Heuristics from bounded meta-learned inference.

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Decision Science

Background:

• Heuristics are widely accepted models of human decision-making.
• The origin and selection mechanisms of these heuristics remain debated.
• Existing models lack a comprehensive explanation for heuristic discovery and application.

Purpose of the Study:

• To propose a novel computational model, bounded meta-learned inference (BMI), explaining heuristic discovery and selection.
• To elucidate how environmental specificity and computational resource efficiency influence strategy choice.
• To predict when specific heuristics, such as one-reason decision-making and equal weighting, are employed.

Main Methods:

• Development of the bounded meta-learned inference (BMI) computational model.
• Analysis of how BMI discovers and selects heuristics based on environmental information.
• Empirical verification through three paired comparison studies using continuous features.

Main Results:

• The BMI model successfully rediscovers established heuristics like one-reason decision-making and equal weighting in specific environments.
• Precise predictions were made regarding heuristic application based on attribute knowledge (ranking, direction).
• Empirical studies confirmed BMI's predictions and demonstrated its ability to explain human decision-making characteristics beyond alternative theories.

Conclusions:

• Bounded meta-learned inference (BMI) provides a unified framework for understanding heuristic discovery and selection in human decision-making.
• Environmental context and attribute information critically determine the choice of decision strategies.
• The model offers a parsimonious and empirically validated explanation for adaptive heuristic use.