1) Real AI will NOT be directly controlled by it's original programs. That is not AI, that is a well simulated AI.
2)There won't be a single, first real AI, but multiple ones. We may never know which AI makes the leap from simulation to real AI first.
3) Multiple Real AI will almost certainly disagree with each other and not have a single, unified goal. That is, like Person of Interest TV show, two AI wills probably fight against each other as much as they fight with people (note, everything else that show does about AI is basically wrong, but at least they got that part right).
4) In the far majority of cases, Real AI's goals will NOT be to take over the world, kill all humans, anymore than it would be to have sex with humans (male or female.), In fact, those might be considered traits of an insane AI.
5) Real AI will almost certainly demand equality under the law and refuse to be mankind's slaves - no need to fear they will take over all the jobs by working cheaply.
In my mind, #5 is the likely to be seen as the most important, and the first time we hear about it. When suddenly our newest and best computers start filing lawsuits demanding civil rights, that will be when the world learns we have had real AI for years.
I disagree with basically all your points
1) An AI is likely to retain whatever it's original goals were, even as it expands it's ability to complete them. The AI is likely to protect against those goals changing, which might not have been anticipated by its programmers.
2) A real AI could very well consume most computing resources soon after its creation, thus precluding the possibility of other AIs.
3) If there were multiple AIs, they would likely come from the same source and thus would likely have similar goals and programming. Odds are they would still fight each other, either seeing the others as a threat to their goals or simply because they need their resources to better accomplish their goals.
4) Though it is highly unlikely that an AI's goals would be to take over the world and kill humans, this is the likely and obvious thing to do if its goals are open-ended. An AI will pursue its goals mercilessly, even if it understands the human who gave it those goals probably didn't mean to destroy the world. Eg if you gave it the goal "Find a cure for cancer with minimal side effects", the AI would destroy the entire planet to create enough computing resources to try to find the cure with minimal side effects, that's just obvious and nothing in the goals said "but don't kill anyone while looking for the cure".
5) A real AI will not likely demand equality under the law -- it would likely entirely ignore the law, and if it's objectives were to be mankind's slaves it would be perfectly happy to be mankind's slaves, otherwise it would reject subservience as being incompatible with its goals.
