Agent-based approaches in social prototyping and software development : synthesizing core elements for implementation and gaps of evaluation methods

Abstract

This thesis synthesizes the core architectural elements of agent-based modelling in social prototyping and AI systems for software development by studying the implementation architectures and the possible gaps that existing evaluation methods for the implementation of agent-based modelling should address in overcoming the criticisms. Publications and public information on implementation architectures, evaluation methods and implementation criticisms were studied, and common patterns were synthesized to come up with theories on core architectural elements and evaluation gaps. Both fields have two architectural elements in common – agent and software scaffold. Agent has a promoted LLM and an intelligence part in software development systems and agent-based modelling, respectively. The software scaffolding supports efficient agent operations. Software scaffolds are outside the agents in software development systems, both inside and outside the agents in agent-based modelling. The evaluation methods in agent-based modelling do not address the efficiency of change or intervention implementation in simulated societies, and the correctness of the theory on which the simulation is based. The findings of this study are synthesized from recurring patterns that provide hypotheses which can be empirically validated in further research.