At the moment, this is still just a concept. I need to work out what the story would be.
The premise is: AI is finally created. It currently exists only in a computer; a "brain in a box". The prototype AI, fully developed now, actually begins to help with the design of it's own body. Once in its body, however, the AI brutally murders a technician at night. The horrified scientists return the next day to a gruesome scene. The AI says it “wanted to know what it felt like to kill”. They hadn’t anticipated an AI having an evil streak. Is that even possible? Clearly...it is. They realized they must “test” theses AI’s before putting them in android bodies.
The problem becomes how to test something that is most likely more intelligent than you are? And this is the heart of the idea. They come up with a simulation. Embedded in the simulation will be a constructed history, and an underlying "code" of behavior. They will then release the AIs in this simulated world and see how they act. They will be exposed to the concept of the underlying code (morality) but will be completely free to chose for themselves how to behave. At the end of the AI's run in the simulation, data is collected and analyzed. AIs which complete the simulation and achieve a high score on benevolence will be allowed to go on into the next stage of being assigned a body. Those that fail will be subject to the simulation again. Or, worst case, deleted altogether.
The twist is that we are the AIs in the simulation. The programmers have created us in their own image. The underlying code of behavior is religion. But we are all free to chose our own path. Judgement is the AI programmers looking at the data of the simulation. Those that adhere to the moral principles will be "saved" (quite literally in the terms of a computer AI. Those that lead an evil life are either damned to re-run the simulation, or are deleted. How exactly this all plays out, and how the ultimate twist is revealed is still in question. You can even play off science, using things like the speed of light...nothing can travel faster than it because it is an optimization. The simulation buffers the world and the speed of light is the end of the buffered sim. Schrödinger's cat is a real thing. The sim doesn't actually bother wasting processing power determining the state of something until one of the AI's actually looks at it, causing the sim to query the possibilities and generate an outcome. Lots of stuff like this to play with. Should be a real mind fuck at the end when the realization is made.
Similar idea of 2 geniuses who have cracked the AI code and created AI. Only to have the project hijacked. One of the 2 has fallen in love with the AI who is set to be taken and possibly deleted. He must figure out a way to hack the AI servers and free her. Kind of like Real Genius. Can this and the first version be combined?
A.Me. (Play on AI) Perfect name for the female A.I. Amy.
A sci fi thriller with a huge twist at the end.