Meta-learning ecological priors from large language models explains human learning and decision making

The paper introduces Ecologically Rational Meta-learned Inference (ERMI), a new class of learning algorithms that utilizes large language models to generate ecologically valid cognitive tasks and employs meta-learning to optimize rational models for these environments. ERMI demonstrates superior performance in capturing human behavior across 15 experiments related to function learning, category learning, and decision-making, outperforming traditional cognitive models in trial-by-trial predictions. This framework highlights the potential for large language models to inform and enhance our understanding of human cognition by aligning learning processes with the statistical structures of real-world tasks.