LARA

Lightweight Architecture for boundedly Rational citizen Agents

Target

We develop a lightweight agent architecture which is intended for the simulation of citizens’ decision making in the arena of policy modeling. Being a Java library that integrates well with existing ABM frameworks such as Repast Simphony, LARA fills a gap between such frameworks with no built-in psychological foundations and full-fledged cognitive architectures. Its components of perception, memory, preprocessor, decision-making, and postprocessor interact via interfaces and thus, the implementation of each one can be adapted or replaced independently.

  • Perceptions of the biophysical, socioeconomic (i.e. political, legal or economic matters) and social environments (e.g. social networks which support social appraisal, comparison, influence, and opinion dynamics) are stored in an agent’s memory which supports limited capacity management and sophisticated recall of content.
  • The preprocessor chooses one of several decision modes (habit, applying heuristics such as a decision tree, or comprehensive evaluation) and prepares behavioural options the agent may decide on (e.g. recalling currently applicable options from memory and calculating their utility considering current environmental factors).
  • The decision-making component performs the agent’s selection process. For instance, it ranks the behavioural options according to their utility with respect to the agent’s preference structure that is defined for the goals the agent pursues.
  • Among other purposes, the postprocessor can be used to evaluate the number of behavioural options an agent has at its disposal at any given time and how well these options perform with respect to the agent’s preferences. Such an assessment of the freedom of action that citizens have under different scenarios can serve as an important indicator for political decision makers.

Result

As a start, LARA is used in simulations of climate adaptation measures and diffusion of changed patterns of behaviour in the face of climate change with several thousand agents. Its scope might be far more general, though. The code is going to be published under the Gnu Public Licence (GPL) and thus applicable in wide range of social simulation.