Trust-driven consensus reaching in human-AI hybrid large-scale group decision-making

Area of Science:

• Artificial Intelligence
• Decision Science
• Human-Computer Interaction

Background:

• Human-AI hybrid group decision-making faces challenges in trust representation, feedback during opinion conflicts, and consensus efficiency.
• Existing methods often lack robust mechanisms for dynamic trust and opinion evolution in collaborative settings.

Purpose of the Study:

• To propose a trust-driven consensus-reaching method for human-AI collaborative decision-making.
• To enhance trust representation, feedback in opinion conflicts, and consensus convergence efficiency.

Main Methods:

• A unified collaborative framework integrating human experts and large language models, centered on trust modeling.
• Dynamic trust relationship construction and coupled evolution of trust mechanisms and opinion dynamics.
• Differentiated opinion updating based on trust propagation, consensus measurement, and feedback regulation for an iterative consensus process.

Main Results:

• The proposed method was validated using the Zhengzhou "7.20" rainstorm case.
• Preliminary results show potential advantages over traditional weight adjustment and deep reinforcement learning methods.
• The method demonstrated a trend towards higher consensus quality with lower intervention costs.

Conclusions:

• The trust-driven method effectively addresses key limitations in human-AI collaborative decision-making.
• It offers a promising approach for achieving stable consensus with improved efficiency and quality.
• The framework supports dynamic trust and opinion evolution in complex collaborative environments.