The swarms of complexity

This article was first published in The Guardian, 30 March 1989:

Army ants have been parallel processing for millions of years. Matt Nicholson reports

Nigel Franks, at the University of Bath, studies the behaviour of army ants in Central America and like many biologists he finds a computer useful. But his work is also producing results that could have important implications for the design of computer systems.

Biologists have long been fascinated by social insects such as ants, bees, and termites because while the individual insect is such an elementary creature, their societies can exhibit very complex behaviour. Army ants are rudimentary even for social insects, and yet the tactics they have evolved to search for food are extremely effective and exhibit a remarkable sense of direction. The whole is indeed greater than the parts.

A swarm can consist of half a million army ants moving across the forest floor in a swathe 20 metres wide, covering 200 metres a day in its search for food. In Africa swarms can contain up to 20 million ants. They live off other insect species, together with larger creatures such as spiders and scorpions. The army ants periodically swarm from a nest, rapidly depleting the area of food, and as a result are nomadic, frequently moving to fresh ground.

Nigel Franks found that the ants would swarm from the same nest 15 times before moving on, and that each swarm would be along a compass bearing roughly 123° anti-clockwise from the previous swarm. The swarm sticks to the chosen compass bearing and if it meets a stream, for example, the column will move along the bank until it finds a fallen branch where it can cross. Once across, the swarm will turn so as to maintain the previous bearing.

Working with a programmer and a mathematician, Simon Goss and Jean Louis Deneubourg of the department of physical chemistry at Universite Libre de Bruxelles, Dr Franks has used a computer to model the behaviour of these ants on the assumption that each individual is itself a simple processing unit, much like the basic computer proposed by Alan Turing in 1936.

Turing described an imaginary device, now called the Turing Machine, that moves along a tape reading, writing and processing the data on it. Each army ant leaves a pheromone chemical trail on the ground, and each ant has sensors in its head that can detect or ‘read’ the trail left by ants further up the trail.

It was assumed for the computer model that each ant in the swarm is exactly the same, and that each follows the same behaviour pattern when it comes to laying and following a pheromone trail. It was assumed that a simple left/right decision is taken by the ant, either moving randomly if there is no trail or following and reinforcing the strongest trail it encounters. It was further assumed that both the rate of decay of the trail and the speed of each ant is determined by the strength of pheromone; that each ant takes up a finite, discrete space; and that the ants leave the nest at a fixed rate.

When this simple model was run, using the computer to simulate the activities of around 100,000 ants, the result was initially disappointing. On Barro Colorado Island in the Republic of Panama, Nigel Franks had observed the swarms creating beautiful fractal structures on the forest floor, culminating in a dense, wide front; on the computer just one main stem led from the nest to the sweeping head of the swarm.

However, the picture changed when the team added food, and the assumption that an ant returns to the nest as soon as it encounters food, obeying similar rules. This resulted in something much more like the swarm patterns found on the island. Three species of army ant have been studied on Barro Colorado, each with a different balance of prey and each with a different swarm pattern. Changing the arrangement of the food in the simulation changed the simulated swarm to patterns backed up by observations. Allowing for the simplicity, the simulation was remarkably accurate.

The implication is that a simple processing unit, obeying a few simple laws, can cooperate in complex problem-solving behaviour when interacting with a few hundred thousand similar units. However, the chemical trail left by each ant could well be more than a simple on/off device. Although the trail is very simple, involving just one or two chemicals, this would be enough to convey quite complex messages.

Nigel Franks is now looking at the uncanny sense of direction exhibited by the ant swarm. The ant’s ancestors had compound eyes but now they are just one facet on each side of the head. The forest floor is very dark, with only occasional patches of sunlight breaking through the canopy. In these conditions an individual ant is virtually blind. What is needed is a large compound eye, sensitive to the light’s polarity, that can make the most of what light there is. He suggests that the 20-metre wide front of the ant swarm could be just that, with the swarm integrating and processing the information gathered from the front.

Whether this is actually the case, he has shown that a sufficient number of simple processing units, coupled with an efficient method of transferring data between them and a certain random nature to their behaviour, can build powerful solutions to problems. He sees the ant swarms as potentially “the ultimate neural network”.

Neural networks are of particular interest to computer scientists. Microprocessors like the Transputer from Inmos give us the means to break away from serial computing, where only one process happens at once, towards parallel or distributed processing, where many elements can all be working on a problem at the same time.

Army ants achieve this level of mutual cooperation without any administrative hierarchy, too. This has considerable implications for collective problem solving, indeed for a whole host of situations in our own society and industry.

But despite the increasing relevance of his work to the computer industry, he is not finding funds easy to come by. It may be difficult to see the relevance of research into a particularly obscure insect species; but it looks as though the computer industry would I do well to look at what biologists can offer. After all, this work demonstrates that “social insects have been parallel processing for 60 million years.”