SBE-UKRI:A Novel Theory of Ordered Judgment Processes

决策有三个核心要素：判断、偏好和选择（Fischhoff & Broomell, 2020）。判断代表了人们如何理解与选择相关的结果及其发生的概率。该提案关注的是人们如何通过汇总来自不同信息源的多个证据来形成判断。许多关于判断的文献都依赖线性模型作为理解这类判断的主要理论基础（例如，Broomell & Budescu, 2009; Dawes, 1971; Karelaia & Hogarth, 2008）。我们提出了一种新的有序判断理论，它融合了心理学理论和计算机科学在数据聚合和人工智能领域的操作和理论进展。我们建议通过(1)利用具有统计上理想的信息聚合属性的计算简单排序过程，(2)模仿判断中感知的相对性质，以及(3)将这种方法与先前使用的线性理论无缝集成，来推进心理学理论超越线性模型。研究人员认为排序是认知的一个潜在组成部分，但迄今为止，缺乏对其存在进行实证测试的能力。我们的有序判断理论将提供第一个操作框架，用于实证检验排序在判断及其以外的作用。我们将通过三个研究目标来发展和检验这一新颖的理论。第一个目标是根据初步工作开发判断的预测模型（Broomell & Wagner, 2023）。这样的模型将促进有针对性的实验，以检测有序的判断过程是否可以解释人类的行为。第二个目标是使用实验室实验来了解基于顺序的过程与判断过程自然匹配的程度，并可以预测人类行为。这些研究将揭示系统的和可预测的行为，判断如何对环境中的瞬间变化作出反应。第三个目标是开发从观察判断中估计有序判断模型自由参数的方法。这种估计程序的发展将允许将判断更详细地分解为驱动判断的稳定和动态优先级。此外，这样的估计过程将对心理学和计算机科学工作中的模型拟合产生广泛的影响。更广泛的影响我们预计，这种理论整合将对心理学和计算机科学研究产生远远超出智力价值的影响。对于心理学而言，我们的有序判断理论对如何显示数据以促进准确处理有许多启示，这将有助于决策支持和从图形用户界面到操作机器等环境中的人为因素工作。对于计算机科学，我们预计这项工作将通过阐明普遍的线性和非线性聚合过程与人类推理之间的直接联系，为可解释人工智能的关键领域做出贡献。这可以提供机制来理解、评估和验证关键应用（如安全和国防、能源和医疗保健）中机器学习驱动的决策方法。此外，正如Kreinovich（2022）所暗示的那样，所提出理论的算法实现有可能提供在机器学习中聚合信息的有效手段，包括神经网络。本提案中的工作还将用于培养博士生和博士后研究人员，利用数学，计算和实证方法进行跨学科和国际合作研究。这项工作的结果将补充PI和联合PI在美国和英国的本科和研究生阶段的教学和指导。

OverviewThere are three core elements to decision making: judgment, preference, and choice (Fischhoff & Broomell, 2020). Judgments represent how people come to understand the outcomes associated with choices along with their probabilities of occurrence. This proposal focuses on how people form judgments by aggregating multiple pieces of evidence gathered from different information sources. Much of the literature on judgment has relied on linear models as the prominent theoretical basis for understanding such judgments (e.g., Broomell & Budescu, 2009; Dawes, 1971; Karelaia & Hogarth, 2008). We propose a new theory of ordered judgment that fuses psychological theory with operational and theoretical advances from computer science in the areas of data aggregation and artificial intelligence.Intellectual MeritWe propose to advance psychological theory beyond linear models by (1) leveraging computationally simple ordering processes with statistically desirable properties for information aggregation, (2) mimicking the relative nature of perception in judgment, and (3) seamlessly integrating this approach with linear theories previously used. Researchers have implicated ordering as a potential component of cognition but to date, lack the ability to empirically test for its presence. Our theory of ordered judgment will provide the first operational framework for empirical tests of the role of ordering in judgment and beyond. We will develop and test this novel theory through three research objectives. The first objective is to develop a predictive model of judgment based on preliminary work (Broomell & Wagner, 2023). Such a model will facilitate targeted experimentation to detect whether ordered judgment processes can account for human behavior. The second objective is to use lab experiments to understand the degree to which order-based processes naturally fit with judgment processes and can predict human behavior. These studies will reveal systematic and predictable behavior in how judgments react to momentary changes in context. The third objective is to develop methods for estimating the free parameters of the ordered judgment model from observed judgments. The development of such an estimation procedure will allow for a more detailed decomposition of judgments into stable and dynamic priorities that drive judgment. Additionally, such an estimation procedure would have implications for model fitting broadly in psychological and computer science work.Broader ImpactsWe anticipate that this theory integration will have impacts for both psychological and computer science research that go well beyond the intellectual merits. For psychology, our theory of ordered judgment has many implications for how to display data to facilitate accurate processing that will be useful for decision-support and human factors work in contexts ranging from graphical user interfaces to operating machinery. For computer science, we anticipate that this work will contribute to the crucial area of explainable artificial intelligence by articulating direct links between pervasive linear and non-linear aggregation processes and human reasoning. This can afford mechanisms to understand, evaluate and validate machine-learning driven decision-making approaches in critical applications such as security and defense, energy, and healthcare. Further, algorithmic implementations of the proposed theory hold the potential to offer efficient means of aggregating information in machine learning including neural networks, as alluded to in Kreinovich (2022). The work in this proposal will also serve to train doctoral students and postdoctoral reasearchers in interdisciplinary and internationally collaborative research leveraging mathematical, computational, and empirical methods. The results of this work will complement the PI and co-PI's teaching and instrunction at undergraduate and graduate levels in the US and the UK.