IV. Adjacent Territory Essay 26
Wholes No Part Contains
Patterns, jams, and things nobody is in charge of
A flock of birds turns. Hundreds of them, and every bird changes direction at the same moment, and the turn travels across the flock as a wave, and there is no bird in the flock that knows the flock is turning. Each bird is following three rules about the birds near it. That is the whole mechanism. The turning is not in any of them.
The same is true of a crowd walking to the left of a pavement, of a shoal of fish, of the cells in a growing thing, of the traffic on a motorway when a jam forms out of nothing and then moves backwards against the direction of travel while every single driver is trying to go forwards.
Some behaviour belongs to the arrangement and not to any of the parts. That is a strange sentence and it is worth slowing down on, because the whole of system thinking depends on it, and because the alternative view, that everything has a cause somewhere in the parts, is much more comfortable and much more often wrong.
Take the jam. Cars are moving at a decent speed on a motorway, and then, with no obstruction in sight, the cars at the back begin to slow. There is a reason, and the reason is visible: the car in front braked. The driver braked for a reason too, and so on, back to some small disturbance near the front.
Now look at what actually happens next, which is the part that makes the jam interesting. The jam does not stay where it started. It travels backwards along the road, at roughly the speed of the flow, from the front of the queue to the back. Every car in it is stationary or nearly so, and the jam itself is moving at fifty miles an hour in the opposite direction to the traffic.
That is a property of the whole. No car is moving backwards. No car is experiencing the jam moving. There is no signal and no coordinator. The jam is a shape that the arrangement of cars produces, and the shape has a direction of travel, and the direction of travel is the opposite of everybody's intention. The intention of every driver is forwards. The behaviour of the jam is backwards. Neither fact contradicts the other, because the jam is not a thing that any driver is doing.
Where the whole is not in the parts
It is worth being careful, because the idea is easy to run with and easy to run too far.
The weak claim is that the whole is more than the sum of the parts, which is vague and says almost nothing. The strong claim, the one worth having, is that there are behaviours which cannot be located in any part. Not "cannot be predicted from" the parts, which is a different and weaker statement. Cannot be located in them. A flock's turning is not in any bird. A traffic jam's backward movement is not in any car. A market's panic is not in any trader. When you ask any single part what it is doing, it says it is doing its own thing, and that is true, and the thing is not there.
And here is the important consequence, which cuts straight across the ladder. If the behaviour is not in the parts, then asking the parts will not find it, and asking the parts is what the abstract level does. "What do the cars do?" is a question about the parts, and the answer is "they brake when the car in front brakes", and that answer contains no jam. The jam is in the relationship, and the relationship is a different question.
How a whole gets made
Here is the part that is genuinely strange, and that has a large practical payoff.
You do not need a boss. You do not need a plan. You need a rule for each part, and the rules need to be local, and the whole behaviour needs to be more than the intention of anybody. A flock turns because each bird matches the speed and heading of its neighbours. Traffic flows because each car keeps a gap and brakes for the car in front. A market clears because each seller lowers the price until somebody buys.
And in each case, if you want a different whole, you change the local rules, not the whole. This is a genuinely useful idea, because whole-level control is nearly always unavailable and almost always slower. You cannot instruct a flock. You cannot instruct traffic. What you can do is change what each part does when it is in a particular situation, and the whole changes as a consequence.
What if a motorway had a rule that said brake earlier, leave more space when the road is empty, and slow down smoothly rather than sharply? The individual rule is a small change, applied by every driver, and the effect on the whole is large, because the whole is sensitive to the delay between one car's braking and the car behind reacting. This is the thing people find surprising: the road does not behave like a sum of drivers. It behaves like a fluid, and the fluid has properties that are not in the drops.
Now change the condition. What if instead you tried to control the whole directly, with signs and signals and instructions about when to brake? You would be trying to steer a thing that has no steering wheel, and the instructions would be received by parts, each of whom would interpret them, and the interpretations would sum to something you did not intend. The local rule is not a compromise. It is the mechanism, and it is the only place you can put your hand.
"Emergent behaviour means nobody is in charge, so it means nobody is responsible." The behaviour of the whole is not in the parts, but the rules in the parts are put there by people, and they are changeable by people, and a rule that produces a jam can be rewritten. The system is not in charge and the system is not innocent. Asking who is responsible for a traffic jam is a good question, and the answer is usually something like whoever set the following distance and the reaction time and the way the road is signed, which is to say a set of small local rules that somebody chose.
The other misconception: "if the behaviour is not in the parts, it cannot be explained by the parts." This is the error that makes people frightened of complexity, and it is simply wrong. The parts explain it, in the same way that the local rules explain the jam. What the parts cannot do is contain it. The explanation and the location are different questions, and almost all the confusion in this area comes from running them together.
Designing wholes you cannot draw
And this is where the idea stops being a curiosity and starts being a method, which is the part that separates this essay from a description of a fun phenomenon.
If you cannot draw the whole, you cannot manage it either. You cannot set a target for a flock, because the flock has no representation of a target. You cannot tell a traffic jam to go faster. But you can set the local rules, and the whole will move, and the surprise is that it moves in a direction you did not design and cannot fully predict.
That is a very different job from control, and it is a job that fits the system level rather than the level above it, because it is about the arrangement and not about the purpose. And it is a job that people are already doing, mostly unconsciously, in every organisation that has ever let a team decide how to do its own work.
Limits
This essay is careful with the word emergent, and so should you be. Not every behaviour that appears in a group is emergent in the interesting sense. A crowd moving to the left of a pavement is a shared convention, and a convention is entirely in the parts. A crowd surging towards a door is a shared convention too, until it is not. The test is whether the behaviour could be predicted by a rule applied to each part, or whether it requires the whole.
Second limit: the descriptions of flocks and traffic here are simplified to the point of being sketches. Real traffic has heterogeneous drivers with different reaction times and different risk attitudes, and much of the interesting behaviour comes from that heterogeneity rather than from the local rule alone. The sketch is a model, and it omits, and the omissions are load-bearing.
Third limit: an emergent behaviour is not automatically a good one, and the idea can be misused to justify leaving things alone. The claim that nobody is steering a system is very often true, and it is very often used as an argument that the system is fine. The rules are the steering. If the rules are wrong, the whole is wrong, and nobody steering is not the same as nobody responsible.
Stuck
You are stuck in system thinking when you are looking for the thing that is causing the problem, because there is no such thing, and you will not find it, and you will keep looking. The tell is a long investigation of individuals that comes back with a list of things each of which is doing something reasonable.
The way out is to change the question. Not who or what is causing this, but what local rule would have to be different for this behaviour not to happen. That question is answerable, it points at something you can change, and it is usually much smaller than the behaviour you were trying to explain. If the answer is a rule about a gap on a road, or a rule about how a team is given work, or a rule about what a bird pays attention to, you have found something you can work on, and you found it by stopping the search for a cause that was never there.
Sources
- Traffic jams that move backwards against the flow of traffic have been studied and reported in the transport literature for some time. No specific study, author, date, or figure is cited here, and no numerical claim is made.
- Emergence as a general idea, and the terms used here for local rules and whole behaviour, are used in a general way rather than as quotations from any particular text.
- All examples are ordinary and hypothetical.