Here's a typical market place interaction: turtles A and B meet. A agrees to purchase an item from B. There are four possibilities:
I. A pays B and B gives A the item
II. A pays B with a bad check and B gives A the item
III. A pays and B gives A a defective item
IV. A pays B with a bad check and B gives A a defective item
Clearly scenario II works the best for A and the worst for B, while scenario III works the best for B and the worst for A. A and B both benefit somewhat in scenario II, while neither benefits in scenario IV.
This is the classical Prisoner's Dilemma game. Imagine A and B are prisoners being interrogated separately about a crime. There are four possibilities:
I. A and B both confess
II. A denies and B confesses
III. A confesses and B denies
IV. A and B both deny
In scenario I A and B both receive 3 year sentences. In scenario II B receives a 5 year sentence while A goes free, while the sentences are reversed in scenario III. In scenario IV both receive a 1 year sentence.
It's easy to see that the best strategy is to deny. If the probability that your opponent denies is p, then your expected jail term is:
3p + (1 – p) = 2p + 1
If you confess, then your expected jail term is:
5p + 3(1 – p) = 4p + 3
In this variation of PD A (the human) and B (the computer) are awarded points according to the following payoff table:
B cooperates B defects
A cooperates A: 3, B: 3 A: 0, B: 5
A defects A: 5, B: 0 A: 3, B: 3
A and B play each other repeatedly. This is called iterated prisoner's dilemma. The goal is to maximize the average payoff.
Try different strategies against a random computer or a computer that always defects.
Suckers always cooperate. Cheaters always defect. Impulsives cooperate randomly. Reciprocators always do what their opponents did previously. Try my version of this: