I think the link goes like your first guess. Also, I find that one of the things I am most interested in learning about the people around me is whether they are disposed to respect my preferences. If I rely in their ability to read me—which I expect, for good reasons, to approach nil—then what they do isn’t informative about that disposition. They might be trying to do exactly what they think I prefer, and be annoying me because they have bad information, not because they choose to act at cross purposes. If I tell them what I want (and that I’m unreadable, etc. etc.), then their behavior becomes informative about their disposition to respect my preferences. Then, if they demonstrate that they have this disposition, I can choose to be around them more, and if they demonstrate that they don’t, I can avoid them and try to limit their influence on my life.
I agree with all of your numbered remarks and the summary at the end, except for a small caveat about (2). While it is true that MHW will articulate preferences that may not resemble the ones they reveal through behavior, and true that their articulated preferences are suspect, I think their revealed preferences are suspect too. The most obvious case of this is the known tendency of abuse victims to re-create the patterns that have been characteristic of their prior relationships. This looks like a revealed preference to be an abuse victim. I consider this little to no evidence in favor of the conclusion that the people exhibiting this behavior prefer to be abuse victims.
This leaves a bit of a muddle the question of where reliable information about common MHW preferences might be obtained. It looks sort of grim. You could try to extrapolate from people more likely to have accurate articulated preferences (like me), but the very factors that make me more likely to have accurate articulations probably also affect what it is that I prefer. (For instance, I prefer that people take my articulations at face value, which someone without good articulations might well not!)
Some avenues of possible investigation:
Try different ways of requesting preference articulation. (The question “what do you want?” is apt to get a cached list of social-circle-approved adjectival criteria. I think one might have better luck asking a MHW to say what her favorite scene in her favorite romance story is; or what three things she’d like to change about her current/most recent boyfriend; or an example of a couple she’s friends with being adorable/compatible/mutually supportive/something like that.)
Extrapolate from some subset of the MHW population likely to have especially… revealing… revealed preferences. (People with very high subjective happiness ratings; people with long-lasting relationships that don’t exhibit signs of abuse; there are probably other categories I’m not thinking of.)
Forget about learning MHW preferences and adopt conservative, general rulesets designed at minimizing risk to vulnerable people. (This is what I’ve advocated historically. I still think it has the best chance of avoiding the worst failure modes; but it is probably possible to do better on net.)
I think the link goes like your first guess. Also, I find that one of the things I am most interested in learning about the people around me is whether they are disposed to respect my preferences. If I rely in their ability to read me—which I expect, for good reasons, to approach nil—then what they do isn’t informative about that disposition. They might be trying to do exactly what they think I prefer, and be annoying me because they have bad information, not because they choose to act at cross purposes. If I tell them what I want (and that I’m unreadable, etc. etc.), then their behavior becomes informative about their disposition to respect my preferences. Then, if they demonstrate that they have this disposition, I can choose to be around them more, and if they demonstrate that they don’t, I can avoid them and try to limit their influence on my life.
I agree with all of your numbered remarks and the summary at the end, except for a small caveat about (2). While it is true that MHW will articulate preferences that may not resemble the ones they reveal through behavior, and true that their articulated preferences are suspect, I think their revealed preferences are suspect too. The most obvious case of this is the known tendency of abuse victims to re-create the patterns that have been characteristic of their prior relationships. This looks like a revealed preference to be an abuse victim. I consider this little to no evidence in favor of the conclusion that the people exhibiting this behavior prefer to be abuse victims.
This leaves a bit of a muddle the question of where reliable information about common MHW preferences might be obtained. It looks sort of grim. You could try to extrapolate from people more likely to have accurate articulated preferences (like me), but the very factors that make me more likely to have accurate articulations probably also affect what it is that I prefer. (For instance, I prefer that people take my articulations at face value, which someone without good articulations might well not!)
Some avenues of possible investigation:
Try different ways of requesting preference articulation. (The question “what do you want?” is apt to get a cached list of social-circle-approved adjectival criteria. I think one might have better luck asking a MHW to say what her favorite scene in her favorite romance story is; or what three things she’d like to change about her current/most recent boyfriend; or an example of a couple she’s friends with being adorable/compatible/mutually supportive/something like that.)
Extrapolate from some subset of the MHW population likely to have especially… revealing… revealed preferences. (People with very high subjective happiness ratings; people with long-lasting relationships that don’t exhibit signs of abuse; there are probably other categories I’m not thinking of.)
Forget about learning MHW preferences and adopt conservative, general rulesets designed at minimizing risk to vulnerable people. (This is what I’ve advocated historically. I still think it has the best chance of avoiding the worst failure modes; but it is probably possible to do better on net.)