Gary Watson and Irving Thalberg both noted this problem many years ago--in the 1970s. (A very cool decade, by the way.) But it has been ignored or, in my view, underappreciated, in favor of the "regress problem" and "manipulation worries"--admittedly, big problems for the hierarchical picture of agency. Here's the issue. On the hierarchical model, in particular, Frankfurt's, it seems that deliberation or practical reasoning is conceptualized as about one's own mental states, rather than courses of action in the world. But this just seems distorted or in any case inaccurate as a picture of what's going on in practical reasoning.
So, I have a first-order desire for that (tantalizing) second martini, and a first-order desire to resist the temptation and not have a second martini (the first was great, but surely enough!). Now Frankfurt conceptualizes my practical reasoning as focussed in the first instance on MYSELF, i.e., on my first-order desires. My deliberation is about which of my first-order desires should lead me to action. But this just seems inaccurate and, well, somewhat narcissistic. Isn't it more accurate in general (maybe there are special cases) to describe me as deliberating in the first instance about courses of action, and not my mental states?
As I wrote above, Gary Watson and Irving Thalberg noted this early on. It seems to me to be very problematic for the hierarchical theorist, and yet the problem gets put to the side. Only a few people have wrestled with it, including Michael Bratman, but I'm not sure anyone has said anything desively to assuage the worry.
What do you think? Is this a fatal problem for a hierarchical theorist? Why is so much attention paid to the regress problem and manipulation problems, and so little attention to weakness of the will and the apparently distorted picture of practical reasoning?? And, as a bit of a plug for my own preferred model of agency, shouldn't all of these issues make a cumulative case to look elsewhere for a plausible model of agency, free will, and responsibility--such as a reasons-responsiveness model? (Well, don't I get to do that at least ONCE this week??)