Approaches and conceptualisations
2.5. Agent-based modelling (ABM) of NRM systems
2.5.2. Agent-Based Models
Agent-based models have emerged alongside cellular automata models and involve diverse and heterogeneous agents that can interact with one another and with their environment. They allow the study of the local and the small scale in spatial systems as well as giving an overview of patterns and concepts in complex, interconnecting systems. The idea of ABM is to “understand properties of complex systems through the analysis of simulations” (Axelrod, 1997:3) with simulations involving “individual actions of diverse agents” and the measurement of the “resulting systems behaviour and outcomes over time” (Crooks and Heppenstall, 2012:86). The communication and interaction of agents between each other and their environment (O’Sullivan et al., 2012) is a key facet of the agent based modelling approach (Millington and Wainwright, 2016). The emphasis is on the ‘bottom-up’ (Epstein, 2006) approach of ABMs, which is central to its opposition to aggregate mathematical models with central controls on large homogenous populations.
2.5.2.1. The concept of an agent
At the most basic level Millington and Wainwright (2016:5) describe an agent as “an object with defined attributes capable of executing functions autonomously”. Based on the original ideas of Wooldridge and Jennings (e.g. Wooldridge and Jenninngs, 1995; Wooldridge, 1999). Jennings et al. (1998:276) state that: ‘an agent is [essentially] a computer system, situated in some environment
65 that is capable of flexible autonomous action in order to meet its design objectives’. Crooks and Heppenstall (2012) list that common attributes of an agent are:
Autonomy: they are free to interact with other agents and there are no central controls on agents except for the influence of social norms and institutions that have built up through previous agent interaction (Epstein, 2006).
Goal directed: each agent has a set of goals to fulfil.
Reactive: they have a sense of their surroundings and can react to changes.
Bounded rationality; each agent’s behaviour is based on the rational choice paradigm (Axelrod, 2007), where agents make choices that are adaptive and inductive in order to move towards their goal, but which are bounded by the use of only local information to inform choices.
Interactive: agents can communicate with others.
Mobility: agents are free to roam through the space.
Adaption/ learning: agents can be programmed to change their state depending on previous states to simulate a learning process.
Agents can be organisms, humans, businesses, institutions, or any other entity that pursues a certain goal (Railsback and Grimm, 2012) and can be considered as the ‘people’ of artificial societies (Epstein and Axtell, 1996). Each individual is placed in a simulated environment, and each has a set of states, some of which are fixed for the agent’s life and some of which can be changed. Through the concept of ‘weak’ agency (as opposed to strong agency involving emotions and human cognitive characteristics in artificial intelligence), the interaction of agents and their associated and changing states and the states of the local environment can represent systems ranging from the operation of markets, movement of traffic, animals in an ecosystem or behaviour amongst government institutions, among others (O’ Sullivan, 2008).
2.5.2.2. Early models and the growth of ABMs
One of the first applications of the concept of ABMs in social science was by Thomas Schelling in the 1960s and 1970s, with his classic series of papers: ‘Models of Segregation’ (1969), ‘On the Ecology of Micromotives’ (1974), and ‘Dynamic Models of Segregation’ (1971). Among these studies Schelling created a spatially distributed model of the composition of neighbourhoods, which involved agents having some level of preference for their neighbours being of the same
‘colour’ as them and moving neighbourhoods accordingly to maintain the preferred ratio of neighbours of the same colour as themselves. He found that even fairly ‘colour blind’ preferences produced segregated neighbourhoods (Epstein and Axtell, 1996). The experimentation with small
66 scale social rules to observe the emergent patterns was a new way to do social research and opened up issues for debate and further research that might not previously have been identified in the same way.
Early use of ABM in terms of social-environmental systems experimented with the combination of complex interdisciplinary systems. For example the SugarScape model (Epstein and Axtell, 1996) is made up of a spatial distribution of a resource (“sugar”) that agents use as food, which is relatively rich in some places in the landscape and relatively impoverished in others. Agents in the system have metabolism, vision and ability to reproduce and follow simple local behavioural rules that equate to searching for the richest areas of sugar, travelling to them and consuming them.
Movement uses up energy and when energy is depleted the agent dies. The purpose of Epstein and Axtell creating this hypothetical situation was to observe the emergent behaviour of the
‘society’ that they had created through the simple behavioural rules. They observed that the concept of ecological carrying-capacity was important and evident, equally that when seasons were introduced the concept of migrant communities emerged. They also found that patterns of distribution of wealth (in terms of sugar collection) emerged and found that they appeared to mirror society in that there was a distinct skew, where most of the population had relatively little wealth, demonstrating that there are similarities between human societies and the artificial society in SugarScape (Epstein and Axtell, 1996).
ABM models can also be used in a more specific sense to try and understand patterns of past societies and can, according to Epstein (2006:12) provide a “powerful new way of doing empirical research”. In a study attempting to reconstruct the Anasazi population dynamics, who lived in Arizona between 800-1300AD, and who disappeared from the valley after that time, Dean et al.
(2000) used ABMs as an experimentation ground for possible theories. They derived that it could have been predominantly non-environmental, sociological and ideological factors that were responsible for the complete abandonment of the area. Dean et al., acknowledge that although the model may never explain the ‘real’ history, it provides an instrument for making progress in a replicable and cumulative way in formulating principles and hypotheses about systems of interest.
2.5.2.3. Realism and simplicity
One issue that is constantly apparent in ABMs and throughout their development has been the issue of realism. Models are always an abstraction of some ‘real’ system and in general, most modellers appear to hold realist ontologies, in that they believe that there is some reality out there, for example, O’ Sullivan (2008) states that modellers assume that there is at least some truth that
67 is truer than others. Although there is recognition now that there are limits on the understanding that can be gained about that reality from models because the conception is influenced by human interpretation (cautious realism): each model is only an interpretation of a reality, and can be manipulated and constructed by the researcher (O’ Sullivan, 2008). Models, therefore, are often described as falling along the scale of realism, from essential to complex (Dietrich et al., 2003) and with each scale of realism comes a different function or development. Davidsson (2002) states that ABMs are most appropriate in situations where decisions are concentrated on particular locations, in which the structures and patterns of the observed actions are seen as emergent. This PhD research takes a cautious realism stance and expects a model to be a representative of an interpretation of one view of reality and acknowledges the role of construction in the interpretation of the model.