This is my first sd journal entry. I'm taking a class on agent based electronic commerce and we spend a fair amount of time on ame theory. I posted the following as a response to a comment about the traffic light switchers. I know it is going to get buried, and I thought I'd save the post here.
Classic prisoner's dilemma has a dominant strategy. If you are only playing one game against one opponent, your best strategy always to defect (vs cooperate).
you C 2 0
___ D 3 1
If you are playing multiple times against the same opponent, then the game is called Iterated Prisoner's Dilemma. The analysis here is a little trickier since it depends on the strategy that your opponent is playing.
If your opponent always cooperates, you should always defect.
If your opponent always defects, you should always defect.
Pure strategies are not very interesting in this game.
There are a LOT of other strategies.
Unforgiving strategy: Cooperate until your opponent defects, and then always defect. If you know this is your opponent, then your best strategy is to always cooperate. This strategy has problems in noisy environments (where you might THINK your opponent defected, but he didn't really)
Tit for Tat: Do whatever your opponent did in the last round. Start with cooperate. If you know that this is what your opponent does, then you should always cooperate until the last round. There is no penalty to defect in the last round. However, if your opponent knows you know this, then they will defect in the last round too. So that means you should defect in the second to last round. Repeat until you are both defecting.
How does this relate to driving? In most large cities, driving interactions are anonymous. If someone defects, you don't get to 'punish' them by defecting the next time. The game is always a one-shot prisoner's dilemma and the best strategy is to always defect.
There are two ways to fix this:
1) Make the interactions so they are not anonymous, or so that defectors can be identified. The paint-balls on top of the fire truck are a good example of this. Another way would be to have some sort of p2p network where you could enter the offending vehicle and identify offending vehicles (not really practical given today's technology, but with good machine vision and smart HUDs then maybe)
2) Change the utility function of the game by adding punishments for defecting. The problem here is that it is usually done with fines. People have different utility curves for money (if you make 7 figures, a $100 fine probably isn't going to phase you much). It is difficult to set up a market mechanism that significantly lowers everyone's payoff for defecting against cooperation.