Exploring the bounded rationality in human decision anomalies through an assemblable computational framework

Abstract

ABSTRACTSome seemingly irrational decision behaviors (anomalies), once seen as flaws in human cognition, have recently received explanations from a rational perspective. The basic idea is that the brain has limited cognitive resources to process the quantities (e.g., value, probability, time, etc.) required for decision making, with specific biases arising as byproducts of the resource allocation that is optimized for the environment. This idea, as a variant of bounded rationality, has grown into a fast-evolving subfield. However, the following issues may limit its development: the assumptions of different models lack consistency, each model typically focuses on one single environmental factor, and the covered decision anomalies are still limited. To address these issues, here we develop a computational framework—the Assemblable Resource-Rational Modules Framework (ARRM)—that integrates ideas from different lines of boundedly-rational decision models as freely assembled modules. The framework can accommodate the joint functioning of multiple environmental factors, and allow new models to be built and tested along with the existing ones, potentially opening a wider range of decision phenomena to bounded rationality modeling. We further apply ARRM to modeling an anomaly in decision under risk (the “peanuts effect”) that we proved to be challenging for all previous decision theories. For one new and three published datasets that cover two different task paradigms and both the gain and loss domains, our boundedly-rational models reproduce two characteristic features of the peanuts effect and outperform previous models in fitting human decision behaviors.