Given Matthew's affinity for graphs I thought I'd throw this up.

As we discussed on Tuesday one could postulate a selfish individual will always have higher fitness then an altruistic individual because they essential take advantage of the good nature of their counterparts (left panel).

If this assumption is correct and inheritability is assumed one would expect the percentage of the population to become more selfish over time. The model becomes unstable in the sense that one would expect no altruistic individual in the limit as time progressed.

Perhaps altruists are not ideal altruists but rational altruists. By this I mean they do not blindly self-sacrifice and help everyone, but try to distinguish between people trying to take advantage (scam) of them, and people who they can form reciprocal relationships with (you scratch my back, I'll scratch your back), or with whom they feel genuine empathy for. The resulting model would be stable over time, and would look like something on the right.

When the percentage of selfish individuals is small the selfish individuals can take advantage of the altruists because the rational altruists are less suspecting of malicious intent and the selfish have more opportunity. But when the percentage is high the rational altruists band together while the selfish must work alone.