In my paper, “Being Familiar with What One Wants”, I argue that there are two kinds of self-ascription of desire. First, there are easy cases where a sincere self-ascription seems to be immediately expressive of self-knowledge. For example, if I believe that I want to eat ice cream then, given the person I am, my self-ascription is true and there is no room for doubt. Second, there are hard cases which lack this kind of immediacy and where one could have easily been wrong about one’s self-ascription. For example, when I believe that I want to have a child then, given the person I am, this self-ascription is not immediately expressive of self-knowledge and can be put under question.
How to explain the difference between easy and hard cases? In particular, what makes a self-ascription of a desire an easy case? Since in the philosophical literature on self-knowledge of desire this question has not been asked, there are no clear guidelines how to answer it. Whichever proposed method of coming to know what one wants we consider it is not sensitive to the difference between easy and hard cases. This is not so much a criticism of any existing theory, but an indication that there is more work to be done.
In my paper, I propose a solution to this problem. My explanation of the difference between easy and hard cases proceeds from the assumption that our self-ascriptions in easy cases are safe self-ascriptions, i.e., they could not have easily been false. For our self-ascriptions of desire to be safe, they must track the constraints under which our desires are formed. What are those constraints? I take it that the generation of a desire is due to the updating of prior reward values of one’s options and these prior values were based on the agents’ past experiences. A self-ascription of desire thus tracks those constraints and is thereby safe only if it properly sensitive to those past experiences.
What does it take for a self-ascription of a desire for x to be properly sensitive to the relevant past experiences? My proposal is that there should be some previous experience with a rewarding option that was similar to x. For instance, I can be sure that I have a desire to eat ice cream because I have had comparable experiences before, and this ensures that I have the self-ascribed desire that is shaped by my actual learning history, and thereby excluding the close possibility of being wrong about my self-ascription. On the other hand, I am not in a position to exclude the possibility that it is false that I want to have a child because I have not experienced as rewarding anything that is relevantly similar to the experience of having a child. It is therefore a close possibility that I have a learning history which did not give rise to a desire to have a child.
The difference between easy and hard cases can thus be defined in terms of familiarity: in easy cases, the agent has experienced some content that is relevantly similar to x; in hard cases, she has not.