Emergence Explained: Weak vs Strong Emergence in Complex Systems

Last updated: 3 November 2026.

Emergence in complex systems is the appearance of patterns, behaviours or properties at the level of the whole that none of the parts has by itself. This article separates the two meanings that matter, weak and strong emergence, shows a small simulation you can reproduce, and explains what each meaning changes for people who model, design or manage complex systems.

Where the idea comes from

The modern argument starts with the physicist Philip Anderson, who in 1972 published “More Is Different” in Science. His point was not mystical. Knowing the laws that govern individual particles does not let you reconstruct the behaviour of a solid, a cell or a society, because each level of organisation has regularities of its own that need their own concepts. Reduction in principle does not mean construction in practice.

Philosophers then split the idea. Mark Bedau formalised weak emergence in the 1990s: a macro-level state is weakly emergent if it can be derived from the micro-level rules and initial conditions, but only by simulating the system step by step, with no shortcut. David Chalmers, in a 2006 essay on strong and weak emergence, used strong emergence for properties that are not even deducible from complete knowledge of the parts, with consciousness as the usual candidate. The distinction matters because the two versions make very different commitments.

Weak emergence: patterns you find by running the system

Take the common cases. Ant trails, flocking birds, market price formation, urban segregation patterns, traffic waves and the glider in Conway’s Game of Life are all weakly emergent. In every one the micro-rules are known and simple. Nothing is hidden. What is missing is a closed-form route from the rules to the pattern. You run the model and see what happens.

Weak emergence has three practical signatures:

1. Different vocabulary at different levels. You describe the trail, the jam or the market with words (width, speed, price) that make no sense for one ant, one car or one trader.
2. Robustness to swapping the parts. Replace individual ants with others of the same type and the trail survives. The pattern belongs to the organisation, not to the specific components.
3. Sensitivity to the rules. Alter a rule slightly, for example how fast pheromone evaporates, and the pattern can change qualitatively or disappear.

Figure 1. Individual ants follow local rules and feel the pheromone trail. A trail between nest and food appears at the colony level, and then shapes what each ant encounters next.

In the figure, no ant knows where the food is or that a trail exists. Each deposits pheromone while carrying food and tends to follow stronger pheromone. A small early advantage in one route gets reinforced, because more ants on a route means more pheromone, which attracts more ants. This is a reinforcing feedback loop of the kind described in reinforcing and balancing feedback loops, and it is the reason self-organisation shows up so often alongside emergence (see self-organisation in complex systems).

A worked example: phantom traffic jams from one small rule

The Nagel-Schreckenberg model, published in 1992, is the cleanest demonstration of weak emergence you can run on a laptop. Cars sit on a circular one-lane road made of cells. Each time step, every car applies four rules:

1. Speed up by one unit, up to a maximum (here 5).
2. Slow down if needed so it does not hit the car in front.
3. With probability p, brake by one unit for no reason (imperfect driving).
4. Move forward by its current speed.

I ran this on a ring of 1,000 cells with 150 cars (a density of 15%), discarding the first 200 steps and averaging over the next 400. The results are from a single random seed, so treat the numbers as illustrative rather than precise.

Setting	Average speed	Share of cars stopped
No random braking (p = 0)	5.00 (free flow)	0%
Random braking (p = 0.3)	about 3.05	about 19%

At the same density, with the same cars and the same road, a small amount of noise turns a smooth flow into one where roughly one car in five is standing still at any moment. The jams are not located at any fixed place. They form, drift backward as cars at the tail stop and cars at the head leave, and dissolve. No individual rule mentions jams.

Three lessons carry over to organisations and technical systems:

• The cause is distributed. Ask “which driver caused the jam?” and you get a nonsense answer. A single brake tap propagates because the system is close to its capacity limit, which is why the density matters as much as the noise.
• Fixing individuals does not fix the pattern. Telling every driver to “drive better” changes little if the density stays near the threshold. Changing density (metering on-ramps) or speed limits does.
• Prediction needs simulation. You can estimate the onset of jams analytically for simple versions, but the detailed pattern for a given road, density and noise level comes from running the model.

Strong emergence: the contested claim

Strong emergence asserts something bolder: that the whole has properties with causal powers that are not fixed by the parts and their interactions, so complete micro-level knowledge would still not let you derive them. The favourite example is subjective experience. Some philosophers hold that no arrangement of physical processes explains why there is something it is like to be an organism. Others argue that this reflects a gap in our theories rather than a gap in nature.

Strong emergence is tied to the idea of downward causation, the claim that the macro level pushes back on the micro level in a way that is not just a consequence of micro-level laws. Donald Campbell used the term in 1974 for biological systems. Weak emergence has a milder version of the same idea: a pattern constrains and shapes the behaviour of its parts (the trail shapes what each ant encounters, the jam shapes what each driver does). The difference is that in the weak case you can express that constraint in micro-level terms by simulation, while the strong case denies this.

For anyone building models, the safe stance is clear. Strong emergence offers no handle: if a property cannot be derived from the components, no model of the components can include it, and no intervention can target it through them. Weak emergence offers every handle, because the pattern is a function of rules you can change.

How weak and strong emergence differ in practice

Question	Weak emergence	Strong emergence
Derivable from the parts?	Yes, but only by simulation	Claimed not, even in principle
Can you reproduce it in a model?	Yes	No
Typical examples	Traffic waves, ant trails, flocking, markets, culture	Consciousness (disputed)
How to intervene	Change rules, feedback, constraints; re-simulate	No agreed method
Status	Mainstream in complexity science	Debated in philosophy of science

Emergence versus reductionism, correctly stated

A common mistake is to treat emergence as an argument against analysing parts. It is not. Weak emergence is compatible with the claim that everything the whole does is fixed by the parts and their relations. What it rejects is the idea that studying the parts in isolation is enough, because the interactions carry the information. The same parts connected differently give a different whole. This is the practical content behind the argument in holism versus reductionism: you need both the parts and the structure that connects them.

It also explains why complex adaptive systems are hard to manage by command. The parts adapt to each other and to the pattern they jointly produce, so the pattern changes as the parts respond to it.

Emergence in organisations and AI systems

Organisational culture is the classic emergent property in the workplace: it cannot be found in any one person’s behaviour or in any policy document, but it is stable, recognisable and shapes what newcomers do. The discussion in culture as an emergent property follows the weak sense: culture arises from repeated interactions, rewards and stories, and you change it by changing those, not by announcing a new set of values.

In AI, groups of interacting agents show collective behaviour that no single agent was designed for, including coordination, conflict and unexpected strategies. This is weak emergence at work, and it is why emergent behaviour in multi-agent AI is a design and testing problem rather than a mystery: you probe it by simulation and by varying the rules of interaction.

How to work with emergence

1. Name the level. State precisely what the emergent pattern is (a jam, a trail, a norm) and at what scale it appears. Vague claims of “emergence” cannot be tested.
2. Write down the micro-rules. List what each part senses, what it does and what it ignores. Most surprises come from rules that were assumed rather than stated.
3. Find the feedback. Look for loops in which the pattern feeds back on the parts. Reinforcing loops create and amplify patterns; balancing loops limit them.
4. Build a minimal model. The smallest model that reproduces the pattern shows which rules are necessary. Remove a rule and see whether the pattern disappears.
5. Find the control parameters. Density in the traffic example, evaporation rate in the ant example. These are usually the cheapest levers.
6. Intervene on structure, then re-run. Change a rule, a constraint or a connection. Do not instruct parts individually and expect the pattern to follow.

Common mistakes

• Calling every surprise emergent. A bug or an unmodelled input is not emergence. Emergence needs a pattern at the whole level that arises from interaction.
• Looking for the guilty component. Searching for the single car or employee behind a system-level pattern misreads the structure.
• Treating emergence as unmanageable. Weakly emergent patterns respond to changes in rules, density, delays and feedback. They are not beyond control; they are beyond direct command.
• Importing the strong version by accident. Claims such as “the whole is more than the sum of the parts” are fine as a slogan, but unless you can say what the interactions are, the sentence explains nothing.

Frequently asked questions

What is the difference between weak and strong emergence?

In weak emergence, a whole-level pattern follows from the parts and their interactions but can only be found by simulation or observation. In strong emergence, a whole-level property is claimed to be impossible to derive from the parts even in principle. The first is standard in complexity science; the second is a philosophical position that remains disputed.

Is emergence the same as self-organisation?

No. Self-organisation describes how order arises without a central controller. Emergence describes the appearance of properties at the whole level. They often occur together, as in ant trails, but a system can show emergent properties without self-organising (a crystal’s properties) and can organise itself with little that is surprising.

Can emergent behaviour be predicted?

Often yes, by simulation, and sometimes approximately by analysis. What you usually cannot do is predict it by reasoning about one part alone. Prediction improves when the micro-rules and the control parameters are known.

Does emergence mean the system cannot be controlled?

No. It means control works through rules, constraints and feedback rather than through direct instruction of each part. Ramp metering on motorways is an example: it changes density, an upstream control parameter, rather than commanding drivers.

Conclusion

Emergence in complex systems is best treated as a precise, testable idea: a whole-level pattern produced by interacting parts, visible only when the system runs. The weak sense gives practitioners all they need, because it points directly at rules, feedback and control parameters. The strong sense is an interesting philosophical question that does not change how you model, design or intervene. Next time someone calls a result emergent, ask which pattern, at which level, from which rules. If they cannot say, the word is doing no work.

For the vocabulary used here, see the glossary entries for emergence, self-organisation and complex adaptive systems.

Sources and further reading

• Anderson, P. W., “More Is Different”, Science, 1972.
• Nagel, K. and Schreckenberg, M., “A cellular automaton model for freeway traffic”, Journal de Physique I, 1992.
• Bedau, M. A., “Weak Emergence”, Philosophical Perspectives 11, 1997.
• Chalmers, D. J., “Strong and Weak Emergence”, in The Re-Emergence of Emergence, Oxford University Press, 2006.