Journal on Policy and Complex Systems
other agents , or itself ” ( Wilensky & Rand , 2015 , p . 209 ). We can think of each rule as characterized by a receptor IF and an executor THEN . To make an example , if there is an election , a citizen will use the detector “ IF today is an election day ” to execute the effector “ THEN go to vote .” The rule could also be more precise and use multiple detectors and effectors at the same time . Because of adaptive learning , agents do not always execute the same rule when faced with the same input ; instead , they learn from their past and “ change their behavior in the future to account for this learning ” ( Wilensky & Rand , 2015 , p . 230 ).
To understand system-wide behavior , it is essential to have a clear understanding of the granularity of the system , which is composed of agents , meta-agents , and sub-agents , 9 with agents being the fundamental generators of the emergent behavior of interest . Meta-agents can be quasi-bounded and possess certain degrees of autonomy ; however , their behavior can only be inferred as an outcome of agents ’ actions and interactions ( Harrison , 2006a , p . 9 ). For example , the behavior of a nation-state in the international system cannot be inferred outside of the context that has generated it . Social organizations are neither self-contained nor autonomous . They are meta-agents generated by the primary agent of social systems : humans . In turn , agents are interdependent , and their behavior can only be understood from a system-wide perspective that takes into account coevolutionary dynamics .
In other words , thinking in terms of adaptive agents requires us to consider the international system as a complex whole . A whole that cannot be divided into levels of analyses , as IR scholars often like to do . As Wilensky and Resnick ( 1999 ) wrote , levels of analysis are used to provide three different kinds of views : organization-chart view , container view , or emergent view . The first one is used to think of structural hierarchies within institutions , companies , or organizations . Levels , in this case , serve to conceptualize chains of command and organization-charts . The container view , widely used in IR scholarship , “ is based on the idea of parts and wholes ” ( Wilensky & Resnick , 1999 , p . 5 ). This view “ differs from the organization-chart view ” because “ the lower-level elements are parts of the higher-level elements ” ( Wilensky & Resnick , 1999 , p . 5 ). For example , a month is part of a year ( container view ), but an employee is not part of the employer ( organization-chart view ).
The third view is similar to the container view but focuses on “ levels that arise from interactions of objects at lower levels ” ( Wilensky & Resnick , 1999 , p . 5 ). The difference between emergent and container view is subtle . After all , one might argue that just as a week is made of days , a state is made of people . However , the state – people relationship is significantly different . For a start , the composition of the state keeps changing ; people leave , come , and die , so do companies , organizations , institutions , and laws . Furthermore , the state
9 Agents are composed of sub-agents , and meta-agents are composed of agents . In agent-based modeling , it is often said that “ it ’ s agents all the way down ” ( Wilensky & Resnick , 1999 , p . 1 ).
122