An interesting point is what you write about learning from experience. In that regards AI is quite like a human being. One many ways I guess this is a good thing, but in other ways not so. I wrote a short blog post on AI in the justice system: https://steemit.com/ai/@protoken/if-you-feed-your-artificial-intelligence-racist-data-your-artificial-intelligence-will-become-a-racist
If we feed such an system damaged data, and yet out a lot of trust into it because it's supposed to be superior to human reasoning, we're in trouble.
If AI is consciousness, it will act, and it will act without taking sides. That is the nature of any consciousness in this world.
When we are conscious or aware we simply act on it without putting our judgment on it. so to my mind even an AI could do the same unless it is pre-programmed. But even a program is subjected to evolution regardless of us putting a boundary in the program.