About Me

My photo
Having reviewed over 1100 books on this blog, I have now written one myself. Motherdarling is a story about a search for a missing Will which reveals long-hidden family secrets. It is available on Kindle through Amazon. Read it and find out whether this critic can write. I live in Canterbury, England. I lived for more than thirty years in Bedford. Having retired from teaching; I became a research student at the University of Bedfordshire researching into Liminality. I achieved my PhD in 2019. I am now properly retired. I love reading! I enjoy in particular fiction (mostly great and classic fiction although I also enjoy whodunnits), biography, history and smart thinking. Follow me on twitter: @daja57

Wednesday, 1 June 2011

'Smart Swarm' by Peter Miller

This book analyses animal behaviour and tries to learn lessons for they way humans do things (especially in the business world). He considers ants, bees, termites, birds, fish and locusts. Some of his lessons seem contradictory but I suppose one wouldn't expect behaviours that have evolved would necessarily be the same in species that fill different ecological niches. Thus the haphazard way in which ants pass communicate and therefore manage the nest is very different from the way honeybees do.

Foraging ants wander randomly to find food and then scuttle straight back to the nest, laying down a pheromone trail as they do so. Other ants follow the pheromones, laying their own. Soon the best trails are those most strongly marked with pheromones so the ant paths become marked. I have experienced this; if you disrupt an ant trail using an obstacle such as your foot which they have to go around and you leave that long enough then remove it the ants will continue to loop around the now non-existent obstacle, blindly following the pheromones. That might suggest parallels with business!


Instead of relying on a priori algorithms to lay down procedures top-down, just trying something and scoring how well it succeeds can quickly transform routing problems through complex networks (such as the travelling salesman problem) whether geographical or virtual.

Honey bee scouts search for great places for the swarm to hive and then return home and communicate with the others using dance. Lots of scouts do this, many checking out the places suggested by their rivals and then (maybe) and the hive then decides. They have a sort of distributed democratic decision making procedure. First the scouts possess diversity of knowledge, second the hive only swarms once there are sufficient scouts agreeing on the best place to go (a sort of Tipping Point concept).

Miller explains the Wisdom of Crowds idea in terms of the Who Wants to Be a Millionaire 'Ask the Audience' strategy. Imagine you have to choose which person was NOT a member of the Monkees (A, B, C or D) and you asked 100 random people. Suppose only 7% know the right answer but 10% can narrow down the answer to a 50:50 guess and another 15% can eliminate one wrong answer leaving the remaining 68% to make a pure guess. So the right answer gets 7+5(half of 10%)+5(one third of 15%)+17(one quarter of 68%)=34%. The other options will share the remaining 66% giving 22% each. So there will be a clear majority for the right answer even though 93% of the sample guessed (to some extent)!

But the key to true wisdom is to ensure a useful diversity of knowledge and that this is expressed. There are lots of irrationalities to which humans are prey when making decisions such as anchoring (eg when a salesman suggests an initial figure for the cost of a product so that all other prices are compared to this 'anchor'). Other decision traps include preferring the status quo and the 'sunk-cost' trap. Miller diverts to talk about the Beer Game ( from Peter Senge's The Fifth Discipline) and then explains how town meetings in Vermont are run according to Robert's Rules of Order (whose best rule is that everyone who wants to should have the chance to speak before anyone may speak twice).

Miller next compares the way termites repair a damaged hill with the chaos that ensues when a highly connected system such as a power network fails and a small initial failure cascades through the system (which made me think of what happens when a person dies). Termite hills apparently are designed to regulate the environment inside the nest (which is actually below ground) by using resonating air columns to catch and tame turbulent winds. Miller suggests we should develop buildings where the walls are porous and thus able to regulate the internal environments of the buildings without using energy eg for air-conditioning.

Termites work using stigmergy,  a form of positive feedback in which the action that one agent performs makes it more likely that the same or other agents will perform the same action, such as when you get an intrinsic reward from the performance of a task. Stigmergy involves indirect collaboration which, Miller implies, could solve failures of highly connected networks, although he never really suggests how. It seems to be connected to wikipedia.

Miller then compares two forms of networks: small world and scale free (which is one in which a proportion of nodes are many times more connected than other nodes in a 1/r way). Random failures affect small world networks most but scale free networks are very vulnerable to hub failures (if one airport in the world closes then air travel is unlikely to be affected because most airports handle a few planes to weird places but if that airport is Heathrow then there is big trouble). In scale free democracies a small group of well connected individuals are able to impose their will on everyone else. Later Miller shows that informed individuals can also effect big changes on inter-connected systems as the Wisdom of Crowds meets Tipping Point meets Critical Mass.

Generally "it doesn't make much sense to discuss the properties of network structures ... without also discussing what it is that you want them to do." (p149)

He also explains about the power of mimicry in social animals when discussing the flocking of birds or fish. He discovers that starlings use a sort of small worlds network in which they pay attention to 7 other starlings (presumably because they cannot cope with the amount of information processing that paying attention to more than 7 implies). This creates swirling behaviour and also the sort of phase transition discussed by Philip Ball in Critical Mass. In schools of fish communication seems to travel in a wave that moves up to 15 times faster than any single fish. Because this form of communication means that not every animal needs to be on high alert against predators, all animals can spend more time feeding and therefore each animal is more likely to survive. The phase transitions seen between one sort of shoal formation and another are classic threshold situations: Critical Mass meets Tipping Point.

Society "consists of institutional arrangements to overcome those divergences between perceived individual interest and some larger collective bargain." (pp265-266)

I learnt the expression 'ground-truth' which means to apply a theory or simulation to the real situation to see whether it works in practice.

Other great books in this area include:

  • Six degrees about small world networks by Duncan Watts
  • sync by Steven Strogatz
  • At Home in the Universe by Stuart Kauffman about fitness landscapes
  • How Nature Works by Per Bak about sandpiles and self organized criticality; an excellent explanation of complexity science
  • Deep Simplicity by John Gribbin which is a brilliant introduction to this whole field
  • The Information by James Gleick although his Chaos (not reviewed on this blog) is perhaps better

Other books not reviewed on this blog on this topic include:

  • The Wisdom of Crowds 
  • Tipping Point by Malcolm Gladwell about fads
  • Ubiquity which is brilliant about fractals and power laws
  • Critical mass by Philip Ball which is a brilliant explanation about phase changes

Fascinating although a lot of what is said are points better made elsewhere.

May 2011; 269 pages

No comments:

Post a comment