A novel integration strategy for uncertain knowledge in group decision-making with artificial opinions: A DSFIT-SOA-DEMATEL approach

Abstract

The inherent knowledge limitations possessed by human experts often yield suboptimal accuracy in linguistic decisions within traditional Decision Making Experimentation and Evaluation Laboratory (DEMATEL) methods. Conversely, artificial decision-making approaches, employed to mitigate these limitations, often introduce substantial instability due to technical constraints and ethical biases. This study introduces a novel decision aggregation approach, termed Dempster-Shafer Fusion fitness function (DSFIT)-spatially Optimal Aggregation (SOA), with the primary objective of mitigating uncertainty in DEMATEL assessment decisions through synergizing human expertise and Artificial Intelligence (AI) capabilities. First, artificial decision-maker (ADM) evidence was acquired through talking with an open-source large language model (LLM); Second, DSFIT quantified uncertainty using the equivalent uncertainty and amplifying uncertainty algorithms, thereby assigning appropriate weights; Third, the SOA method combined evidence from all experts into a comprehensive group perspective. Finally, an analysis of human decision-makers' satisfaction factors in interactive multi-objective optimization, utilizing the DEMATEL method, probed the influence of uncertainty operators on ADM weights. Furthermore, we compared ADM knowledge supplementation, ADM acquisition prompt design, and uncertainty assessment methods to enhance the methodology's accuracy during ADM intervention. This work introduces innovative methods to mitigate decision instability during human and LLM collaboration, providing valuable support for strengthening human–machine partnerships.