Over at Language Log, we have a discussion of how to get discussion of how to get rid of Microsoft's Clippy by cussing him out. Being scientists and all, this leads them into deeper questions about whether it's possible to make a computer recognize when it's pissing the user off.
This, really, is a central feature of conscious thinking. Douglas Hofstadter addresses this extensively in Godel, Escher, Bach, an Eternal Golden Braid. When we answer a question or take an action two or three times then recognize that we're not getting a positive response, we "move up a level" in thinking and re-address what the questioner wants.
There's a downside to this. Part of what we like about computers is that we can anticipate that they will always respond the same way to the same commands, whether we like the response or not. If it starts trying to anticipate what we want, and gets it wrong, this makes it less predictable, which is not necessarily what we want from a computer.