you do not need to define an "interest" as you called it. it is enough to allow units to have some behaviour which might change slightly over generations (mutation). you should be able to observe certain behaviors which have an interesting property. they might not be optimal for the individual, but if everybody does it, it would be stupid to do anything else. the utility function of the cooperators would exceed that of defectors. that's the basic scenario. beyond that, it gets complicated, but extremely interesting.
one very simple example. imagine baby birds in the nest. when mommy bird comes to feed them, they all scream. you scream louder, you increase the chance of getting food, but also the chance of mr. fox hearing you and putting an end to our discussion. how loud should you scream? surprisingly enough, you should scream 1/n (or something like that)of your full potential, assuming the rest of the baby birds are your siblings and have the same screaming capacity, and there are a total n of you chicks. want to make it complicated? sure. what if you are a cookoo bird chick, and these are not your brothers? does mr. fox eat all of you, or just a random chick? what if you are not brothers but cousins? etc.

you got right to the point. you need to make your decision in advance. the thing about evolution is that there is no decision making. they simple make as much noise as they make, and any other behaviour would be a disadvantage.
if the decision was cyclic, you could have an ever increasing (by epsilon) series of screams. chicks that are programs to scream 1/n are cool the rest suck.