Active Inference Modeling of Socially Shared Cognition in Virtual Reality.

Area of Science:

• Cognitive Science
• Human-Computer Interaction
• Artificial Intelligence

Background:

• Sharing ambiguous concepts in virtual reality (VR) poses challenges for collaborative tasks.
• Active inference provides a computational framework for understanding goal-directed behavior and learning.
• Modeling joint attention and agreement-seeking is crucial for effective human-robot interaction.

Purpose of the Study:

• To propose and validate a process model for sharing ambiguous category concepts in VR using active inference.
• To investigate the role of gaze synchrony in collaborative learning and agreement formation.
• To enhance active inference models by incorporating social cognitive mechanisms.

Main Methods:

• Developed a dual-layer Bayesian update model within an active inference framework.
• Incorporated disagreement in category judgments as a risk term, weighted by gaze synchrony (measured via Dynamic Time Warping).
• Conducted a VR object classification experiment with ambiguous items, comparing bot and human-human interaction conditions.

Main Results:

• The active inference model demonstrated high prediction accuracy for observed values as learning progressed.
• Introducing gaze synchrony weighting significantly improved prediction accuracy, achieving optimal performance.
• The model successfully captured the collaborative learning process leading to agreement in ambiguous situations.

Conclusions:

• The proposed active inference model effectively captures socially shared cognition in VR environments.
• Gaze synchrony is a key factor in improving collaborative accuracy and agreement in ambiguous concept sharing.
• This framework offers a novel approach for modeling human-robot interaction and social learning.