Swarms in the Abstract

Swarms in the Abstract

In an age of booming complexity, where governments, economies and societies alike have become impossible for any individual to comprehend, where software engineers wrestle with intractable computer systems and, as we've seen, computer systems wrestle with intractable problems, the study of social insects can do more than just pave the way to more powerful algorithms and robotics.

The promises and successes of collective intelligence and the swarm paradigm are a powerful demonstration of complex and useful systems arising from easily comprehended rules and habits. The units of computation are as dumb as ever, but the overall result we unhesitatingly term 'intelligence'. But what is it that these systems and our minds have in common?

The Swarm Organism

At the start of The Extended Phenotype, Richard Dawkins challenges the assumption that as living organisms we are delimited from the world by our skin. Why not talk about organisms in terms of individual organs which happen to interact, or cells which rely on each other for survival and reproduction? Maybe it would make more sense to go the other way and cite a mated pair as an organism (after all, try as they might a male or female alone cannot fulfil all of the biology lesson criteria for life), or perhaps(though Dawkins probably wouldn't extend the idea this far) a family group or the whole biosphere of the planet, each animal with their respective food chains?

If a swarm is simply a set of self-organising agents capable of performing distributed problem solving, the body of a multicellular organism can be seen to constitute such an entity. This idea is put forward by biologist Jesper Hoffmeyer in his paper, The Swarming Body[12]. Hoffmeyer's main area of interest is biosemiotics, the study of living things from the perspective of signs. Rather than study the interactions of organic chemistry, biosemiotics looks at the meanings taken on by elements of a living system, and how the relationships between those signals culminate in a working whole. In this paper, he points out that it is not the case that “our different body parts love each other”, and really the whole system of the body, an overlapping 'swarm of swarms' in his analogy, is maintained in precarious harmony by the communication of the different systems which make up the whole.

His paper also touches, tantalisingly but briefly, on the evolution of intelligence within this swarm organism. Initially, the simplest organisms begin to solve problems by manipulating “stupid molecules”, then as evolution grinds on, the problem solving capacity of the organism increases ('problem solving' in this case refers to the trials and tribulations of survival), but all the 'intelligence' exists in an entirely devolved way, within the interactions of the swarm constituting the organism. At this level of development, decisions are delegated to the individual cells comprising the whole organism, but at some later point the organism develops to a level of complexity where it is advantageous to defer certain decisions to the organism as a whole. The same swarm-based problem-solving is going on, but now one swarm has the new and peculiar property mediating between, or even directing, the other swarms. Bit by bit, out of the swarm is borne subjectivity.

The Swarm Mind?

Hoffmeyer's analogy is hardly a solution to the mind-body problem, as he would be the first to point out. However his paper demonstrates two key ideas: the near-global applicability of the swarm paradigm as a means of understanding complex systems, and the emergence of one kind of system (be it organism, brain or society) out of another (the swarm).

A better example of the emergent mind is given by Marvin Minsky in his seminal work, The Society of Mind. Although his book makes no specific reference to swarms, it is probably the best exposition of how simple agents can give rise to complex mental states. Minsky’s hypothesis differs slightly from a swarm-based equivalent because it does not capitalise on self organisation, although there are undoubtedly elements of SO present within the workings of neurons.

A fascinating, if somewhat oblique, example of equivalence between what looks like collective intelligence and what (theoretically) looks like conventional intelligence is the fact that certain implementations of Ant Colony Optimisation are very nearly equivalent to neural network approaches to solving the TSP. Both are connectionist systems, and both learn by increasing the strength of the connection, be it pheromone or a weight, between the nodes.

The Swarm Economy

There are definite parallels between certain features of social insects and aspects of human social systems. One striking example is the way in which ants distribute tasks between themselves. In any ants’ nest there is always work to be done, be it organising the brood, taking out corpses and other debris, collecting food or maintaining the nest. There is no centralised distribution of tasks, the ants must simply get on with the work in the most efficient manner possible, which turns out to be a complex interplay of stigmergy (seeing what needs to be done), self organisation (seeing how many other ants are doing something) and probability, all carefully balanced to ensure that jobs get done without causing large numbers of ants to attempt the same task at the same time. This partially stochastic approach to job scheduling has already been effectively employed in factories and distribution companies, but is interestingly analogous to how human job markets and international trade operate. It’s all a matter of distributing resources.

Grassroots Swarm Thinking

It doesn’t necessarily take complex computer systems, specialised robots or even any research to reap the benefits of ‘swarm thinking’, as a wet cement delivery company in rural northern Mexico has proven. Cemex (Cementos Mexicanos) used to have a huge amount of difficulty scheduling deliveries to construction sites in the Guadalajara region, bearing in mind the quality of the road network, traffic and unreliable contractors. Many cement companies attempted strict advance planning, but with scant success. Cemex however, took the opposite approach and devolved all planning to the level of the drivers themselves. They implemented a powerful telecommunications network, offering the benefits of self organisation to the truck drivers, rather than plans passed down from above. Drivers and dispatchers were able to make instant decisions about how to serve the customers in hand and as a result, deliveries went from 35% on-time to 98%, an amazing success.

It’s possible that this type of thinking will characterise the 21st century, where mighty corporations and institutions evaporate into networks of nodes and sub-nodes. Where individuals, dwarfed by the social, political, economic and informational networks they comprise once again surface as the collective masters of those networks.


  1. Marais, E. N., de Kok, W.(trans.) 1937: The Soul of the White Ant http://journeytoforever.org/farm_library/Marais1/whiteantToC.html
  2. Maeterlinck, Maurice. 1927: The Life of the White Ant. Allen & Unwin 
  3. Grassé, P.-P. 1959: La Reconstruction du nid et les coordinations interindividuelles. La théorie de la stigmergie, Insectes Sociaux 6: 41-84.
  4. Goss. S., Aron. S., Deneubourg, J.L. and Pasteels, J.M. 1989: Self-organized shortcuts in the Argentine ant. Naturwissenschaften 76, 579-581.
  5. Benzatti, D. 2002: Emergent Intelligence, AI Depot http://ai-depot.com/CollectiveIntelligence/Ant.html
  6. Dorigo, M. and Gambardella, L.M.. Ant Colony System: A Cooperative Learning Approach to the Traveling Salesman Problem. IEEE Transactions on Evolutionary Computation, 1(1):53-- 66, 1997. http://citeseer.nj.nec.com/article/dorigo96ant.html
  7. Schoonderwoerd, R., Holland, O., Bruten, J. and Rothkrantz, L., 1996: Ant-based load balancing in telecommunications networks, Adaptive Behavior, vol.5, No.2, .
  8. Di Caro, G and Dorigo, M. 1998: AntNet: Distributed Stigmergetic Control for Communications Networks. Journal of Artificial Intelligence Research, 9:317--365, . http://citeseer.nj.nec.com/dicaro98antnet.html
  9. Eberhart, R. C., and Kennedy, J. 1995: A New Optimizer Using Particles Swarm Theory, Proc. Sixth International Symposium on Micro Machine and Human Science (Nagoya, Japan), IEEE Service Center, Piscataway, NJ, 39-43.
  10. Collective Robotic Intelligence Project (CRIP) - http://www.cs.ualberta.ca/~kube/research.html
  11. http://www.swarm-bots.org
  12. J. Hoffmeyer, The swarming body, Proc. 5th Congress of the International Association for Semiotic Studies, Berkeley, 1994. http://www.molbio.ku.dk/MolBioPages/abk/PersonalPages/Jesper/Swarm.html

Particle Swarm Optimisation Demo

I have prepared a simplistic visual demonstration of Particle Swarm Optimisation in action. It does not aim to solve any difficult problem, instead I wrote it out of curiosity about how the particles of PSO actually move through the problem space.

The program initialises with a 3D cube and a bunch of 30 red balls, representing solutions. Each ball has a different quality value, shown by the size of the ball and the number next to it. 200 particles start off at random positions within the search space and should gradually congregate around the highest (or one of the highest) solutions. Note that the smooth movement of the particles does not happen in real PSO applications - this has been implemented to clarify how the particles are moving. They simply move in a straight line from their position on one iteration to their position on the next.

You can pause, stop and restart the demo to see the particles search new spaces, and there is also a 'spin' button to get different views of the cube. It's dead simple, almost totally useless but looks quite pretty.

Download the demo.

Remember you can visit the Message Store to discuss this essay. Comments are always welcome!. There are already replies in the thread, why not join in?