An Agent-Based Model for a collective, group-based decision-making process.

This thesis aims to provide a qualitative analysis of the influence of various factors in a collective, group-based decision-making process, where the group must establish its priorities regarding a multi-objective problem. To this end, an Agent-Based Model is developed to simulate the discussion process, grounded in the Complex Adaptive Systems (CASs) approach. In the first chapter, the theoretical approach of CASs is presented, with Agent-Based Modeling (ABM) as the implementation method. Its relevance in describing human-oriented organizational systems is highlighted through a literature review, with a focus on collective decision-making process. In the second chapter, the multi-objective problem at the center of the discussion is outlined. It is a simplified representation of an investment problem (e.g., in terms of economic or human resources) in short- and long-term impact policies within the context of a company. Their combination defines the company's chosen strategy. The three objectives are the strategy's effectiveness after one, two, and three years. Given the three-component vector of a decision maker's priorities, the chosen strategy and the corresponding effectiveness at the end of the three considered time periods are determined. The Global Utility function weights these impacts according to the decision maker’s priorities, providing a measure of their satisfaction. The definition of the Agent-Based Model is outlined in the third chapter. An important focus is placed on the Agents' (i.e., decision makers') learning and adaptation process. It consists of evaluating and responding to solutions proposed by others (Agents' reactivity). This leads to the individual reconstruction of the Global Utility function that each Agent wants to maximize (Agents' proactivity). The overall structure and results provided by the model are then discussed. This thesis originated from a collaboration with the company Nova Analysis, and a strong emphasis has been placed on the model's usability as an educational tool. As a result, it is straightforward to create a group with various number of Agents, combinations of individual attributes (e.g., varying degrees of Agents' talkativeness and openness), and hierarchies (e.g., democracy, oligarchy, or a single leader). Moreover, it is also possible to follow Agents' paths during the discussion, in terms of individual learning processes and relationships among them. Furthermore, the effect of Agents' openness toward other opinions has been studied in an aggregate way by considering three kinds of groups with varying degrees of Agents' average openness and different sizes, all with an internal democratic structure. The main findings are that groups benefit from Agents' openness, regardless of size, in terms of the quality of the group solution (i.e., its Global Utility) and level of agreement. Concurrently, in the case of high average openness, larger groups achieve better results than smaller ones. On the other hand, smaller groups perform more effectively in the case of lower average openness. Moreover, the case of a single leader has been considered in two variants: a high-openness leader and a low-openness leader. The first case outperforms the second in both solution quality and Agents’ satisfaction. Open single leadership is then presented as a possible alternative to enhance group performance in larger groups in the case of lower average openness. Finally, some possible extensions of the model are outlined.