This discussion is particular to Internet think tank(s). Refer to the book "Emergence" for a good overview of emergent optimization, if you are having difficulty infering the technique.
Note the word "route" in the reference on Internet think tank(s). This implies a degree of 'memory', that is, a dependance on the past. A simple idea would be, and here I invent a term(I think), each "user navigation session" would create a "thread", which is simply an ordered history of the pages that they've visited; the hops they made, whether through static links or searches or what-have-you. Now "route tables" would be constructed from the composite of all the "threads", and this is what we mean by emergence. The "route tables" would generate (in a very simple and straightforward way) "suggestions" for the user on where to go next, by showing them, essentially, the most often visited pages from this page; the most probable "routes".
Now here's the "memory" part: what if the user was researching a particular topic, and the current page they were on was relevant to multiple topics. The suggestions would be a composite of all topics that the page is relevant to. But can this situation be improved; can the page automagically know what topic the user was researching, and narrow down the suggestions accordingly?
To do this, it could use not only information about what page the user is on, but also what pages the user has just visited, to develop a more precise 'route prediction'. Thus, the route tables would have to have induced the appropriate information from the previous users. This is more complex than the simple "memoryless" version introduced in the first paragraph, and there are many perspectives from which to attack this goal. Suggestions for how to make both memoryless(simple) and memorypotent(complex) "route predictors/inducers" are appreciated. They don't have to be technical.