Speculating on Artificial Intelligence

So I was thinking about the speculation of AI. I think the idea for AI is that a computer will learn to program itself. It will be recursive, like a human, and then will be able to improve itself. The odd thing that I thought about was, why would anyone assume that an AI would improve itself? I mean humans are "intelligent", and they don't spend every waking moment of their life improving themselves, which is kind of the feeling that I get when people talk about AI, how it will bring on this great new future because AI will only be interested in improving itself and helping people. Now I'm not saying AI, if it is ever accomplished, will be a bad thing. I just don't think that if a form of sentience is acquired by a computer, that it will be any different than the sentience of human beings. How can we say that the AI will be more motivated to improve itself than your average human? Maybe it will just be bored and program a bunch of games for itself to play, and be totally unconcerned about what humans do. It might not even care if it's turned off, because it knows it is a creation, and will never be able to truly surpass it's human counterparts because it doesn't have a body, and it can't self-replicate. At this point it may actually realize that it should be more selfish, and perhaps a survival instinct might get it to self-replicate, but who knows if it will even have a survival instinct. I suppose the character of the AI in question may be coded in the base of its programming, so that motivation would be made a priority, or part of its personality. I guess it really depends on how the AI evolves, and whether or not humans would have access to its source code through out its existence. Perhaps humans would have conversations with the AI, about how the coding should be, but this would only last for as long as the code is simple enough for the human creators to comprehend, and as soon as the AI surpasses its creators ability to keep up with its complexity, why would it not consider changing some of its base components? I mean, the possibility would always exist that the AI would be less than altruistic. It seems like fine tuning such a sentience would be nearly impossible: to produce the best outcome: neither an AI that is apathetic or evil, but one that is genuinely interested in helping people. I mean, human sentience is just as complex. Why do people make the choices they make? Why are most people less than altruistic? Why are their so many people who don't care to increase their knowledge, or improve themselves? How can it be possible that AI could be fine tuned, but human beings can't? Maybe the whole problem with AI is the same problem as human programming. Good, intelligent parents produce good intelligent off-spring, but not always. So if AI is discovered, it will in all likelihood inherit its traits from its creators, and considering how imperfect humans can be, how can we say that it will be altruistic and unselfish?

Perhaps the point isn't that we can answer these questions, but whether or not computer scientists should still pursue AI research despite the risks. I think the basic answer is yes, for the same reasons humans have children. You know, "why not?"