I love the idea in this Slashdot post: show pictures instead of (or alongside?) error messages. From what I’ve seen, it’s quite true that most people do not read/remember error messages at all, just keep clicking buttons until they get a response they recognise. The result of having switched from a character-based to a graphical interface, I guess – like switching from needing to understand the context when writing an essay to providing random answers via multiple choice, perhaps.  People often seem to regard what I do as magic, when in reality it’s simply a combination of paying attention to what happens on the screen and “reading the manual”.