Emergent Tendencies in Multi-Agent-Based Simulations: Using Constraint-Based Methods to Effect Practical Proofs Over Finite Subsets of Simulation Outcomes
Centre for Policy Modelling Discussion Papers
296 Pages Posted: 29 Nov 2001
Date Written: 2001
This thesis suggests a methodology for studying complex systems. This method is intended to be particularly useful for searching and proving tendencies (whether considered emergent or not) in those systems whose dynamics seem to be strongly dependent on the system's components' interaction (such as social systems). These systems are commonly simulated in Multi-Agent Systems (MAS).
It begins by examining the formal notions of simulation, modelling, and theorem-proving. Then it reviews some notions of complexity and proposes a notion of the emergence of tendencies as based on the trade-off between subjective and objective factors of complexity. It next moves on to investigate the dynamics of a system via a platform consisting in a (logical) model constraint-based exploration of the dynamics of a simulation. This platform is suggested for systematically exploring the subspace of simulation trajectories associated with a range of parameters of the model, a range of choices of the processes (e.g., agents' choices), and the logic of the simulation program. Following this, we suggest using this architecture in addition to the higher architectural level given by a MAS and an even lower level, a syntactic constraint-based architecture, as complementary means to investigate aspects of the dynamics of a MAS simulation. The proposed methods are compared with other approaches for exploring the dynamics of a simulation. In particular, differences in terms of the notions of morphism among models, the generality of the conclusions, and the measures of behaviour that each approach allows are emphasised. In addition, enveloping the simulation outputs is proposed as an alternative to statistical summaries. This seems to be especially convenient for studying complex systems and for analysing outputs in case of applying theorem-proving techniques.
This model constraint-based architecture is applied to a MAS-based model exemplifying a typical interaction trader-distributor. A tendency is identified in the MAS-based model and then a constrained proof is performed in the model constraint-based architecture. Afterwards, some implications of this thesis for related areas of science are reviewed. Finally, the appendices include an analysis of the complexity of this model constraint-based exploration of trajectories and two papers particularly relevant to the social simulation and the MAS communities.
Suggested Citation: Suggested Citation