Joho the Blog
An Entry from the Archives

Eric Bonabeau is talking about "Inspiration from nature for designing computing systems."

He begins with an exceptionally clear example of how ants "solve" the problem of finding the shortest route to a sugar cube. Those that discover the shortest path deposit the freshest pheromones. He now applies this to the Salesman Problem (shortest routes among multiple cities). [It involved math so I played Minesweeper until the scary symbols went away.]

He shows how this form of emergent programming can result in more efficient routing of packets.

Now he's onto the "second lesson from nature: simple rules rule." With ants, a small ant carrying a load will give up that load when it comes across a larger ant. This is apparently quite efficient as proven at CVS drugstores where workers were ranked in terms of productivity and organized into ant-like "bucket brigades." [How appealing!]

Eric asks us to imagine a game in which we are assigned an aggressor and a protector. Each person has to try to put his protector between himself and his aggressor. The overall pattern of behavior is very hard to predict. Change the rules a little and you get a different outcome. And individuals' explanations don't really account for the way it turns out: people don't know what they're part of. But you can simulate this via bottom up modeling, and you can affect behavior through minor adjustments of the rules. E.g., SW Airlines did this for cargo routing and saved $10M a year with a 71% improvement at the biggest hubs.

Eric tells a cuationary tale about blindly following simple rules. For example, army ants may sometimes form a circular path and follow each other around, putting down more and more pheromone that causes them to march ever faster, until they die. [Ah, a metaphor for life.]

Lesson #3: No one has to be in control. His example: Ants randomly position themselves around a big item until by chance they are all pushing in the same direction. Also, nest construction by wasps.

How do we shape emergence? Great demo at the end: Give people a slip of paper saying who they like and who they hate. They are to move towards those they like and away from those they hate. With random initial conditions, we get the same pattern: a rotating line within a following circle. Eric ran the simulation and it was, well, cool.

Practical results: Self-organized satellite deployment, swarm-based senor networks, self-healing networks. Social control and world domination. Coming soon.

Glenn Fleishman asks: How do you model perversity, i.e., "particles" that purposefully don't follow the rules. Eric says this asks what you do if the particles don't like your rules. Answer: you can make the rules robust against perturbation. [Doesn't this mean that you can make rules that result in particular behaviors that the particles — we — don't recognize as bringing about that group behavior and that we don't even want to follow?]

[Would someone like to tell me if this shows the importance of Wolfram or his irrelevance? Or neither, of course.]