| 群智反馈的人机协同决策多属性价值对齐方法 |
| Multi-attribute value alignment for human-machine collaborative decision-making: collective intelligence feedback |
| 摘要点击 3 全文点击 0 投稿时间：2025-01-05 修订日期：2026-04-03 |
| 查看/发表评论 下载PDF阅读器 |
| 中文关键词 群体决策; 人机协同; 决策属性权重; 大语言模型; 反馈传播算法 |
| 英文关键词 group decision-making; human-machine collaboration; decision attribute weights; large language model; feedback propagation algorithm |
| 基金项目 国家自然科学基金重大项目(72293574; 72091515); 国家自然科学基金面上项目(72301298); 湖南省自然科学基金项目(2024JJ6723); 湖南省研究生科研创新项目(CX20230141). |
| 投稿方向 群体决策、人机协同决策 |
| 作者 | 单位 | 邮编 | | 郭栋炜 | 中南大学 | 410083 | | 徐选华* | 中南大学 | 410083 | | 孙妍妍 | 中南大学 | | | 王宗润 | 中南大学 | | | 赵程伟 | 中南林业科技大学 | |
|
| 中文摘要 |
| 针对当前机器智能辅助决策对人机分歧考虑的不足, 提出了一种基于人类群体智慧反馈优化的人机协同多属性价值对齐方法(MuAVA). 首先, 设计了由数据计算和模型推理两个模块构成的属性权重挖掘方法, 建立了辅助决策的机器智能模型. 其次, 引入人类共识达成过程, 基于群体智慧实现人类属性熵权的计算. 然后, 测度人机一致性水平, 建立专家群智的反馈传播模型, 实现人类群体智慧向机器智能的传递. 结果表明, MuAVA方法可以融合人机各自优势为决策提供建议策略, 同时有效避免了人机协同带来的不稳定性风险, 能够为重大事件决策的智能化提供方法支持. |
| 英文摘要 |
| To address the current inadequacies in aligning human-machine opinions in decision-making aided by machine intelligence, a human collective intelligence feedback optimization method for Multi-Attribute Value Alignment (MuAVA) is proposed. Initially, a machine intelligence model is developed for decision support, designing a method to extract decision attribute weights through two modules: data computation and model reasoning. Then, the human consensus reaching process is introduced, employing collective intelligence to calculate human attribute entropy weights. Next, the human-machine consistency level is measured, and a feedback propagation model of expert collective intelligence is established to facilitate the transfer of human Collective Intelligence (CI) to Machine Intelligence (MI). The results show that the MuAVA method combines the respective advantages of humans and machines to provide decision-making strategies, effectively avoiding the instability risks associated with human-machine collaboration and offering methodological support for intelligent decision-making in major events. |
| 关闭 |
|
|
|
|
|
