A new perspective on human judgement and decision making as optimal: A framework for behaviour change

心理学和经济学通常认为人类的判断和决策（JDM）是有缺陷的。JDM通常被描述为非理性的、有偏见的或次优的，有许多观察似乎支持这一观点。例如，当被要求对一次投掷硬币的可能结果做出判断时，在之前的一长串抛硬币都是正面朝上的情况下，许多人认为下一次抛硬币会是反面，而实际上，之前的抛硬币并没有改变50:50的正面：反面的概率。这种误解被称为赌徒谬误（GF）。然而，如果我们承认人类的注意力集中范围很窄，记忆能力有限，比如只能记住最后5次投掷，这将限制我们从长序列中学习到的东西。这个有限的经验窗口意味着，从数学上讲，并不是所有给定长度的序列都有相同的机会出现在更长的（但有限的）序列中，因此赌徒对连续出现正面后出现反面的期望可以被证明是完全合理的。在这个项目中，我们的目标是表明，如果我们考虑到人类认知能力的限制或环境施加的限制，人类JDM并没有缺陷，实际上在这些限制下是最优的。我们将用两个例子来证明这一点，上面的GF和“偏好逆转”（PR）是JDM通常被描述为有偏见的另一类现象。此外，我们将证明，如果我们改变认知或环境约束，人类将适应并在新的约束下保持最佳状态。这将使我们能够测试操纵人类JDM性能的因果决定因素的后果，因此将允许更好的政策制定。为此，我们将进行行为实验，并制定和评估我们的实验任务的计算模型。该项目包括两个主题，在18个月的项目中并行运行。主题1将探讨人类对随机性的感知以及相关的GF现象。人类感知环境中的结构并将其与对应的随机性区分开来的能力，是我们在事件之间建立因果关系的能力的核心。因此，研究对随机性的感知对于我们理解人类如何理解世界至关重要。利用GF和我们的理论分析，我们将进行一系列行为实验来解释GF在约束条件下是最优的——即短期记忆限制作为个人经验窗口的代理。我们将使用计算模型来测试在这种约束下观察到的行为是否是最佳的。进一步的实验将模拟改变关键约束的干预措施，例如，通过改变认知负荷暂时减少经验窗口的大小。我们假设行为会随着模型预测而改变，并保持最佳状态。这将是一项重大发现，也是对拟议框架的有力检验。主题2将探讨语境效应和偏好逆转（PR）。当更多的选择改变了一个人的偏好时，PR就产生了，即使新的选择更糟糕。尽管公共关系被用作人类非理性的证据，但我们的试点工作表明，引入新的选择提供了关于先验的额外信息，从而使公共关系成为一种最佳行为。我们将操纵选择问题的分布，并通过测量对上下文依赖的变化来衡量对不同任务分布的适应。

Psychology and Economics typically characterise human judgement & decision-making (JDM) as flawed. JDM is commonly described as irrational, biased or sub-optimal and there are many observations that appear to support this viewpoint. For example, when asked to make a judgment about the likely outcome of a coin toss after a long sequence of previous tosses coming up heads, many people expect a tail on the next toss, when in reality previous tosses do not change the 50:50 head:tail chance. This misperception is known as the Gamblers Fallacy (GF). If, however, we accept that humans have a narrow focus of attention and a limited memory capacity for say, only the last 5 tosses, this will limit what can be learnt from the long sequence. This limited experiential window means that, mathematically, not all sequences of a given length have the same chance of occurring in a longer (but finite) sequence and as a result the gamblers' expectation of a tail after a run of heads can be shown to be entirely reasonable.In this project we aim to show that if we take into account the limits on human cognitive capacity or the limits imposed by the environment, human JDM is not flawed and is in fact optimal given those constraints. We will demonstrate this for two examples, the GF above and 'Preference Reversals' (PR) another class of phenomena in which JDM is commonly described as biased. Further, we will show that if we change cognitive or environmental constraints, humans will adapt and remain optimal given the new constraints. This will allow us to test the consequences of manipulating the causal determinants of human JDM performance and therefore will allow for better policy development. To do so we will conduct behavioural experiments as well as formulate and evaluate computational models of our experimental tasks.The project comprises two themes running mostly in parallel over the 18 months project. Theme 1 will investigate human Perception of Randomness and the associated GF phenomenon. The ability of humans to perceive structure in the environment and differentiate it from its counterpart, randomness, is at the heart of our ability to make causal links between events. For this reason, studying perception of randomness is vital for our understanding of how humans understand the world. Using the GF and our theoretical analysis, we will run a series of behavioural experiments to explain GF as optimal under constraints - namely the short-term memory limitation as a proxy for an individual's window of experience. We will use computational models to test if the observed behaviour is optimal under such constraints. Further experiments will simulate interventions that modify the key constraints, for instance, temporarily reducing the size of the window of experience by varying cognitive load. We hypothesise that behaviour will change in line with model predictions, and remain optimal. This would be a significant finding and a strong test of the proposed framework. Theme 2 will investigate context effects and Preference Reversal (PR). PR occurs when the addition of further choices changes a person's preference, even when the new choices are worse. Although PR is used as evidence of human irrationality, our pilot work shows that introducing new choices provides additional information about priors, such that PR becomes an optimal behaviour. We will manipulate the distribution of choice problems and measure adaptation to the different task distributions by measuring changes in the reliance on context.