People are better at dealing with abstractions when those abstractions are described in terms of persons. The classic example of this is the Wason test, which people find hard until it is rephrased in terms of teenagers trying to sneak into a bar. Even you, my highly intelligent blog reader, will probably have to "think", if only momentarily, to solve the traditional Wason test, but will regard the solution to its bartender translation as instantly obvious.
[Quick sidebar on the Wason test: about half of the versions on the web state it incorrectly! They error by leaving out the "every card has a number on one side and a letter on the other" proviso. Without this, the correct answer is to turn over every card that doesn't have an even number on it.]
Social rules, including morality, are abstractions. The Wason test makes it seem reasonable to suppose that people, including you and I, will be more effective in reasoning about said social rules when they are put in terms of the desires of people. If, as is the case for many social rules, there is no real person who directly desires that you follow the rule (or more precisely, no real person whose desires you *currently* care about), why not make up an imaginary person who does?
This line of reasoning also has an evolutionary psychology reading. (Me #1: But what doesn't? Me #2: Quiet, you!) Here's the Robin Dunbar-esque Just-So Story: Humans start out living in medium sized groups and develop some mental machinery to reason about other people and their desires. People then apply this mental machinery to the desires of imaginary persons, allowing them to remember and reason with more complicated social rules, allowing them to live in even larger groups and eat more yummy fresh wooly mammoth. Later still, some additional (non-imaginary person dependent) mental machinery got layered on to deal explicitly with large group living.
All of this was inspired by Pascal Boyer's "Why is Relgion Natural?" article in the Skeptical Inquirer, in particular the section where he writes:
"Religious concepts do not change people's moral intuitions but frame these intuitions in terms that make them easier to think about. For instance, in most human groups supernatural agents are thought to be interested parties in people's interactions. Given this assumption, having the intuition that an action is wrong becomes having the expectation that a personalized agent disapproves of it. The social consequences of the latter way of representing the situation are much clearer to the agent, as they are handled by specialized mental systems for social interaction."