One of the main ideas we have came across many times in Snark Consulting is this: human brain connect ideas very much in the same way web links do. Roughly speaking we think by jumping between very different ideas, pretty much hypertext is structured. Example? A joke is a superlative jump of faith across several contexts.
Let´s go one step beyond. We like to master Google searches. We feel “at ease” with the web if we find what we are looking for in a couple of seconds. Strangely we find the right content (text, images, you name it) only if we can provide an approximate text of what we are seeking; believe it or not, this has to do with Magic. From Tolkien´s Gandalf to JK Rowling´s Potter, the power is hidden under the name of the spell, or under the precise name of the thing itself. So in the Web you find something in two secs only if you are pretty aware of the exact string and the context of what you´re looking for.
I think the next big thing would be a reverse search, where you could “enter” a first glimpse of the desired information -by singing a small piece of a song, by drawing a sketch, or even by sending a pic-. The output would be the desired mp3 song, a complete graph of the desired effect, or the name of the person we´re looking for. That certainly would be magic! But wait, for many consequences may arise: information much more tagged (who would do this? who would certifiy or approve?), security concern about my personal information scattered on the web, etc, etc. Don´t get me wrong: I don´t know if this has to do with scatter plots and traceback, face recognition methods, optimal projection vectors. All I know is that this feature rocks.
So you wouldn´t need the name of the thing, and all mortals could search with no need of some initial knowledge –or Magic-. Check Retrievr, draw something and let me know. No need to say I´m very curious about reverse searches!
UPDATE from 2014: Google now allow users to search using images as input!