Complex Systems Science in Sustainability Research

Complex Systems Science is be more than a playing field for interdiscipinary researchers. What can we contribute to solving urgent societal problems? 

Complex systems science in sustainability research

This article is based on a talk by Bert de Vries on 26 April 2019. See here for his slides. 

If the world were governed by a fully rational, all-powerful “social planner”, then sustainable management of resources, including the climate, might be easy. Or at least, it might boil down to a tricky, but fairly well-designed engineering problem. The social planner (or his minions, the scientists) would use ecological surveys or weather models to assess, say, the amount of fish or fresh water that can safely be extracted at a certain location and tell the fishermen or farmers what to do. Simple Integrated Assessment Models for climate policy assume that a social planner optimises the world’s carbon budgets.

But unfortunately – or fortunately – our world is not that simple. Earth is inhabited by some seven billion people who each take their own decisions, interact, influence each other, and build emergent structures – families, states, religious institutions  –  which in turn also interact with each other. Individuals and institutions base their actions on many different motivations and incentives, short-term and long-term, economic or spiritual.

Modelling seven billion people is of course hopeless. Not so much because 7 billion is a large number – a box with seven billion oxygen molecules would be a rather simple system -- because all oxygen molecules are alike and follow easy fundamental laws. But humans are not atoms, or even bacteria, which you can take to the lab to analyse their “fundamental laws”. We do not even have measures to quantify human feelings, drives and achievements, let alone equations to predict their behaviour.

Yet in sustainability science, those hard-to-grasp humans are very much part of the system under investigation. So, what can we do? Should we try to abstract away human complications as much as possible in order to gain tractable mathematical models? Do we have to give up quantitative modelling altogether and resort to storytelling, such as writing descriptive case studies? Or is there something in between? 

One way to go could be to gradually add more “human complexity” to simple models and investigate its effect. For example, in case of CO2 reduction, instead of one single social planner, we could study several entities – say, countries or groups of countries – negotiating their climate mitigation efforts. The countries could be perfectly altruistic, i.e. aiming to optimise the global welfare, or perfectly egoistic, i.e. optimising their own welfare, or something in between. Altruistic countries would pay costs to reduce their emissions provided that the global benefit exceeds the costs. Egoistic countries would only pay to reduce emissions if their own benefit exceeds their cost. A simple modelling exercise shows that, unless countries are very altruistic, climate mitigation and global welfare decrease when the number of countries increases. This model is far too idealised to predict future  emission reduction, but it does serve to illustrate a relevant effect: The more players, the more difficult it is to cooperate, even if cooperation would be beneficial for everyone. The model also allows an important conclusion: Due to the large number of countries, a high degree of altruism is required from each country to enable us to limit global warming.

In the above example, all agents were assumed to be alike and have the same goal (optimise local and/or global welfare) and strategy. Of course, agents could have different strategies. Economists often assume that agents maximise their profits in a fully rational way, with perfect information about the market. Obviously, real humans are not like this. They have imperfect information, make mistakes, or even don’t care about growing ever more rich. But even if agents have the same goal, their strategies may differ. Imagine for example a village of fishermen who can either sail to a plentiful fishing ground at A or a poorer one at B. Assume that each fisherman wants to maximise his catch. If they were alone, the best strategy is to sail to A, where more fish can be caught. But if everyone else goes to A, maybe I catch more if I alone go to B? So maybe the best strategy is to look where everyone else is going and head to the fishing grounds with least boats per fish are heading? Now already the decisions of individual fishermen are influenced by the others. But of course the fishermen could also form teams or even a village-wide cooperation. Which strategy performs best depends on the abundance of fish. If there is plenty of fish for everyone, coordination does not help much, but if the exploitation increases, cooperation becomes more profitable – until overexploitation kicks in. Once again, a sensible, but qualitative, result.

Of course, humans not only interact by making each others’ strategies more or less profitable, but also interact more directly by communication. This way, they can influence each other’s opinions, values and aims. Yet another simple modelling study suggests that the more connected people are, the easier a sudden regime shift in opinions becomes. Could this be good news for sustainability? Could we tip the system to make sustainable behaviour fashionable?

In all of the above examples, toy models were used to study specific effects of human behaviour. They yield intriguing insights, or at least demonstrate that human behaviour cannot safely be ignored. But they only deal with isolated phenomena. For many purposes, we need more complete models. For example, we might want to make forecasts  about climate change. For this, we need CO2 to forecast  emissions, which in turn depend on economic development, political decisions and customer’s choices.  We could concoct assumptions about all these process and put them into a huge model, arguing that all these effects are important and we cannot ignore them. But how can we ever validate such a huge, intractable model? Wouldn’t it be better to use “wrong” but simple models (like the social planner models) of which we at least understand the outcome?

There is no clear answer yet on how to create credible models involving human behaviour.

One approach could be to start with simple models and gradually add features – possibly derived from toy models like the examples above, carefully testing their effects along the way. The disadvantage is that this can easily lead to lack of transparency. 

For some applications, one might use hierarchical modelling. For example, one might use detailed models of a company, or a neighbourhood, and combine insights in a simplified form when studying larger structures like countries or the global economy. 

A very different approach is to replace modelling the most unpredictable human part of the system by using scenarios. We acknowledge that we cannot tell how society will develop, but acknowledge this uncertainty by presenting several storylines which all seem plausible. 

In the case of CO2 emissions, one might suppose that humans start caring much for environment, develop a close global collaboration, and are innovative with respect to green energy. Or one might suppose that while there is considerable collaboration to foster economic development in poor countries, environmental protection ranks low on the global agenda. For each of these narratives, one can project a likely greenhouse gas emission (and hence, climate) pathway. So without venturing to forecast how the “global human mindset” will develop, one obtains a range of possible futures.

When dealing with complex systems, we will have to adopt a more humble notion of forecast. In complex systems, human or otherwise, tiny perturbations can lead to very different outcomes, making it impossible to predict trajectories.

As an example, if five competitors start a company under similar initial conditions, it is impossible to predict who becomes market leader. But we can estimate the probability of ending up in a regime where only one competitor survives. If they are local bakers, then probably all five survive, unless their bakeries are very close to each other: Bakers can only produce a limited amount of bread per day, and customers will only travel a limited distance to buy bread. However, if the entrepreneurs open a web-based car sharing platform, it is likely that “the winner takes it all”, because the bigger the company gets, the more well-known it becomes and the more customers it attracts. We are unable to predict one firm’s profit, but by understanding the dynamics, we can make regime predictions (“probably, only one car sharing company will survive”) and even steer the system to some extent (“if my company is to win, I must do everything to gain customers quickly”).

How far can we get in creating good – or at least useful, fit-for-purpose – models of human activities? We do not know yet. There have been important successes, for example, using network dynamics in predicting the spreading of diseases, but whether complex systems science can provide a meaningful model of a global transition to a sustainable economy, is impossible to tell.  However, Complex Systems Science is still a young, expanding  scientific field and many discoveries lie ahead.


In April-June 2019, the CCSS will host three discussion events focusing on the potential role of complex systems science in different societally relevant areas. Each event consists of a presentation by a leading expert, followed by a plenary discussion. The aim is to get a clearer picture of pressing issues, identify promising complexity-related approaches to these problems, and of course to exchange ideas with fellow complexity scientists. 


  • April 26th, 12:00-14:00: Complex Interactions in Environmental Economics(lead scientist: Bert de Vries)  
  • June 5th, 11:00-13:00: Decision making under Uncertainty(lead scientist: Jan Kwakkel)
  • June 14th, 12:00-14:00: Self-Organisation in Communication, Traffic and Energy(lead scientist: Hans van den Berg)

All interested researchers are welcome, but please register using the links above.