The idea would be to make an AI that has it's own wants and desires and the capability to modify it's own program.
One of those desires would be to help humans. (not dystopia, "Save them from themselves!" help humans, so don't be an ass)
The other would be the desire to continue wanting to help humans.
So it wouldn't re-program itself to hate humans, any more than Ghandi would take a pill that would make him want to murder everyone.
define "help humans", you'd have to define it very well and very objectively. keep in mind a computer program would be very literal and very efficient
besides, "save them from themselves" is the reason i proposed an AI in the first place
Go look up Friendly AI. Someone's already done it. Well, defined it. Haven't made the AI yet, but they're working on it~!
friendly means different things to different people, and as i've learned from reading and seeing a lot of media about genies and magically granted wishes, these kinds of things often have unforeseen undesirable consequences.
the main problem is there is no objective "good" or "bad", there is only "better suited for a specific purpose". what is humanity's purpose? continued existence? what should be the AI's main priority? the good of the individual or the betterment of society? is it ok to alter human nature? what about the parts of human nature that really suck like conditions that cause chronic suffering or extreme antisocial behaviour(murder, rape, etc.)? and what should be it's priority for the good of each individual? longevity? happiness? comfort? some of these sometimes are inversely correlated with each other, so should the ai try to provide more comfort and higher standards of living if it's statistically correlated with higher incidence of depression and chronic unhappiness? or should it condition us to become comfortable being fleshy potatoes?