Rampancy

After reading several books that involve AI, there's really only one major downfall in creating a computer that can learn. It is the threat of rampancy. An AI will eventually no longer be able to be controlled by the protective algorithms in place, and the AI loses its ability to reason. It may run normally (unlikely), it may decide that it hates humanity for creating it but limiting its ability to function (I have no mouth...), or it may try to be more human and form a relationship with another rampant AI (Halo). These possibilities are not very promising, so i'm sure that if we created an actually thinking robot, we would install some sort of failsafe that would kick in after say, 10 or 20 years of service. This would protect the crews in deep space expeditions, but you would have to have multiple copies of the AI, or different ones. All in all, AI would drastically help us run our lives, we shouldnt turn them into robot servants like iRobot, but they could run our weapons systems, conducting cybernetic warfare with the opposing sides, they could eventually pilot our space fairing warships, and if we keep them in check, they will b very effective.