As a dilemma, it depends mainly on your framework. If you are following game theory, then the optimal strategy is to press the red button. (You survive in either outcome if you press red, but if you press blue you may or may not survive.) However, if you are answering on the basis of morality, then most systems of morality indicate you should press blue. It's equivalent to should you take a risk to save someone, or sit back and let them die if there is any risk to you.

It's got some resemblance to the famous "Prisoners Dilemma" game. In both cases the fact that the game is one-off alters the situation. If games like the Prisoners Dilmma are iterated ie played repeatedly, then it becomes rational to co-operate. This has been demonstrated with programs of different strategies playing against each other. The most successful (when I read about it) was "Tit for Tat": the program assumes co-operation. If an opponent defects, the program defects once in retaliation but then reverts to co-operation.