The use of agent-based simulation to assist in environmental management and decision-making has increased markedly in recent years. Simulation models representing various facets of human behaviour have emerged (Hare and Deadman 2004). Social scientists use this kind of simulation for several purposes, including the discovery of collective regularities. Some are built on strong behavioural foundations, where mental models of the silicon agents are based on empirical data or stakeholders’ views. In other cases, agents’ strategies are selected stochastically or even arbitrarily and the outcomes tested in computational experiments.

What is arbitrarily decided in such simulations is how to drive the artificial learning process that enables each agent to adapt to other agent strategies interactively and to decide whether and how to modify its own strategy over time. Imitation of other (e.g. nearby/visible) agents is a simple and common method of introducing adaptation over time. In other simulations, agents choose from a suite of heuristics or mental models made available to some or all of them. More sophisticated learning methods (such as neural networks and genetic algorithms) have also been adopted.

Largely overlooked in the simulation work to date is the fact that the intrinsic structure of the system being simulated can have a direct influence on ways in which the human agents are likely to behave, i.e. react and adapt over time. Some social systems that fall into this category are called self-referential problems—situations in which the forecasts made by agents serve to create the very world they are trying to forecast. In these complex systems, the best thing to do depends on what everyone else is doing.

The purpose of this paper is to look briefly at a special class of self-referential problems called self-defeating systems. A typical example is the bar problem, in which a simulated population of attendees oscillates in an apparently random manner around a critical congestion level. It is suggested that several resource management problems involving human interactions with ecosystems have a self-defeating character, posing new challenges for integrated resources management. An example is the potential over-fishing of fisheries, behavioural aspects of which are discussed in the paper and likened to a minority game. Such commons problems will require new institutional arrangements to overcome the self-defeating character of these situations.