It would be interesting to simulate the prisoner's dilemma through AI agents and see if they develop a sense of trust over time. Maybe they would adopt a cooperative strategy eventually.
There is an iterated version of the prisoner's dilemma and you don't need a clever AI to win. The name of the strategy is tit for tat and it basically wins over the long run.
Edit: This paper seems to go in that direction http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.43....