Coherence-Based Reasoning and Rationality: A Neural Network Modelling Approach to Decision Making

生态合理性的一个关键方面是在复杂环境中快速准确地学习。因此，现实世界的决策者经常面临反馈稀少、嘈杂和动态的环境。这些因素可能会使线索-标准关系的学习复杂化。相比之下，其他因素，如最佳学习顺序，可能会促进学习。在阐述并行约束满足决策模型(PCS)的基础上，我们将开发和研究PCS的学习机制，并将其与其他模型(即用于快速而经济的启发式策略选择的强化学习理论、前馈网络、回归和贝叶斯模型)进行比较。首先，为了评估生态合理性的相对程度，将在一系列蒙特卡洛模拟中比较这些模型在这些真实世界条件下的性能。其次，这些模型预测行为的能力将在随后的一系列实证研究中得到检验。我们假设，所建议的PCS学习机制应该允许学习，即使在动态环境中也可以学习，因为a)内部反馈机制，b)补偿性线索整合，c)对所有信息进行监控。

A key aspect of ecological rationality is fast and accurate learning in complex environments. Real-world decision makers are thereby often confronted with environments that are scarce in feedback, noisy, and dynamic. These factors are likely to complicate learning of cue-criterion relations. Other factors such as an optimal learning order could, in contrast, facilitate learning. Elaborating on previous work on the Parallel Constraint Satisfaction model of decision making (PCS), we will develop and investigate learning mechanisms for PCS and compare them to alternative models (i.e., reinforcement learning theory of strategy selection for fast and frugal heuristics, feedforward networks, regression and Bayes models). First, to assess the relative extent of ecological rationality, the models’ performance under these real-world conditions will be compared in a series of Monte-Carlo simulations. Second, the models’ ability to predict behavior will be tested in a series of subsequent empirical studies. We hypothesize that the suggested learning mechanism for PCS should allow learning even in dynamic environments with scarce and noisy feedback due to a) mechanisms of internal feedback, b) compensatory cue integration, and c) monitoring of all pieces of information.