Theory of Mind From Observation in Cognitive Models and Humans

Area of Science:

• Cognitive Science
• Artificial Intelligence
• Computational Psychology

Background:

• Developing artificial intelligence (AI) with theory of mind (ToM) is a significant research challenge.
• Existing ToM models often rely on complex assumptions or are task-specific.
• A cognitive model that learns and predicts behavior like humans is needed.

Purpose of the Study:

• To propose a novel cognitive framework for computational theory of mind (ToM).
• To develop a ToM model using instance-based learning theory (IBLT) from observed agent behavior.
• To evaluate the model's ability to infer and predict behavior across varying decision complexities.

Main Methods:

• Utilized instance-based learning theory (IBLT) to create a cognitive model for ToM.
• Trained the IBL model on goal-directed agent behavior in a gridworld navigation task.
• Tested the model's predictive accuracy on new scenarios and a false belief task (Sally-Anne test).

Main Results:

• The IBL model successfully inferred and predicted agent behaviors in novel gridworlds.
• Predictions aligned with human observer performance across different decision complexity levels.
• The model demonstrated capability in a classic false belief test, validating its ToM inference.

Conclusions:

• The proposed IBL framework offers a robust computational representation of ToM.
• This approach avoids limitations of previous ToM models, providing a more human-like cognitive development.
• The IBL observer model holds potential for enhancing human-machine interactions.