[izaidi]

I think the tricky part is that peak shift relies on generalizations (like "rectangleness") being made about a whole set of objects or experiences, so you first need some kind of foundation (like Pavlovian training) from which a generalization can be constructed.

When I'm designing an interface I try to make sure it has a visual vocabulary that's pervasive and consistent enough for the user to learn it from very little experience. So for example, the opacity of visual elements can be used to represent their importance; important stuff is brighter and clearer while unimportant stuff fades into the background. Once the user picks up on that, they can use it to make decisions about the importance of what they see: they'll focus extra attention on the text that's 100% opaque and ignore the text that's half-transparent.

Simple rules like this have existed in design for hundreds of years -- it's why newspapers give 72-point headlines to "big" stories. What's interesting now is that interfaces are becoming increasingly dynamic, and the tools we're using are letting us broaden the design vocabulary beyond just size and placement and typography.