Theory of Adaptive Decision Making

本研究的目的是调查个人如何根据过去决策的后果所产生的经验来学习提高他们的决策技能。该研究旨在对数学学习理论进行实证评估，该理论可以总结为三个原则：第一原则描述了如何对每种情况进行分类，以确定一套适当的决策规则；第二个原则描述了如何根据过去的平均表现从一组相互竞争的规则中选择一条特定的规则；第三个原则描述了每个决策规则的性能是如何通过由先前结果的反馈指导的微调过程来缓慢改进的。前两个实验将测试一个假设，即使用一个简单的“爬坡”机制来微调和提高给定规则的性能。实验1将研究在医学诊断任务中如何学习诊断标准，实验2将研究个人如何学习改进资源分配以实现利润最大化。接下来的两个实验将检验规则是根据最近加权平均机制产生的性能评估来选择的假设。实验3将研究个体如何学习在不同疗效的虚构医疗方案中进行选择，实验4将研究个体如何学习在虚构患者分类的诊断规则中进行选择。最后两个实验将检验在规则选择之前对决策情景进行分类的假设，并且对每个类别分别评估规则性能。实验5将检验未经训练的决策者依赖于具体的表面特征，而经过训练的决策者依赖于抽象的结构原则对决策情景进行分类的假设。实验6将研究个体如何在环境线索的基础上学习区分规则执行。总的来说，这个研究项目将提高我们对决策策略的演变，最优和非最优策略的发展，以及新手决策者通过实践成为专家的过程的理解。

The purpose of this research is to investigate how individuals learn to improve their decision-making skills on the basis of experience resulting from consequences of past decisions. The research was designed to evaluate empirically a mathematical learning theory that can be summarized by three principles: The first principle describes how each situation is categorized for the purpose of identifying a set of appropriate decision-making rules; the second principle describes how one particular rule is selected from a set of competing rules on the basis of average past performance; and the third principle describes how the performance of each decision rule is slowly improved by a fine tuning process guided by feedback from previous consequences. The first two experiments will test the hypothesis that a simple "hill-climbing" mechanism is used to fine tune and improve the performance of a given rule. Experiment 1 will investigate how diagnostic criteria are learned in a medical diagnosis task, and Experiment 2 will investigate how individuals learn to improve resource allocations to maximize profits. The next two experiments will test the hypothesis that rules are selected on the basis of performance estimates produced by a recency-weighted averaging mechanism. Experiment 3 will investigate how individuals learn to choose among fictitious medical treatments that differ according to efficacy, and Experiment 4 will investigate how individuals learn to choose among diagnostic rules for categorizing fictitious medical patients. The last two experiments will test the hypothesis that decision situations are categorized prior to rule selection, and rule performance is assessed separately for each category. Experiment 5 will test the hypothesis that untrained decision makers rely on concrete surface features, whereas trained desision makers rely on abstract structural principles for categorizing decision situations. Experiment 6 will investigate how individuals learn to discriminate rule performance on the basis of envirionmental cues. Overall, this research program will improve our understanding about the evolution of decision strategies, the development of both optimal and non-optimal strategies, and the process that novice decision makers use to become experts through practice.