What is Swarm Intelligence?

If you watch an ant try to accomplish something,” says Deborah M. Gordon, “you’ll be impressed by how inept it is.[1] When the biologist at Stanford University made this seemingly cursory remark, she was not renouncing the subject that she has devoted her life to. The fact is that ants are a massively successful species; more than 12,500 species have been classified, and they span six continents. [2] Although an ant has a tiny brain and very limited intelligence, the colony as a whole is capable of remarkable tasks: they build intricate nests, find the shortest route to food, and in the event of flood build rafts out of their own bodies.

It is clear then that there is some form of greater intelligence which arises in large groups of more simple animals, and we call it Swarm Intelligence.

The subject has been studied under the guise of Swarm Intelligence only since 1989, although swarming animals have been examined for centuries. [3] The distinction is that the new field does not just comprise biologists documenting animal behaviour. Computer scientists, physicists, and the like are using the lessons learnt from the swarm behaviour in nature to solve our own problems. For example a swarm-based internet search engine has been researched, based on the idea that a human could lay a pheromone trail when searching for information on the web much like an ant does when searching for food. [4]

It is important to be clear on what we mean by swarm here, because with Swarm Intelligence we are not just talking about the mechanisms animals use to travel in large groups. In fact the field encompasses all the mechanisms that allow colonies of animals to perform tasks difficult or impossible for the individual alone, such as foraging for food. And we might leave the word animal behind, for there are no animals living inside computer programs based on Swarm Intelligence. The word agent is better suited. So a swarm is a group of agents. But there is more, because back in nature ants are not being told what to do by some kind of chief ant. There is a queen, but her function is not to give orders. What we find is that the organised behaviour of the colony results from the local interactions between the ants themselves. These interactions result from simple rules - in some cases merely ‘follow the ant in front of you’. So we say that the ants are self-organised, and a swarm is a group of self-organised agents. By giving a set of simple rules to each agent in a group, the group becomes capable of complex tasks. This is a powerful tool, especially in computing. Instead of trying to face a difficult task head on, now we can just set out a simple set of rules for a large group of agents to abide by such that the desired result emerges from the system.

And this brings us to another word: emergence. Emergence involves local, and to some extent random, interactions between agents resulting in intelligent global behaviour emerging. (The previous sentence is worth re-reading: it is an important concept. A more detailed analysis of emergence is left to the next chapters, with examples in the following chapter.) Put simply, hundreds of simple events can add up to make something complex happen. In nature, this global behaviour is usually beyond the scope of understanding of the simple agents, so they are just doing what they are hard-wired, or ‘programmed’, to do. In the case of computer software the agents are directly programmed by a computer programmer, but in nature the rules must arise by different means. The complex, emergent behaviour renders a selective advantage as simple organisms are able to punch well above their weight - a swarm of ants might repair a nest very quickly for example. So animals hardwired to follow the right set of simple rules are more likely to survive, and emergence evolves.