Model based planners reflect on their model-free propensities.

Area of Science:

• Cognitive Neuroscience
• Computational Neuroscience
• Decision Science

Background:

• Dual-reinforcement learning theory posits two systems: model-free (MF) for value caching and model-based (MB) for prospective planning.
• A key question is how MB systems integrate influences from MF systems during planning.

Purpose of the Study:

• To investigate whether MB planners exhibit self-reflection by anticipating MF influences.
• To explore the integration of MF tendencies within MB planning processes.

Main Methods:

• Development of a novel bandit task allowing participants to design their environment.
• Analysis of reward assignments in relation to MF tendencies within the task.

Main Results:

• Participants' reward assignments were consistent with an MB system anticipating its own MF propensities.
• Higher rewards were assigned to bandits associated with stronger MF tendencies.

Conclusions:

• MB planning systems can be sophisticatedly self-reflective, incorporating anticipated MF influences.
• Findings have implications for understanding decision-making in areas like addiction, pre-commitment, and economic choices.