Attention, Memory, and Judgment

人类决策的一个基本组成部分是人们如何产生、评估和测试诊断假设。以临床诊断医师的任务为例。临床医生的任务类似于侦探，因为他或她被迫寻找线索（即数据），这些线索可用于生成和测试对当前症状的可能解释。临床医生假定产生可能的诊断（即疾病假设），并积极寻求信息以评估产生的诊断。临床医生在环境中寻找信息可能不是随机的，而是由他或她目前正在考虑的诊断假设指导的。通过搜索过程新发现的信息或数据用于评估当前正在考虑的诊断并生成新的诊断。新诊断的产生可能反过来导致新的信息搜索线索，在这些线索中可能会产生新的假设。在某些情况下，要么是搜索空间耗尽，要么是临床医生不断地无法产生额外的可信假设，要么是一个特定的假设获得了足够的证据支持，临床医生可以自信地做出诊断。拟议研究的目标有两个方面。其中一个目标是更好地理解控制人类假设生成、概率判断和信息搜索的认知约束。了解这些认知约束将使我们能够开发方法或技术来改进现实世界决策任务中的诊断决策，如医疗诊断或智能分析。第二个目标是开发人类判断的认知模型，该模型可以将一段数据（例如，症状）作为输入，并生成一组诊断假设来解释该数据，提供生成假设的概率估计，并随着新数据的体验迭代地修改生成的假设和概率判断。这种类型的系统对于理解人类在动态任务中的决策以及开发能够超越人类决策者的人工智能系统都具有自然的意义。

A fundamental component of human decision making is how people generate, assess, and test diagnostic hypotheses. Take the task of a clinical diagnostician as an example. The clinician's task is similar to that of a detective in that he or she is compelled to search for clues (i.e., data) that can be used to generate and test possible explanations of the presenting symptoms. The clinician presumably generates likely diagnoses (i.e., disease hypotheses) and actively seeks information to evaluate the generated diagnoses. The clinician's search for information in the environment likely is not random, but is instead guided by those diagnostic hypotheses he or she is presently entertaining. The information or data newly revealed through the search process is used to both evaluate the diagnoses currently under consideration as well as generate new diagnoses. The generation of new diagnoses may, in turn, lead to fresh information search threads where new hypotheses might be brought to mind. At some point, either the search space is exhausted, the clinician continually fails to generate additional plausible hypotheses, or one particular hypothesis gains enough evidential support that the clinician can render a diagnosis with confidence. The goal of the proposed research is two fold. One goal is to better understand the cognitive constraints that govern hypothesis generation, probability judgment, and information search in humans. Understanding these cognitive constraints hopefully will enable us to developed methodologies or technologies to improve diagnostic decision making in real-world decision tasks such as medical diagnosis or intelligence analysis. The second goal is to develop a cognitive model of human judgment that can take as input a piece of data (e.g., a symptom) and generate a set of diagnostic hypotheses to explain that datum, provide probability estimates of the generated hypotheses, and revise both the generated hypotheses and the probability judgments iteratively as new data are experienced. This type of system has natural implications both for understanding human decision making in dynamic tasks and for developing artificial intelligence systems that can outperform human decision makers.