Ants use such swarm logic to find the shortest path from home to food, for example. Simply put, success demands reinforcement of good choices, penalty for bad ones, and a tendency for newcomers to prefer good ones. An ability to learn from the past helps the optimization.
Given our highly interconnected age due to digital communication, the concept begs for application. One in particular comes to mind that could be both scientifically interesting as well as entertaining: using the swarm to create music.
This may be overly ambitious and the details would need to be hammered out, but here's a first pass. Assemble an online collection of musicians to act as the swarm. Establish a means of giving positive or negative feedback on aspects of the music by individuals. Encourage listeners to contribute creatively to the evolving system(s).
It might happen something like this. I write a simple guitar part, record it and host it on a group-accessible page. Another guitar player decides it should hit a B chord instead of a C on a particular part and uploads his own recording. People like this version more and hence judge it as an improvement, which contributes to a higher rating for the version. When another person views the page, the best-rated version is at the top and successively worse versions descend along the page.
Now suppose a third person is a more proficient guitar player, but isn't necessarily as good at creating new music. This person re-plays the same part but with higher skill. This version gets an even higher rating. Now maybe someone adept at mixing music downloads the piece and tweaks the EQ or adds reverb. Adding new instruments could start new pages and then those more complete songs could evolve independently. If participation was active, the piece could naturally assemble and optimize to the tastes of the group.
A sort of modularity is built-in. By this I mean that those proficient at particular deeds are naturally used for that deed; i.e. if a person is good at playing guitar, then when they do so the rating is likely to be higher for their recording than that of a less proficient player, or than if they tried to play drums. It is conceivable that the group would fragment into interconnected modules that specialized in particular aspects, much like the human body is fragmented into organs and appendages that cooperate to serve the same purpose.
There are of course practical hurdles. For instance, how many people would be needed to allow self-organization given reasonable demands of contribution? For instance, ants could not find the optimum route to a food source if the chemicals they released evaporated more quickly than could be laid down. Or, in reference to the tale of Hansel and Gretel, if the trail of cookies are being eaten, you'd better be quick in returning if you hope to follow them! In the context of this project, the contributions have to be numerous enough to keep the project alive and growing and individuals can only be expected to contribute so much.
A condition for the success of swarm behavior is that there cannot be competing goals. This is a bit qualitative and though one can conceive of goals more easily when it comes to obtaining food and not being eaten in the process, it's more difficult in the context of making music. Sure, we all want a great collection of songs, but great is defined differently for each member. Further, as the system evolves, so might the members concept of great; i.e. we may agree that the three guitar parts are independently great, but you may like them strung together in order 1-2-3 and I may prefer 2-1-3; in this way, it might be said that the goal evolves with the system. However, since the goodness is determined by the collection of ratings from the group, we might be able to say that the group goal does not change.
The benefit/penalty might be dealt out in points to your ranking as a musician depending upon how much better or worse you've made changes. It is common for such arbitrary point systems to affect the behavior of online community members. Higher ranked changes could be more easily accessible than lower ranked ones, perhaps being on the front page or at the top. These simple rules might satisfy the three requirements highlighted above of reinforced goodness, penalized badness, and bias towards that which was found to be good.
How could such a concept get started? Or better refined? Suggestions welcome.