CNH: Fine-Scale Dynamics of Human Adaptation in Coupled Natural and Social Systems: An Integrated Computational Approach Applied to Three Fisheries

本项目的目的是更好地了解个体渔民之间的竞争如何导致出现与资源养护相一致（或不一致）的私人激励和非正式社会安排。这些非正式安排和激励措施很重要，因为它们帮助我们了解私人利益可能在多大程度上加强或削弱正在进行的资源管理，从而了解耦合的人类和自然系统的可持续性。推动这项研究的广泛假设是，渔民之间的竞争互动所产生的非正式社会结构反映了自然系统的特殊情况。在某些情况下，成功的竞争需要秘密的非合作行为;在其他情况下，合作往往会产生更好的竞争结果。这些不同的结果对资源管理的可行性和有效性有不同的影响，但并不总是明显的。我们认为，相关的人类社会过程是这样一个过程，在这个过程中，个体通过耗时且昂贵的方式获得关于复杂资源的有价值的知识，从而相互竞争。为了成功地竞争，个人必须在利用他们目前掌握的知识所带来的直接利益与探索新知识的昂贵需求之间取得平衡;此外，在寻求新知识时，个人必须在通过合作或自主搜索获得知识的成本和收益之间取得平衡。为了模拟这种竞争过程，我们采用了一种从计算机科学中借来的技术，称为学习分类系统（LCS）。LCS使用遗传算法来模拟智能体（这里是渔民）使用他的经验来不断完善他对自然和社会环境的知识和决策的方式。 LCS的重要性在于，它允许模拟自利渔民的共同发展战略互动，这些渔民仅部分了解他们正在开发的资源状况以及与他们竞争的渔民。理解这些竞争动态的问题在几乎所有耦合的自然和人类系统中都很明显。我们采用的方法进行比较研究的缅因州渔业的特点是显着不同的时间和空间动态-海胆，龙虾和鳕鱼的三个海湾。 每个渔场将使用自然系统的生物物理模拟器和模拟渔民学习和互动的紧密集成的多智能体学习分类器系统进行建模。每个模型的设计将部分基于对渔民的广泛采访，了解他们对所从事渔业动态的了解。我们将使用这些模型来探讨每个渔业过去和未来的政策问题。除了这些探索的直接适用性之外，我们希望这个项目将为在其他耦合系统中更广泛地使用多智能体学习模型提供基础。项目成果将定期传达给行业和管理人员。主要研究人员包括经济学家、生物学家、人类学家和计算机科学家。所有的PI都有多年的缅因州湾渔业经验，并与个别渔民和管理人员建立了良好的关系。一个海洋政策的硕士生，一个博士。该项目将雇用一名计算机或海洋科学专业的学生和一名计算机科学博士后研究员。此外，该项目还将开设一门复杂的适应性社会生态系统的本科课程，并在同一领域举办一个研究生/教师讲习班。

The purpose of this project is to gain a better understanding of the way competition between individual fishermen lead to the emergence of private incentives and informal social arrangements that are (or are not) consistent with conservation of the resource. These informal arrangements and incentives are important because they help us understand the extent to which private interests might strengthen or weaken on-going resource management and, consequently, the sustainability of coupled human and natural systems. The broad hypothesis driving the study is that the informal social structure that emerges from competitive interactions among fishermen reflects the particular circumstances of the natural system. In some cases, successful competition requires secretive non-cooperative behavior; in others, cooperation tends to yield better competitive results. These different outcomes have different, and not always obvious, impacts on the feasibility and effectiveness of resource management. We think of the relevant human social process as one in which individuals compete with one another through time-consuming and costly acquisition of valuable knowledge about a complex resource. To compete successfully, individuals must balance the immediate benefits that come from exploiting knowledge they currently hold with the costly need to explore for new knowledge; additionally, when seeking new knowledge, individuals must balance the costs and benefits of acquiring knowledge through cooperation or through autonomous search. In order to model this kind of competitive process, we employ a significantly modified version of a technique borrowed from computer science called a learning classifier system (LCS). LCS uses a genetic algorithm to mimic the way an agent (here a fisherman) uses his experience to continuously refine his knowledge and decisions about his natural and social environment. The importance of LCS is that it permits simulation of the co-evolving strategic interactions of self-interested fishermen who are only partially informed about the state of the resource they are exploiting and the fishermen with whom they compete. The problem of understanding these kinds of competitive dynamics is evident in almost all coupled natural and human systems. We apply the approach to a comparative study of three Gulf of Maine fisheries which are characterized by significantly different temporal and spatial dynamics - sea urchins, lobster and cod. Each fishery will be modeled using a biophysical simulator of the natural system and a tightly integrated multi-agent learning classifier system that simulates the learning and interactions of fishermen. The design of each model will be based in part on extensive interviews with fishermen about their knowledge of the dynamics of the fisheries in which they work. We will use these models to explore past and prospective policy problems in each fishery. Beyond the immediate applicability of these explorations, we expect this project will provide a foundation for the wider use of multi-agent learning models in other coupled systems. Project outcomes will be transmitted regularly to industry and managers. Principal investigators include economists, biologists, anthropologists and computer scientists. All the PIs have years of experience in the fisheries of the Gulf of Maine and have well developed relationships with individual fishermen and managers. A masters level student in marine policy, a Ph.D. student in computer or marine science and a post-doctoral researcher in computer science will be employed on the project. In addition, the project will develop an undergraduate course in complex adaptive social-ecological systems and a graduate student/faculty workshop in the same area.