基于智能体的犯罪倾向动力学
Agent-based dynamics of criminal propensity
浏览论文内容
中文总结 AI 辅助
该研究将进化犯罪学的RRM形式化为智能体动力学系统，扩展有界置信意见动力学并证明收敛条件，发现系统主导行为为持续振荡，揭示了不同环境感知下的群体倾向及两极分化机制。
中文摘要 AI 辅助
报应与互惠模型（Retribution and Reciprocity Model, RRM）是进化犯罪学中提出的一种新框架，用于通过合作视角理解犯罪成因。我们将RRM形式化为一个基于智能体的动力学系统，其中犯罪倾向通过成对互动和对他人互动的观察而演变。该数学模型扩展了有界置信意见动力学，包含两个新特征：代表环境感知的智能体特定吸引子，以及第三方目击者的参与。我们证明个体倾向保持有界，并建立了收敛到极值和智能体环境感知的充分条件。模拟显示，系统的主导行为并非收敛，而是持续振荡，这是有界置信模型中罕见的现象，此前需通过自适应置信界获得，而本模型无需该界即可产生。我们推导了振荡的必要条件，发现具有均匀但中性环境感知的群体倾向于两极分化至极端，而均匀非中性感知则倾向于达成共识。两极分化机制局限于近中性感知的窄带，其宽度随群体规模减小而减小，随互惠和报应倾向强度增大而增大。
英文摘要
The Retribution and Reciprocity Model (RRM) is a novel framework presented in evolutionary criminology to understand causes of crime through the lens of cooperation. We formalise RRM as an agent-based dynamical system in which criminal propensity evolves through pairwise interactions and through observations of others' interactions. The mathematical model extends bounded confidence opinion dynamics with two novel features: an agent-specific attractor representing the perception of the environment, and the participation of third-party witnesses. We prove that individual propensities remain bounded and establish sufficient conditions for convergence to an extreme value and to the agent's perception of the environment. Simulations show that the dominant system behaviour is not convergence but persistent oscillations, a rare phenomenon in bounded confidence models arising here without the adaptive confidence bounds through which it has previously been obtained. We derive a necessary condition for oscillations, and find that a population sharing a uniform but neutral perception of the environment tends to polarise to the extremes, while a uniform non-neutral perception tends to yield consensus. The polarising regime is confined to a narrow band of near-neutral perceptions, whose width decreases with population size and increases with the strength of reciprocal and retributive tendencies.
