With desktop software, usability is largely operating with WIMP (Windows, Icons, Menus and Pointers). Generally graphical user interface toolkits always share certain conventions. Following conventions in itself is not a bad thing - after all, familiarity acts as great leverage for learning new things. However, it might become troublesome if the conventions themselves become outdated by new technology. WIMP is not designed to deal with tangible interfaces, voice input or gestures.
This has been noted by HCI researches in at least several papers, most likely more, using different names for things. One fundamental limitation in our current applications is that they are designed for exactly one pointer. Multitouch screens on recent smartphones and PCs are slowly paving the way for multiple pointers, allowing a user to touch several points at once. Tangibles, researched since mid 90's, will move beyond touch screens, allowing virtual objects to be controlled by multiple physical objects. Which is faster: dragging objects on the screen with a mouse or moving multiple physical objects on a flat surface?
Technologies for tangible input have been here for some time. Same can be said of voice input. Gestures are making their way, receiving a huge boost from Microsoft Kinect. Combining these technologies with the traditional keyboard (still king for typing) should result in better interaction overall. I'm not ready to kill the mouse either, it is still pretty good for pointing. Sometimes touch screen can replace it, but I don't think there are too many as effective instruments for certain tasks (especially certain games).
So, point? I don't think that our current conventions are able to adapt to these new technologies. Trying to fit these new possibilities within known conventions is more likely to hindrance improvements in overall user experience. Learning new things cannot be avoided forever. Of course in creating new interaction models it is our responsibility to make them easy to learn. Conventions from the physical world should be useful for us.
Changing the desktop world is probably too late though. That is why I've set my sights to ubicomp, where conventions don't really exist yet. It should also be easier for people to accept learning new ways of doing new things, than it is replacing their comfortable old ways of doing things with new ways. Tear down the wall, let creativity reign!