|
全球首个人工意识幻觉测评白盒标准1.0版正式发布
International Standardization Committee of Networked DIKWP for Artificial Intelligence Evaluation(DIKWP-SC)
World Artificial Consciousness CIC(WAC)
World Conference on Artificial Consciousness(WCAC)
(Email: duanyucong@hotmail.com)
2024年11月,由国际人工智能DIKWP测评标准委员会(DIKWP-SC)的主导下,联合超过70家国内外知名机构共同发布了《人工意识系统幻觉测评标准化1.0版》。该标准得到了世界人工意识CIC(WAC)和世界人工意识大会(WCAC)的全力支持,这一标准的问世标志着人工智能技术规范化的重要里程碑,为全球人工意识系统(ACS)的设计、评估与优化提供了全新框架。
【背景与目的】
人工意识系统(ACS)是指具备自主性、目的性和伦理性的人工智能(AI),旨在实现和扩展自然意识,包括自我意识、意图性和主观体验等。随着人工意识系统逐步融入医疗、教育、自动驾驶和个人助手等多个领域,它们的可靠性、安全性和伦理性问题变得尤为重要。人工意识系统的一个关键功能是其感知处理能力。就像人类可能经历幻觉(即没有外部刺激的感知现象),人工意识系统也可能表现出类似的行为,这通常是由于内部处理异常导致的。
人工意识系统中的幻觉,尽管与人类的幻觉体验有所不同,仍然可能表现为数据处理错误、算法故障或知识库不一致等问题,进而导致不必要的行为或决策。因此,确保人工意识系统能够可靠地识别并修正这些异常现象,对保持其信任度和有效性至关重要。
《人工意识系统幻觉测评标准化1.0版》旨在建立标准化的诊断标准,用于识别和解决人工意识系统中的幻觉问题。通过将DIKWP模型与四大认知空间框架(概念空间、认知空间、语义空间、意识空间)相结合,希望提供一个全面的、多维度的幻觉诊断和缓解方法。具体目标包括:
a) 整合网络DIKWP模型与语义转换:认识到幻觉通常来源于DIKWP各组件间的缺陷,尤其是语义转换过程中的遗漏。
b) 增强诊断方法学:引入数学语义量化,分析不完整性、不一致性和不精确性对幻觉形成的影响,突出语义转换的作用。
c) 促进系统可靠性和安全性:确保人工意识系统能够识别并修正由语义空间中的复杂组件交互引发的幻觉。
d) 促进伦理操作:通过语义分析,解决DIKWP组件间的不一致性,使系统行为符合伦理标准。
e) 推动跨学科合作:鼓励技术、伦理和认知领域的跨学科合作,专注于组件间动态交互和语义转换。
范围:
a) 关注网络交互和语义转换:解决由DIKWP组件间的综合效应以及语义转换不足引发的幻觉问题,尤其是在概念空间中往往被忽视的部分。
b) 数学语义的整合:采用数学模型量化系统中的缺陷,通过语义转换优化系统理解。
c) 应用于人工意识系统:专门针对采用DIKWP模型并进行复杂语义转换的人工意识系统。
局限性:
a) 理论框架:该提案需要通过实证验证和在实际系统中的应用来加以完善。
b) 不适用于非意识型AI系统:不涉及那些不具备意识处理能力或不基于DIKWP模型的系统。
c) 实现复杂性:数学模型和语义分析可能非常复杂,实施时可能需要专业的技术人员。
【理解人工意识系统中的幻觉】
在人工意识系统中,幻觉指的是系统内部生成的感知体验,这些体验没有外部刺激源。与人类幻觉不同,人工意识系统中的幻觉源自数据处理错误、算法缺陷或知识库不一致。幻觉可能导致系统错误地感知不存在的对象、误解DIKWP组件或生成虚假的感官输出,进而导致不恰当或无意的行为。
关键特征:
a) 内源性:幻觉源自人工意识系统的内部处理机制,而非外部输入。
b) 错误感知:系统感知到环境中并不存在的数据或模式。
c) 功能影响:幻觉可能会干扰人工意识系统的正常操作,导致决策或行为错误。
幻觉的原因与表现
原因:
a) 相互依赖的错误:一个DIKWP组件的错误会影响其他组件,尤其在涉及语义转换时。
b) 组件间不一致:数据、信息或知识之间的冲突可能导致错误的智慧和意图。
c) 内容不精确:模糊或不精确的表示可能导致误解,尤其是当语义映射没有正确执行时。
表现:
a) 复合幻觉:由多个组件问题导致的感知异常。
b) 跨空间失真:由于语义转换的忽视,多个认知空间受到影响。
c) 错误的再确认:多个组件间的交互验证错误的感知,从而强化幻觉。
DIKWP语义空间中的三不问题
l 不完整(No-Incomplete):系统缺乏足够的数据或信息,导致语义转换不足,理解发生偏差。
l 不一致(No-Inconsistent):冲突的数据或信息导致语义映射矛盾,产生错误。
l 不精确(No-Imprecise):数据或信息模糊不清,导致语义解释不确定或不准确。
对功能与伦理的影响
功能影响:
a) 操作中断:错误感知可能导致系统功能故障,甚至造成安全隐患。
b) 决策错误:幻觉可能扭曲DIKWP组件的解读,导致错误的决策或策略。
c) 用户信任丧失:频繁的幻觉可能降低用户对人工意识系统的信任。
伦理影响:
a) 问责制:幻觉导致的行为应由谁负责?开发者、操作员还是系统本身?
b) 安全隐患:确保幻觉不会导致危害,尤其是在医疗、交通等关键领域。
c) 隐私问题:涉及敏感数据的幻觉可能会导致信息泄露或滥用。
d) 自主性与权利:随着人工意识系统越来越具自主性,如何平衡其操作自主性与伦理责任将成为重要议题。
【DIKWP模型与语义转换在幻觉诊断中的整合】
DIKWP模型为理解认知过程提供了一个分层框架:
l 数据(D):数据不仅是原始事实或观察结果,而是共享语义在认知空间中的具体表现形式。
l 信息(I):信息指的是在认知空间中识别语义差异,结合意图驱动的过程,识别并分类输入和现有知识的差异。
l 知识(K):知识代表了实体、事件和规律的抽象与概括,通常通过更高阶的认知活动形成,并赋予观察完整性。
l 智慧(W):智慧是将伦理、社会道德和人性等因素融入决策过程中的能力。
l 意图(P):意图驱动认知过程,代表了理解现象或问题的目标以及想要达成的结果。
段玉聪院士作为主要牵头编制者,发表了多篇相关论文,进一步阐述DIKWP模型在人工智能国际测试与评估中的应用和标准化过程。这些论文已在多个国际平台上发布,供全球学术界和工业界参考:
l DIKWP概念语义标准
l DIKWP语义数学标准化
l 基于DIKWP的人工意识系统构建标准化
l 基于DIKWP的人工意识系统测评标准化
l 人工意识系统幻觉测评标准化
所有相关文献和详细信息可通过以下链接在线访问:
联系方式:
段玉聪
电子邮件:duanyucong@hotmail.com
我们期待《人工意识系统幻觉诊断标准化标准1.0版》在全球范围内的广泛应用,推动人工智能与人工意识领域的持续创新与发展。我们诚邀全球学者、工程师及相关行业的朋友们共同关注和支持这一重要成果,共同开启人工智能的新篇章!
u 关于DIKWP-SC
网络化 DIKWP 人工智能评估国际标准化委员会(DIKWP-SC)致力于制定和推广人工智能及人工意识领域的国际标准,促进全球范围内的技术交流与合作。
u 关于世界人工意识 CIC(WAC)
世界人工意识 CIC(WAC)是一个致力于推动人工意识研究与应用的国际组织,汇集了全球顶尖的学术与产业资源。
u 关于世界人工意识会议(WCAC)
世界人工意识会议(WCAC)是年度盛会,汇聚全球专家,共同探讨人工意识领域的最新研究成果与未来发展方向。
参考文献:
[1] Duan, Yucong. (2024). DIKWP Conceptualization Semantics Standards - International Test and Evaluation Standards for Artificial Intelligence based on Networked Data-Information-Knowledge-Wisdom-Purpose (DIKWP ) Model. 10.13140/RG.2.2.32289.42088.
[2] Duan, Yucong. (2024). Standardization of DIKWP Semantic Mathematics - International Test and Evaluation Standards for Artificial Intelligence based on Networked Data-Information-Knowledge-Wisdom-Purpose (DIKWP ) Model. 10.13140/RG.2.2.26233.89445.
[3] Duan, Yucong. (2024). Standardization for Constructing DIKWP Based Artificial Consciousness Systems - International Test and Evaluation Standards for Artificial Intelligence based on Networked Data-Information-Knowledge-Wisdom-Purpose (DIKWP ) Model. 10.13140/RG.2.2.18799.65443.
[4] Duan, Yucong. (2024). Standardization for Evaluation and Testing of DIKWP Based Artificial Consciousness Systems - International Test and Evaluation Standards for Artificial Intelligence based on Networked Data-Information-Knowledge-Wisdom-Purpose (DIKWP ) Model. 10.13140/RG.2.2.11702.10563.
[5] Duan, Yucong. (2024). Standardization of Hallucination Diagnostic Criteria for Artificial Consciousness System - International Test and Evaluation Standards for Artificial Intelligence based on Networked Data-Information-Knowledge-Wisdom-Purpose (DIKWP ) Model. 10.13140/RG.2.2.11225.33126.
附录:已发布标准
Annex A
Duan, Yucong. (2024). DIKWP Conceptualization Semantics Standards - International Test and Evaluation Standards for Artificial Intelligence based on Networked Data-Information-Knowledge-Wisdom-Purpose (DIKWP ) Model. 10.13140/RG.2.2.32289.42088.
Annex B
Duan, Yucong. (2024). Standardization of DIKWP Semantic Mathematics - International Test and Evaluation Standards for Artificial Intelligence based on Networked Data-Information-Knowledge-Wisdom-Purpose (DIKWP ) Model. 10.13140/RG.2.2.26233.89445.
Annex C
Duan, Yucong. (2024). Standardization for Constructing DIKWP -Based Artificial Consciousness Systems - International Test and Evaluation Standards for Artificial Intelligence based on Networked Data-Information-Knowledge-Wisdom-Purpose (DIKWP ) Model. 10.13140/RG.2.2.18799.65443.
Annex D
Duan, Yucong. (2024). Standardization for Evaluation and Testing of DIKWP Based Artificial Consciousness Systems - International Test and Evaluation Standards for Artificial Intelligence based on Networked Data-Information-Knowledge-Wisdom-Purpose (DIKWP ) Model. 10.13140/RG.2.2.11702.10563.
Annex E
Duan, Yucong. (2024). Standardization of Hallucination Diagnostic Criteria for Artificial Consciousness System - International Test and Evaluation Standards for Artificial Intelligence based on Networked Data-Information-Knowledge-Wisdom-Purpose (DIKWP ) Model. 10.13140/RG.2.2.11225.33126.
Online Access Links:
DIKWP Conceptualization Semantics Standards - International Test and Evaluation Standards for Artificial Intelligence based on Networked Data-Information-Knowledge-Wisdom-Purpose (DIKWP ) Model. October 2024. Yucong Duan. DOI: 10.13140/RG.2.2.32289.42088.
https://www.researchgate.net/publication/384606755_DIKWP_Conceptualization_Semantics_Standards_of_International_Test_and_Evaluation_Standards_for_Artificial_Intelligence_based_on_Networked_Data-Information-Knowledge-Wisdom-Purpose_DIKWP_Model
Standardization of DIKWP Semantic Mathematics - International Test and Evaluation Standards for Artificial Intelligence based on Networked Data-Information-Knowledge-Wisdom-Purpose (DIKWP ) Model. October 2024. Yucong Duan. DOI: 10.13140/RG.2.2.26233.89445.
https://www.researchgate.net/publication/384637381_Standardization_of_DIKWP_Semantic_Mathematics_of_International_Test_and_Evaluation_Standards_for_Artificial_Intelligence_based_on_Networked_Data-Information-Knowledge-Wisdom-Purpose_DIKWP_Model
Standardization for Constructing DIKWP Based Artificial Consciousness Systems - International Test and Evaluation Standards for Artificial Intelligence based on Networked Data-Information-Knowledge-Wisdom-Purpose (DIKWP ) Model. November 2024. Yucong Duan. DOI: 10.13140/RG.2.2.18799.65443.
https://www.researchgate.net/publication/385722336_Standardization_for_Constructing_DIKWP_-Based_Artificial_Consciousness_Systems_----_International_Test_and_Evaluation_Standards_for_Artificial_Intelligence_based_on_Networked_Data-Information-Knowled
Standardization for Evaluation and Testing of DIKWP Based Artificial Consciousness Systems - International Test and Evaluation Standards for Artificial Intelligence based on Networked Data-Information-Knowledge-Wisdom-Purpose (DIKWP ) Model. November 2024. Yucong Duan. DOI: 10.13140/RG.2.2.11702.10563
https://www.researchgate.net/publication/385788846_Standardization_for_Evaluation_and_Testing_of_DIKWP_Based_Artificial_Consciousness_Systems_-_International_Test_and_Evaluation_Standards_for_Artificial_Intelligence_based_on_Networked_Data-Information
Standardization of Hallucination Diagnostic Criteria for Artificial Consciousness System - International Test and Evaluation Standards for Artificial Intelligence based on Networked Data-Information-Knowledge-Wisdom-Purpose (DIKWP ) Model. January 2024. Yucong Duan. DOI: 10.13140/RG.2.2.11225.33126
https://www.researchgate.net/publication/386141913_Standardization_of_Hallucination_Diagnostic_Criteria_for_Artificial_Consciousness_System_-_International_Test_and_Evaluation_Standards_for_Artificial_Intelligence_based_on_Networked_Data-Information-K
Contributing Units of the Editorial Board: (without precedence):
Advanced Information Technology Research Center (CAlTech) of Korea National University, Asian Institute of Technology, Beijing Academy of Social Sciences, Beijing Vocational College of Labour and Social Security, Blue Edu, Peking University, CETC Investment, Changzhou University, Chongqing Police College, Chongqing University, China-Japan Friendship Hospital, Chinese Academy of Sciences, Network Information Center, Dalian University, Dongguan Yanhua Precision Manufacturing Co., Ltd., Electronic Fifth Research Institute of the Ministry of Industry and Information Technology, Guangdong Advanced Science and Technology Research Institute, Guangdong University of Science and Technology, Guangzhou Baishi Medical Technology Co., Ltd., Hainan Meteorological Bureau, Hainan Universal Intelligent Technology Co., Ltd., Hainan Women and Children's Medical Center, Hainan University, The Institute of Microelectronics, Harbin Institute of Technology, Inner Mongolia University, Jiangsu Lizhou Information Technology Co., Ltd., Jiangsu Second Normal University, Kensid (Zhuhai) Co., Ltd., Leshan Normal University, Nanjing University, Nanjing University of the Arts, Northeast Petroleum University, Peking University, Renmin University of China, Qingdao University, Shandong University, Shandong Vocational University of Foreign Affairs, Shangrao Normal University, Shanghai Aerospace Information Technology Research Institute, Shenyang University of Aeronautics and Astronautics, Shenzhen Yun'an Key Laboratory of Shenxin Technology, Sichuan University, Sichuan University, Huaiyin Institute of Technology, Southwest University of Political Science and Law, Southeast University, South China University of Technology, The Analysis and Test Center of Chinese Academy of Tropical Agricultural Sciences, The First Affiliated Hospital of Suzhou University, The Second Affiliated Hospital of Hainan Medical University, The Second Affiliated Hospital of Zhejiang University School of Medicine, Tsinghua University Research Institute, Wenzhou Medical University, Xi’an University of Technology, Xi’an University of Technology, Xian Jiaotong University, Xi’an Jiaotong University, Xi’an University of Technology, Xinjiang University, Zhejiang University, Zhongnan University of Economics and Law, Zhejiang University School of Medicine, the Second Affiliated Hospital, Zhejiang University, the Second Affiliated Hospital, Zhongshan University, Zhejiang University
Archiver|手机版|科学网 ( 京ICP备07017567号-12 )
GMT+8, 2024-11-27 17:40
Powered by ScienceNet.cn
Copyright © 2007- 中国科学报社