Dear Hugo,
I appreciated your unconventional approach to the problem.
The question of when a system has not just goals or desires, but also "reasons for", seems particularly interesting. I think most would accept that the fitness function in an evolutionary model provides the (illusion of) a goal for a population, or at least a signal of differential goal-achievement.
But it is certainly the case that evolution doesn't give organisms "reasons for" doing something. They do them, or do not, and thrive accordingly. Your rational choice example of there being reasons-for is nice; yet it's not directly applicable because it's a model we build ourselves. We're using rational choice to determine reasons for doing something, but that's because we're interested in reasons. How does that reason-making itself arise? I think you punt a bit on this one, cramming it into the last two paragraphs.
Years ago I was at a lecture where speaker suggested that the neutron "calculated its mass every X nanoseconds" (or something). We (a roomful of physicists) were all mystified about what he meant--he was trying to make a too-cute point about how long it took *us* to calculate the mass using computers. In the end I think he was wrong: it is one thing to be X, or have property X; another to infer, calculate, or compute it. I think you make a similar slip in the last two paragraphs.
Perhaps I'm missing something here. You've taken an approach that reminds me of my data with the analytic philosophers. One of their great virtues, in addition to high alcohol tolerance, is to take a non-scientific idea like "having a reason for doing something" and just relentlessly refusing to cash it out in scientific terms.
Yours,
Simon