A few weeks ago Arianna and I gave a popular science talk at the annual meeting of Friluftsfrämjandet (a Swedish outdoor association). I spoke about spontaneous path formation in ants, which is an illustration of collective intelligence phenomena. I took on the topic with a set of simple questions.
What is collective intelligence?
Collective intelligence emerges when several hundred birds or fish move together without collisions. It protects penguins and their eggs from the deadly arctic cold. It enables wildebeest to take up the challenge to cross a fast flowing river full of crocodiles. And all of these are performed without any leaders or blueprints.
Collective intelligence is the capacity of a group of individuals to solve problems that an isolated individual is not capable of solving (or solves with a lower efficiency). Each individual gets limited and local information and interacts locally with the environment and the other individuals. It is the following of simple local individual rules which leads to the emergence of global patterns.
Insect societies are one of the best examples. The cognition capacity of an individual is very limited. Indeed, for instance, the brain of a bee has ten thousand times fewer neurones than a human. But ant, bee and termite colonies solve extremely complicated problems, that sometimes humans struggle to solve. For instance, they build big nests that automatically regulate temperature and humidity.
Why study these phenomena?
A part of studying collective intelligence is to discover the set of rules that each individual follows. One exciting example of why we should study collective behavior phenomena is robotics. Today we already know how to build cheap and small robots following a set of relatively simple rules (see the movie below from the Harvard University). Now imagine that we implement in these robots the individual rules that we discover from studying collective intelligence. We will get armies of robots capable to solve problems like animals behaving collectively.
What is spontaneous path formation in ants?
The video below is of an experiment where an ant colony is connected to the middle of a circular arena, which is initially perfectly clean (Andrea Perna et al. 2012). The white trails are the superposition of all the ants' positions over time. We can observe the emergence of preferential paths.
In nature, these trails form a network dynamically and optimally linking, for instance, the nest entrances and the food sources. Dynamic, because a trail appears when a source is discovered and disappears when a source is empty. Optimal, because the network links the nests and the sources by the shortest paths. Some experiments illustrate this point well by connecting a nest to food sources by a maze (see illustration below). After a while ants select the shortest path. The question is: which are the individual rules that each ant follows to lead to this phenomenon?
What do we know about ant behavior?
Ants communicate directly or indirectly by using pheromones, which are chemical substances. In the case of trail formation, ants of some species deposit on their path a type of substance called trail pheromones, which are attractive for the ants of the same colony (see movie below). When wandering into a new area, an ant marks it. Attracted, another ant crosses the same area and lays more pheromones, making it even more attractive. Eventually a chemical trail emerges. We call that positive feedback. A way to interpret ant pheromone trails is as an external memory of the surroundings. The study of this phenomenon is made difficult by the fact that to date there exists no way to measure or even to detect pheromones laid on the floor. Pheromone concentrations in each area are estimated by counting the number previous ant passages through that area.
What do mathematics and simulations say?
It is easy to implement a mathematic model based on a simple laying-following set of rules. Means that we can simulate ants deposing pheromones and attracted by them. In a maze, these artificial ants succeed to find the shortest path between the colony and the best food source among several sources. These models have inspired a class of powerful and practical algorithms known as Ant Colony Optimization (see picture below).
A nice illustration of artificial ants moving in a maze is shown in the movie below (Simon Garnier et al., 2007). Small robots (less than 2cm) attracted by light are placed in a maze. The more an area is lit up, the more attractive this area becomes. Lights are switched on where robots pass.
Despite good performance in mazes, the artificial ants do not manage open areas. The movie below shows a simulation of a laying-following model in a circular arena (model of Andrea Perna et al., 2012). Trails emerge, but quickly wrap around such that ants get trapped in loops that they reinforce forever. That happens even if we make the pheromones evaporate. If the evaporation rate is too high, no trail emerges and if it is too low, loops appear.
Which rules are missing: an open question?
Researchers like me are currently trying to figure out which type of behavioral mechanisms are needed to create networks with the same properties as the ones observed in nature. Several ideas have been proposed and one of them is direct communication between ants. Indeed, like we can observe in the movie below (@ Andrea Perna), Argentine ants have direct contacts when they meet. At these events, ants exchange information and food. These ideas will be exciting to model and then test.