Collaborative Research: CRCNS Research Proposal: Adaptive Decision Rules in Dynamic Environments

做决定的时候，可能是在哪里吃午饭这样简单的事，也可能是从事什么职业这样复杂的事，时间往往花在考虑不确定的证据和可能的结果上。这种深思熟虑的过程是行为和认知灵活性的核心，但需要克服一个重大挑战：决定何时停止深思熟虑，并承诺采取行动。许多用于研究人类决策和实现机器决策的模型都假设，当累积的证据达到一个固定值或界限时，就会发生决策承诺。这些“累积到绑定”的模型，其在上个世纪的发展帮助形式化和改进了与代码破译、制造和其他现实世界应用程序相关的决策，也为理解大脑中的决策承诺提供了一个有用的起点。然而，固定边界在固定条件下是最合适的，这在实验室实验中经常使用，但在现实世界中很少遇到。这项研究的目标是超越决策承诺的“固定界限”形式，而是考虑大脑使用更灵活的方式来结束审议并做出决定。这项研究从一个新的、以数学为基础的理论开始，该理论描述了在不断变化的条件下使用灵活的承诺形式的优势。该项目的目标是利用这一理论来设计和解释实验，这些实验将提供一个全面的新观点，即人类大脑是如何做出决定的，即使这些决定是基于在不确定和变化的条件下进行的审议。这项工作将为来自不同背景的本科生和研究生提供跨学科的数学、认知科学、心理学和神经科学的交叉训练。该小组将利用与研究相关的活动来鼓励代表性不足的群体参与科学。此外，该团队将开发和传播资源，包括新的数据集和分析工具，这将有利于大脑如何做出决定的研究和教育。最后，该团队将提高公众对计算神经科学的认识，并解决提高科学素养和理解科学如何造福社会的迫切需要。这将通过公开讲座、向全国公共广播电台播放的节目“我们聪明才智的引擎”捐款、大脑意识周活动以及向非专业人士解释大脑研究的网站捐款来实现。深思熟虑的决定将大脑从反射性处理的即时性中解放出来，但也带来了一个关键的挑战：大脑如何决定何时停止深思熟虑，并致力于采取行动？我们对这一承诺过程的理解一直被一个计算框架所主导，该框架假设一旦积累的证据达到预定义的水平或界限，决策就会终止。这些“累积到约束”的模型与规范理论有着密切的联系，可以解释一系列行为和神经方面的发现。然而，它们只有在许多决策研究中使用的高度限制性条件下才是最优的，在这些条件下，证据的信息量、获取率和其他特征是稳定的，并且是事先已知的。目前尚不清楚大脑是如何在更普遍的情况下平衡决策、审议和承诺的，因为暂时延长的决策必须与我们动态和不确定的世界相抗衡。目标是促进我们对大脑在这些条件下使用的决策规则的理解。pi从一个新的理论基础开始，包括灵活的决策边界，这些决策边界不是预定义的，而是可以在决策形成时更新，即使在变化的条件下也可以优化性能。他们使用这个框架来指导人类行为研究的设计和分析，并结合非人类灵长类动物的行为和神经生理学研究，他们用这些研究来检验他们的主要假设，即灵长类动物的大脑使用动态的、适应性的规则来支持在变化和不确定的环境中做出理性决策。该项目由数学与物理科学理事会（MPS）下属的数学科学部（DMS）、计算机与信息科学与工程理事会（CISE）下属的信息与智能系统部（IIS）以及社会、行为和经济科学理事会（SBE）下属的行为与认知科学部（BCS）共同资助。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

When decisions are made, which can be as simple as where to have lunch or as complicated as what career to pursue, time is often spent deliberating over uncertain evidence and possible outcomes. This deliberation process is central to behavioral and cognitive flexibility but requires overcoming a major challenge: determining when to stop deliberating and commit to a course of action. Many models that have been used to study human decision-making and to implement machine decision-making assume that decision commitment occurs when the accumulated evidence reaches a fixed value or bound. These “accumulate-to-bound” models, whose development last century helped formalize and improve decisions related to codebreaking, manufacturing, and other real-world applications, have also provided a useful starting point for understanding decision commitment in the brain. However, fixed bounds are most appropriate under fixed conditions, which are often used in laboratory experiments but rarely encountered in the real world. The goal of this study is to move beyond the “fixed bound” form of decision commitment and instead consider more flexible ways the brain uses to end deliberating and arrive at a decision. The research starts with a new, mathematically grounded theory that describes the advantages of using flexible forms of commitment under changing conditions. The goal of the project is to use this theory to design and interpret experiments that will provide a comprehensive new view of how human brains commit to decisions even when they are based on deliberations that occur during uncertain and changing conditions. This work will provide interdisciplinary training at the interface of mathematics, cognitive science, psychology, and neuroscience for undergraduate and graduate students from diverse backgrounds. The team will use research-related activities to encourage the participation of underrepresented groups in science. Also, the team will develop and disseminate resources, including novel datasets and analytic tools, that will benefit research and education in how the brain makes decisions. Finally, the team will increase public awareness of computational neuroscience and address the urgent need to increase scientific literacy and understanding of how science can benefit society. This will be accomplished via public lectures, contributions to the program “Engines of Our Ingenuity” broadcast by National Public Radio stations nationwide, Brain Awareness Week activities, and contributions to a website that explains brain research to non-specialists. Deliberative decisions free the brain from the immediacy of reflexive processing but pose a critical challenge: how does the brain decide when to stop deliberating and commit to a course of action? Our understanding of this commitment process has been dominated by a computational framework that assumes decisions are terminated once accumulated evidence reaches a predefined level or bound. These “accumulate-to-bound” models have close ties to normative theory and can explain a range of behavioral and neural findings. However, they are optimal only under the highly restrictive conditions used in many decision studies, in which the informativeness, rate of acquisition, and other features of the evidence are stable and known in advance. It is not known how the brain balances decision deliberation and commitment more generally, when temporally extended decisions must contend with our dynamic and uncertain world. The goal is to advance our understanding of the decision rules used by the brain under these conditions. The PIs start with a novel theoretical foundation that includes flexible decision bounds that are not predefined but instead can be updated while the decision is being formed to optimize performance even under changing conditions. They use this framework to guide the design and analysis of behavioral studies in humans and combine behavioral and neurophysiological studies in non-human primates, which they use to test their primary hypothesis that the primate brain uses dynamic, adaptive rules to support rational decision-making in changing and uncertain environments.This project is co-funded by the Division of Mathematical Sciences (DMS) within the Mathematical and Physical Sciences Directorate (MPS), Division of Information and Intelligent Systems (IIS) in the Directorate of Computer and Information Science and Engineering (CISE), and Division of Behavioral and Cognitive Sciences (BCS) within the Directorate for Social, Behavioral, and Economic Sciences (SBE).This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.