Deterministic response strategies in a trial-and-error learning task.

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Behavioral Economics

Background:

• Trial-and-error learning is fundamental for adapting to new environments.
• Simple stimulus-response associations are often insufficient for optimal learning when dependencies exist.
• Previous research indicates humans employ higher-level cognition for efficient learning in complex settings.

Purpose of the Study:

• To analyze human learning strategies during the initial phase of a trial-and-error task with hidden dependencies.
• To compare the explanatory power of standard Q-learning with novel deterministic response models.
• To investigate the transition of learning strategies from initial exploration to asymptotic performance.

Main Methods:

• Computational modeling of human behavior in a trial-and-error learning task.
• Analysis of data from 85 human subjects with deterministic feedback and hidden stimulus-response dependencies.
• Comparison of model fits, including standard Q-learning and deterministic response models.

Main Results:

• Standard Q-learning inadequately explains initial learning strategies.
• Deterministic response models best explain the behavior of 50.6% of subjects during initial learning.
• A significant portion of subjects exhibited generic optimal learning (21.2%) or partial dependency exploitation (22.3%).
• Q-learning effectively explains asymptotic performance after the initial learning phase.

Conclusions:

• Human learning in trial-and-error tasks transcends simple reinforcement learning.
• High-level cognitive processes support sophisticated, efficient learning strategies during initial phases.
• These cognitive strategies are gradually replaced by stimulus-response associations as learning progresses.