Evidence and Uncertainty in Human Reasoning

虽然有些决定是基于对“直觉”或各种选择的效用的简单直接评估，但许多决定涉及冲突的感觉；冲突的解决过程通常会考虑一些不确定事件的证据。一个典型的例子是一个大学生面临的冲突，他第二天有一个重要的考试，但想要放弃学习，以便花一些时间和朋友在一起。在解决这一冲突的过程中，许多学生考虑到即使现在停止学习，他们也会在考试中失败的不确定事件。与这一活动相关的证据包括考试预计有多难，以及学生整个学期在这门课程中做了多少准备。如果考试只有一点点重要，而且证据让学生觉得不用继续学习就能做得很好，那么学生很可能会决定退学。即使考试非常重要，但对成绩很有把握的学生很可能会退学。对于一些决定，部分证据是由科学模型得出的估计概率组成的。医生在进行补救治疗之前可能会考虑估计的癌症复发概率，或者房主在决定是否加固地基时可能会考虑地震的概率。目前的研究是基于最近发展起来的不确定性下的决策理论。其核心思想是，人们对不确定的事件进行分类(例如，“很确定”或“非常确定”的考试成绩)，这种分类是基于各种证据(有时包括从科学模型估计的概率)。这项研究有两个主要分支。其中一个重点是涉及冲突的决策情况。我们研究决策规则是如何根据目标的价值而变化的(例如，学生可能会在一些重要的考试中使用“相当肯定”的规则，但在非常重要的考试中使用“非常确定”的规则)，以及如何使用证据来对不确定的事件进行分类。我们的初步发现表明，不同抽象类别的不确定性使用不同的证据权重。这部分研究还考察了人们做出决定的理由，以便更好地理解理由与选择下的实际决策规则之间的关系。研究的第二个分支专注于证据判断，与最终决定分开--例如，侦探或科学家在形成或确认假设时可能使用的判断。我们感兴趣的是最初产生一个假设和确认它之间的区别；人们对相互冲突的证据的反应；人们如何使用基本比率和相对概率作为证据；以及当相关证据被无关的东西掩盖时，选择相关证据的过程。总而言之，这项研究的主要目标包括测试和发展一种相当新的决策理论；研究科学的概率估计如何用于决策；以及描绘在制定假设时探索性使用证据与假设确认之间的差异。

Although some decisions are made on the basis of simple directevaluation of "gut" feelings, or the utilities of the variousalternatives, many decisions involve conflicting feelings; and theconflict-resolution process often takes into account evidence aboutsome uncertain events.A typical example would be the conflict faced by a college student whohas an important exam the next day but wants to quit studying tospend some time with friends. In the process of resolving thisconflict, many students consider the uncertain event that they will dowell in the exam even if they stop studying now. Evidence relevant tothis event includes how difficult the exam is expected to be and howmuch preparation the student has done all semester in this course. Ifthe exam is only somewhat important, and the evidence leads the studentto feel pretty sure of doing well without further studying, the studentwill probably decide to quit studying. Even if the exam is highlyimportant, a student who is very sure of doing well will probablyquit studying.For some decisions, part of the evidence consists of estimatedprobabilities derived from scientific models. A physician mayconsider estimated probabilities of recurrence of cancer beforerecommending therapy, or a homeowner may take account of earthquakeprobabilities in deciding whether to reinforce the foundation.The present research is based on a recently developed theory ofdecision-making under uncertainty. The core idea is that peoplecategorize uncertain events (e.g., "pretty sure" or "very sure"of doing well on the exam), and that such categorization is based onvarious kinds of evidence (sometimes including probabilitiesestimated from scientific models).The research has two main branches. One focusses on decisionsituations involving conflict. We study how decision rules change,depending on value of goals (for example, a student may use a"pretty sure" rule for a somewhat important exam, but a "very sure"rule for a highly important one) and how evidence is used tocategorize the uncertain events. Our preliminary findings suggestthat different evidence weightings are used for different abstractcategories of uncertainty. This part of the research also examinespeople's justifications for their decisions, in order to understandbetter how justifications relate to actual decision rules underlyingchoices.The second branch of the research focusses on evidence judgment, inisolation from the ultimate decision--for example, the judgmentsthat a detective or a scientist might use in forming or confirminghypotheses. We are interested in the difference between initiallygenerating a hypothesis and confirming it; in how people react toconflicting evidence; in how people use base rates and relativelikelihoods as evidence; and in the process of selecting relevantevidence when it is obscured by irrelevancies.To summarize, the major goals of the research include testing anddeveloping a rather new theory of decision making; studying howscientific probability estimates can be used in policy decisions;and delineating the differences between exploratory use of evidencein formulating hypotheses versus hypothesis confirmation.