I'd prefer alignment to be an after-the-fact statistic. Nobody is born with one. What one does determines where one begins to lean. What one becomes.
I'm with Jonsy on this one. Behavior creates "alignment." Not the other way around. The other way around alignment acts as a guide, but most humans are far too subjective about far too many things for alignment to determine how they behave. You can trust what a man says about himself if ya like, and most men can say a lot of things about themselves, one way or another. Say they are this way or that. But watch how they act. Behavior isn't something you can accurately describe or predict as a concept. It's something people display in a crisis. When theory rightfully means nothing much worth thinking about.
Behavior tells you how your align yourself when it counts. Talk and a concept only tells ya how you hope you'll behave all other things being equal. Action is what actually points the needle of the compass far as I'm concerned. A pre-conceived Alignment doesn't really point to true north, just painted north on the face of the compass.
As a real life example, I'll sorta expound upon what CF said:
5. This gets especially bad when DMs want to include moral choice or shades of gray, and their players do not share identical moral beliefs. At least half of the threads with players claiming that their DM screwed over their paladin, or threads with a DM complaining that his paladin's player isn't living up to his code, are a result of the DM and the Player having different moral beliefs. It usually goes something like this: The DM puts forward a moral problem, like, will you kill the presently innocent goblin child who's prophecied to throw down civilization in a rain of blood? Or will you let him live, and fulfill his destiny? The DM is thinking that this will be difficult, because killing innocent goblin children is wrong, but letting the goblin child live will be worse in terms of consequences. But the player doesn't think about morality the same way the DM does. He kills the goblin child instantly, perhaps because he cares more about consequence than deontology, perhaps because he has a more intentions based morality, or perhaps because he views inaction as the same as action and believes that he would be equally culpable either way. The DM is outraged. The paladin committed an evil act! He must be punished! The rulebook says so!
Because of my occupations and interests a lot of people have always thought of me as a Law and Order guy. That's how they see me. But some of my buddies know for a fact that I don't give a tinker's wink about the law at all, if the law gets in the way of Justice. I'd sacrifice, without a moments hesitation, law on any half-sunny day to achieve real Justice. The law just isn't important to me unless it achieves Justice.
So although a lot of other people see me as a Law and Order guy (even people in my family), I've never ever thought of myself that way. I'm a Justice guy. If the law achieves Justice, I'm all for that. If it gets in the way of Justice, then to hell with the law. So others can see ya one way, and maybe from their point of view, with very good reason. But deep down inside you know you're another way, for at least as good a reason as anybody else has.
That to me is why game alignments are meaningless, even as a gaming concept.
Everybody else has gotta see you exactly the same way you see yourself. How often is that likely to really happen, even as a fantastical concept?
Now if alignment is offered as a guide to think about how people wanna order the behavior of their characters, that's fine by me. Nothing wrong with guidelines in anything. Guidelines tend to be good for stimulating thought and new ideas. But as a description of how people are likely to react (ahead of time, much less always) well I think it's about to be as likely useful and efficient as somebody else thinking of me as a Lawman, and me seeing myself as an Agent of Justice. Yeah, denotatively they're maybe six a one and half dozen of another. But where it counts they're entirely different beasts of burden in my book.
Now, again, if somebody plays in a game where the world is structured so that everybody exactly agrees on what is good and evil, or law and chaos, or what constitutes true neutrality, then I reckon it don't matter much. But aside from fantasy worlds, how often is that likely to really happen? Come to think of it, even in fantasy worlds, how often is that likely to happen?
I'd bet that kinda world is rarer than gold wool and horsefeathers.