Press release from December 12, 2005
"The world is not enough"
Leipzig. An international team of scientists has developed a new strategy for representing agent-based systems of all kinds sufficiently realistically that developments can be understood and predicted with the help of computer models. The scientists in the team led by ecologist Volker Grimm of the Centre for Environmental Research Leipzig-Halle (UFZ) present this form of modelling in Science (Vol. 310, Issue 5750 of 11 November 2005). According to the article, pattern-oriented modelling is not limited to environmental research, but can be used everywhere where there are agent-based complex systems, for instance in financial markets and traffic or town planning.
James Bond as a model
Agents are everywhere: customers in the supermarket, stock brokers on the stock exchange, trees in the forest. Scientists have adopted the term agent to describe individuals of all kinds so that they can be modelled, in order to gain a better understanding of processes in nature and society. When they hear the word agent, many people think of James Bond, the agent on Her Majesty’s secret service. But what is it that makes 007 an agent? "He has a mission, but it’s up to him how he performs it. He operates autonomously and adapts his decisions to the situation. And that is what we assume for ecological agents, individual organisms. They have a mission to pass on their genes, i.e. to produce as many surviving progeny as possible", says Dr. Volker Grimm. "But the organisms must decide for themselves how to achieve this aim: if I’m hungry, I have to take the risk of being eaten while searching for food. If I’m full, I’d better stay safe." What all agents, from James Bond to motorists to brown bears, have in common is that they are discrete autonomous units that follow a goal and adapt their decisions to their environment.
No regularity – no science
Agent-based systems are usually so complex that they can be described only in individual aspects. For instance, no one is capable of simulating on the
computer the complete development of a metropolis with a population in the millions with all its social and economic aspects. Individual aspects, like water
consumption, however can be represented in models. But like everywhere in life, it’s the details that are problematic. Models are only really usable if they
are not too complex. Even in his day, Albert Einstein is supposed to have said: "Everything should be made as simple as possible, but no simpler". Social
scientists in particular often have scruples about describing humans in simple terms, according to Volker Grimm. This leads to models that are either difficult
to understand or which are so abstract it is not clear how they relate to reality. The new strategy described in the Science article leads to models that are
realistic, yet simple enough to be understood.
Using patterns is a way of focusing on the most essential information about the internal organisation of complex systems. A pattern is any kind of observation that reveals non-random structures and therefore contains information about the mechanisms that form them. Complex systems contain patterns on various hierarchical levels and scales. In ecosystems, for instance, patterns can be observed in species composition, regional structures or the behaviour of individual organisms. "Science begins when I see a regularity. If complex systems were just complex and chaotic I would not be able to practise the science of complex systems. We need to learn to see the patterns. What is special about complex systems is that concentrating on one pattern is no longer enough." It is, therefore, not only a question of seeing the forest for the trees, i.e. grasping the patterns that characterise the forest, but also vice-versa: seeing the trees for the forest and describing their behaviour appropriately.
Trees as agents
One example of such a model is a simulated primeval forest. No one knows what the primeval forests that the Teutons found looked like. There are no original
virgin forests left in Germany. So how can we make forests as close to the original as possible? And how should forest management be designed to make it as
close to nature as possible? These are questions that are difficult to answer using practical experiments. After all, in a beech forest, a complete cycle of
growing, flourishing and decomposition takes around 250 years. Computer models can solve this dilemma. Patterns from remains of virgin forests in Slovakia and
the vast knowledge of foresters concerning the fate of individual beech trees were used to develop a simple, yet realistic primeval beech forest model.
The amount of work required to develop and test these kinds of models can be considerable and take years, as in the case of a model for tropical forests with their large number of species. But it is worth it. "We use the tested models as virtual laboratories", explains Grimm’s colleague Dr. Andreas Huth. "With the tropical forest model we can go through the model forest with an imaginary chainsaw every 10, 20 or 30 years and harvest a certain percentage of the trees. Then we simply watch to see how the forest reacts. In this way we discover how the forest can be managed sustainably, so that future generations will be able to use it too, whether for timber or other purposes." The UFZ scientists have also designed pattern-oriented models to predict the spread of brown bears in Austria, the reintroduction of the lynx in Germany and the rabies risk from infected foxes. In the rabies model, for instance, the idea is to determine the best control strategy and to investigate how many rabies baits are necessary. Rabies has been almost entirely eradicated in Germany. Nevertheless, there is still a risk that it could be introduced again by infected foxes from neighbouring countries. The foxes in Eastern Europe are also to be vaccinated using rabies baits over the next few years. The UFZ scientists want to use a computer model to test the most cost-effective way of doing this.
A new systems theory?
"There have been repeated attempts to develop a general systems theory. They were often influenced by extremely simplistic mathematical approaches." This
approach is too inflexible for agent-based systems, according to Volker Grimm, who sees the new strategy as a model example: "We were interested in thinking
outside the ecological box and saying to colleagues from other scientific disciplines: We have a strategy. It can also be used for systems with human agents
to produce models that reveal the inner structure of these systems so that we can understand and control them better." Grimm is optimistic as regards a general
theory of agent-based systems: "Pattern-oriented modelling is only just starting to gain ground and we are expecting rapid development in the future." Our
conception of "understanding" will probably have to be modified for a new systems theory. In any case, American social scientists Epstein and Axtell suspected
as long ago as 1996 that we might one day have to interpret the question: "Can you explain it?" as "Can you grow it?", i.e. are we able to "grow" the thing
to be explained on the computer? Pattern-oriented modelling will contribute significantly to this "growing" of virtual worlds.
Thanks to James Bond.
Further information on this subject is available from
Doris Böhme / Tilo Arnhold,
Public Relations of Helmholtz Centre for Environmental Research – UFZ