摘要
生成式人工智能已经成为人工智能的最前沿技术之一,被认为是通往通用人工智能的重要阶梯。对2023年生成式人工智能的主要发展动向进行了综合评述。首先盘点生成式人工智能技术当前发展状态;其次概述生成式人工智能技术未来发展趋势;最后探讨生成式人工智能的军事应用潜力。综述表明,多模态性和可解释性是当前生成式人工智能技术的热门方向,生成式人工智能技术的应用使强调指令理解推理的机器人、强调归纳推理的基础科学等领域不断取得突破,面向终端用户的“轻量化”和面向开发者的“开源化”势头正盛,与此同时,对生成式人工智能技术应用的监管也迫在眉睫。综述认为,深度融合多模态、建设高质量数据资源及评估手段、强调与环境自主交互、在边缘侧低成本应用部署、平衡技术发展与规制治理等,将是生成式人工智能技术在未来的重要发展方向。
Generative Artificial Intelligence(GAI)has become one of the cutting-edge technologies in Artificial Intelligence(AI),and is considered as an important step towards Artificial General Intelligence(AGI).This paper makes a comprehensive survey of the main development trends in the field of GAI.Firstly,the development trends of GAI are reviewed;Secondly,its further development directions are summarized;Finally,the potential military applications of GAI are explored.This survey shows that multimodality and interpretability are currently the key directions of GAI;GAI is now accelerating the breakthroughs in domains of robots that emphasize instruction comprehension and reasoning,as well as basic science that emphasizes inductive reasoning;lightweight GAI for end-users and open-source GAI for developers have become popular trends;and the supervision of the application of GAI is also urgent.This survey believes that deep integration of multimodality,construction of high-quality data resources and evaluation methods,autonomous interaction with the environment,low-cost application deployment at the edge,and balancing technological development with regulatory governance,will be important directions for the future development of GAI.
作者
王亚珅
方勇
江昊
曾园园
白然
WANG Yashen;FANG Yong;JIANG Hao;ZENG Yuanyuan;BAI Ran(Key Laboratory of Cognition and Intelligence Technology(CIT),Artificial Intelligence Institute of CETC,Beijing 100049,China;Center for Information Research,Academy of Military Sciences,Beijing 100142,China;School of Electronic Information,Wuhan University,Wuhan 430072,China;National Engineering Laboratory for Risk Perception and Prevention(RPP),China Academy of Electronics and Information Technology,Beijing 100041,China)
出处
《无人系统技术》
2024年第2期101-112,共12页
Unmanned Systems Technology
基金
国家自然科学基金(62106243,U19B2026)。
关键词
生成式人工智能
多模态
可解释性
具身智能
轻量化
开源
监管
Generative Artificial Intelligence
Multimodality
Interpretability
Embodied Intelligence
Lightweight
Open-source
Supervision