Distributional dual-process model predicts strategic shifts in decision-making under uncertainty

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Behavioral Economics

Background:

• Human decision-making under uncertainty is complex and involves adaptive strategy shifts.
• Traditional reinforcement learning (RL) models often fail to capture these dynamic strategic changes.
• Understanding how individuals navigate uncertainty is crucial for explaining adaptive behavior.

Purpose of the Study:

• To propose an entropy-weighted dual-process model that integrates value-based and frequency-based decision-making strategies.
• To investigate how uncertainty influences strategic choices in human decision-making.
• To compare the performance of the proposed model against traditional RL models.

Main Methods:

• Developed a distribution-based model incorporating parallel evaluation systems using Dirichlet and multivariate Gaussian distributions.
• Simulated model behavior under varying levels of uncertainty and reward variance.
• Conducted empirical tests with human participants to validate model predictions.

Main Results:

• The proposed dual-process model successfully captured participants' strategic shift from value-based to frequency-based learning under heightened uncertainty.
• Increased reward variance led participants to prioritize reward frequency over actual reward value.
• The model demonstrated superior performance compared to traditional RL models in explaining adaptive strategy changes.

Conclusions:

• Heightened uncertainty prompts individuals to employ compensatory evaluation methods, including using reward frequency as a proxy for value.
• The entropy-weighted dual-process model offers a robust framework for studying multi-system decision-making in complex environments.
• This research advances our understanding of adaptive human behavior in uncertain, multivariable contexts.