Transition dynamics shape mental state concepts

Area of Science:

• Cognitive Science
• Social Psychology
• Computational Neuroscience

Background:

• Humans possess a sophisticated ability to understand others' mental states, including thoughts and feelings.
• This mental state knowledge is organized conceptually, often along dimensions like valence, guiding social interactions.
• The acquisition process of this conceptual structure remains incompletely understood.

Purpose of the Study:

• To investigate the role of observing mental state dynamics in shaping conceptual structure.
• To test whether transition probabilities between mental states causally influence conceptual judgments.
• To explore how artificial intelligence models learn conceptual dimensions from mental state dynamics.

Main Methods:

• Nine behavioral experiments with 1,439 participants examined the impact of observed mental state transitions on conceptual judgments.
• Computational modeling was used to analyze how individuals represent mental state dynamics geometrically.
• Three neural network experiments trained AI to predict human mental state dynamics.

Main Results:

• Observing frequent transitions between mental states led participants to judge those states as conceptually similar.
• Computational models revealed that mental states are embedded in a geometric space where proximity reflects transition likelihood.
• Neural networks trained on mental state dynamics spontaneously acquired human-like conceptual dimensions.

Conclusions:

• Mental state dynamics, and the drive to predict them, are fundamental in shaping the conceptual structure of mental states.
• The findings suggest a mechanism by which experience with dynamic social information informs abstract conceptual understanding.
• This research bridges cognitive science, social psychology, and AI by demonstrating shared principles in conceptual learning.