improve our odds of succeeding at alignment by the time we reach takeover-capable AGI.
Of course you also have to watch closely for motivated reasoning driving you to overestimate that ratio so you can work on cool directions that might make you important/rich/famous.
Personally, I believe it would be helpful for the alignment community to somehow quantify how much of a given piece of research goes directly into alignment versus capabilities. But I have recently heard that this task might itself be an alignment-complete problem, which would mean it cannot be solved before the alignment problem itself.
I do not believe that (0.2), but I do not have many arguments. My position is that even though a large part of past progress came from "random" directions unrelated to the final solutions, we do have grantmakers, foundations, safety labs and many other orgs, as well as the personal intentions of the people who decide which research they will actually do, so there should be implicit or (if we do not believe in orgs) at least heuristic arguments available here.
Secondly, there are some existing thoughts on this: post1, post2. By comparison, these arguments personally do not seem persuasive to me: I can imagine cases where some compute research is useless for alignment, and I can imagine the reverse. Everything depends on how hard it is to convert one into the other and what actually will happen with its usage. Can we quantify that, or do we need aligned AGI first?