And although I'm using the multiple selves / sub-agents terminology, I think it's really just a rhetorical device. There are not multiple selves in any real sense.
I would actually dispute this, but that goes into what you actually mean by a "self". I don't see how it's not obvious that are multiple agents at work; the problem of akrasia is, then, trying to decide which agent actually gets to pilot your brain at that instant. I suspect this is alleviated, to some extent, by increased self-awareness; if you can pick out modes of thought that you don't actually want to "endorse" (like the "I want to be a physicist" versus "I don't want to do physics" example below), you are probably more likely to have the ability to override what you label as "not endorsed" than if you are actually sitting there wondering "wait, is this what I really think? Which mode is me?"
I agree that knowing which is the real me is the first step, and I propose that it has a simple answer: the me not under the influence of immediate consequences. So then the next question is how to make sure that me's decisions are the ones that stick.
If we agree on that much then I suppose the question of whether "multiple selves" are real vs a rhetorical device is moot.
A year and a half ago I wrote a LessWrong post on anti-akrasia that generated some great discussion. Here's an extended version of that post: messymatters.com/akrasia
And here's an abstract:
The key to beating akrasia (i.e., procrastination, addiction, and other self-defeating behavior) is constraining your future self -- removing your ability to make decisions under the influence of immediate consequences. When a decision involves some consequences that are immediate and some that are distant, humans irrationally (no amount of future discounting can account for it) over-weight the immediate consequences. To be rational you need to make the decision at a time when all the consequences are distant. And to make your future self actually stick to that decision, you need to enter into a binding commitment. Ironically, you can do that by imposing an immediate penalty, by making the distant consequences immediate. Now your impulsive future self will make the decision with all the consequences immediate and presumably make the same decision as your dispassionate current self who makes the decision when all the consequences are distant. I argue that real-world commitment devices, even the popular stickK.com, don't fully achieve this and I introduce Beeminder as a tool that does.
(Also related is this LessWrong post from last month, though I disagree with the second half of it.)
My new claim is that akrasia is simply irrationality in the face of immediate consequences. It's not about willpower nor is it about a compromise between multiple selves. Your true self is the one that is deciding what to do when all the consequences are distant. To beat akrasia, make sure that's the self that's calling the shots.
And although I'm using the multiple selves / sub-agents terminology, I think it's really just a rhetorical device. There are not multiple selves in any real sense. It's just the one true you whose decision-making is sometimes distorted in the presence of immediate consequences, which act like a drug.