Human-AI coordination for large-scale group decision making with heterogeneous feedback strategies
Jing Zhang, Ning Wang, Ming Tang
Abstract
In group decision making, human experts are usually susceptible to cognitive biases and information overload. Artificial intelligence (AI) has capabilities in data processing and analysis, but is limited by issues such as interpretability and human adoption. Humans and AI have different problem-solving capabilities, they can benefit from each other. Thus, there is a need to leverage a mechanism to tap into the intelligence of both parties and achieve mutually shared outcomes. In this study, we propose a large-scale group decision-making model with human-AI consensus. First, an improved density-peak clustering algorithm is utilized to classify experts into subgroups based on the Similarity-Trust-Attitude score. Then, weights of experts and subgroups are obtained based on the internal influence of experts and the intuitionistic fuzzy entropy of subgroup preferences. Further, considering three different strategies of human-AI interaction, subgroup consensus and subgroup-AI consensus are calculated. Finally, a minimum cost consensus model with heterogeneous feedback strategies is proposed. The usability of the proposed model is verified through a medical diagnosis case. This study found that the human-AI coordination with heterogeneous feedback strategies can reduce adjustment costs, and different interaction mechanisms have different effects.