27 Feb 2009

Here is a result from the group project.

To summarize shortly, the job description was to take an emergent system (where the exact outcome can't be controlled for in advance) and inject some agents that prevent some behaviour from happening. This could for instance be an effect you as a system designer don't want while you still want to allow the system to freely find other solutions.

In this little animation you can see the first proof of concept in a Particle Swarm Optimization. In the middle is a local optimum (darker background is optimal) and the grey agents try to coordinate in luring normal (red) agents to the hard-to-find global optimum in the upper left corner. Everything still happens decentralised.

It was fun to put it together in roughly two days, but now the tricky parts would begin:

  • This example with only two control agents worked well, but it all depends on the settings: What are good parameters to adjust the control agents (e.g. the ratio of control agents vs normal agents, what is k in the k-nearest neighbour approach, why do sometimes equilibria exist that seem hard to explain right now?)
  • In what direction should research in controlling emergence generally go? Martin Middendorfs initial approach [1] is interesting, but also just playing with the idea.

Update: Jeff Vail has some thoughts on "Guided Emergence" in Biology, the War on Terrorism and the Military.

[1] D. Merkle, M. Middendorf, A. Scheidler. Swarm Controlled Emergence - Designing an Anti-Clustering Ant System. Proc. IEEE Swarm Intelligence Symposium, 8 pp

# lastedited 31 Aug 2009
You are seeing a selection of all entries on this page. See all there are.