I haven't had a chance to test it much, but I think I have an idea. Frame the question as what will happen if I do x that won't happen if I do y. At this point it seems like it should be possible to reuse the mental processes for making beliefs pay rent. Basically typecast "I do x" and "I do y" to beliefs. Then see what experiences I anticipate as a result of the beliefs "I do x" and "I do y". Then determine which set of anticipated experiences has higher utility.