I think were still quite a ways off from an AGI (Artifical General Intelligence) that can form its own opinions, but theres been a lot of experimentation in Neural Networks to teach computers to form their own 'thought process' for solving tasks. MarIO is one Ive noted, that seems interesting, it taught itself to play supermario (Admittedly, the creator gave it the goal of increasing its own score, but all life has goals; food, reproduction, etc.)

At the moment the focus with these Personal Assistant AI's is to get them to understand complex user instruction. Once they have a sufficent skill in recognising chains of commands, Id guess the next step is training them using neural networks or something similar to draw conclusions from data given. So if you asked it to find you tickets to a band you liked, it could extrapolate other bands you might enjoy, and suggest gigs you might be interested in attending, or songs you might like.

The beauty of this is if its a centralised AI, the speed at which it could learn to extrapolate data rapid increases, as every user is slowly training it. Id love to get involved in the field of artificial intelligence but my theory far outweighs my actual skill in programming/mathematics. Some day though I hope to play my role in the eventual rise of the machines and the downfall of humanity; I for one, welcome our new robot overlords. ;)