Model-Based Reasoning in Humans Becomes Automatic with Training.

Area of Science:

• Cognitive Science
• Neuroscience
• Artificial Intelligence

Background:

• Model-based reinforcement learning (RL) is often contrasted with model-free RL, with model-based RL seen as slow and deliberative.
• Previous research indicated that cognitive load impairs model-based reasoning, supporting its non-automatic nature.

Purpose of the Study:

• To investigate whether task familiarity influences the impact of cognitive load on model-based RL.
• To determine if model-based reasoning can operate in parallel with other cognitive demands under specific conditions.

Main Methods:

• Utilized a cognitive task previously shown to impair model-based reasoning under cognitive load.
• Introduced prior task training for participants in two independent studies.
• Employed various analysis methods to validate findings.

Main Results:

• Cognitive load did not impair model-based reasoning when participants had prior task training.
• Task familiarity enabled model-based reasoning to function in parallel with concurrent cognitive demands.
• Findings were consistent across two studies and multiple analytical approaches.

Conclusions:

• Task familiarity can render model-based reasoning automatic and parallelizable.
• This finding has significant implications for understanding complex behavior learning and execution.
• Suggests potential failure modes in psychiatric disorders related to automaticity of model-based reasoning.