Software Agents

I'm very much into this A.I thing these days. Cornell University has had a successful demonstration test of its self-replicating robot. Nothing intelligent there, still, it follows some seemingly simple rules that required people of IQ not less than 146 to design.
With A.I, something else also come into picture: Software Agents. They, being an agent that do some minor stuffs on behalf of us, like getting the best bargains, booking a flight, accomodation, planning a meeting, etc, do require some intelligence to make decisions on behalf of us. Thus, Software Agents have to learn our habits of doing things, our likings and dislikings, our preferences and finally able to predict what we have in mind should we be in its situation and thus act accordingly.
Software Agents possess such qualities as Self-Learning, Collaborating, Autonomous, Proactive, Reactive, etc. My idea of a software agent is a piece of software that has a personality, much like a virtual secretary or receptionist. That will serve as the interface or bridge or medium between us human and the machine. It has to translate our goals and desires and facts from Natural Language into a goal table and eventually end up as bits and bytes of executable codes. Having scrapped away the free-will and emotion part (atleast for now, as they're not considered to be vital components of the software agent), I actually don't think it'll work pretty much the way we envisioned it to be, instead it'll still behave as an ordinary machine, only with some common sense and getting input from voice and facial expression instead of keyboards and mices. It'll serve as the complement of our secretaries. For example, a real secretary would have known that you're happy for some reasons and decided to add some surprises to your business meeting.
I've actually drafted a primitive flow chart of how should a software agent works. Since I don't have that much expertise in A.I, I decided to start from Learning and Automation. The idea is this: It sits there quietly while learning the way you use the computer and slowly, as the frequency of certain repeatative actions exceeded a threshold, it'll automate things for you. Eventually, we can try to add some simple natural language processing. It'll parse simple voice commands, and with some common sense learned from the past, decide what's best for me.
I don't quite like the way A.I is being developed now. I actually think A.I can never progress to that of a real A.I. It'll of course exhibit some very convincing intelligent behaviours that we'll have no qualms of taking it as good as human's but still it's just a bunch of silicones and metal scraps and electrons that roam freely around the circuits. I'm talking about free-will, self-awareness, cognitive abilities, etc. These are the very fundamental components, of which coupled with basic instincts, will create a creature that could adapt to any surrounding its body is permitted.