A Behavioural Decision-Making Framework For Agent-Based Models

Abstract

In the last decades, computer simulation has become one of the mainstream modelling techniques in many scientific fields. Social simulation with Agent-based Modelling (ABM) allows users to capture higher-level system properties that emerge from the interactions of lower-level subsystems. ABM is itself an area of application of Distributed Artificial Intelligence and Multiagent Systems (MAS). Despite that, researchers using ABM for social science studies do not fully benefit from the development in the field of MAS. It is mainly because the MAS architectures and frameworks are built upon cognitive and computer science foundations and principles, creating a gap in concepts and methodology between the two fields. Building agent frameworks based on behaviour theory is a promising direction to minimise this gap. It can provide a standard practice in interdisciplinary teams and facilitate better usage of MAS technological advancement in social research. From our survey, Triandis' Theory of Interpersonal Behaviour (TIB) was chosen due to its broad set of determinants and inclusion of an additive value function to calculate utility values of different outcomes. As TIB's determinants can be organised in a tree-like structure, we utilise layered architectures to formalise the agent's components. The additive function of TIB is then used to combine the utilities of different level determinants. The framework is then applied to create models for different case studies from various domains to test its ability to explain the importance of multiple behavioural aspects and environmental properties. The first case study simulates the mobility demand for Swiss households. We propose an experimental method to test and investigate the impact of core determinants in the TIB on the usage of different transportation modes. The second case study presents a novel solution to simulate trust and reputation by applying subjective logic as a metric to measure an agent's belief about the consequence(s) of action, which can be updated through feedback. The third case study investigates the possibility of simulating bounded rationality effects in an agent's decision-making scheme by limiting its capability of perceiving information. In the final study, a model is created to simulate migrants' choice of activities in centres by applying our framework in conjunction with Maslow's hierarchy of needs. The experiment can then be used to test the impact of different combinations of core determinants on the migrants' activities. Overall, the design of different components in our framework enables adaptations for various contexts, including transportation modal choice, buying a vehicle or daily activities. Most of the work can be done by changing the first-level determinants in the TIB's model based on the phenomena simulated and the available data. Several environmental properties can also be considered by extending the core components or employing other theoretical assumptions and concepts from the social study. The framework can then serve the purpose of theoretical exposition and allow the users to assess the causal link between the TIB's determinants and behaviour output. This thesis also highlights the importance of data collection and experimental design to capture better and understand different aspects of human decision-making.