Skip to content

GCRI · 创新与技术变革

负责任的技术

负责任的技术安全、能很好地服务于人、让人对重大决策保持控制、易于使用并尊重权利。全球风险与创新中心(Global Centre for Risk and Innovation,GCRI)在自身工作中(包括 Foundry)应用这些原则,并研究各机构如何将其应用于所采用的技术。原则、其应用及局限见下文。属于 创新与技术变革.

02Five principles

What responsible technology asks for.

安全

先测试,后信任

系统 在接触真实服务之前,会在受控条件下按停止规则进行评估。

监督

由人来决定

具有重大影响的决策由负责任的人员作出,分析绝不被当作决策呈现。

权利

隐私与同意

个人数据和受保护的知识仅在具有合法依据、明确目的且在需要时取得同意的情况下使用。

  • 社会成果:谁受益、谁承担风险、谁被排除在外,都是每项评估的组成部分
  • Accessibility: interfaces and outputs are designed so that people with disabilities can use them
03工具

赋予各项原则具体内容的文书。

原则其所依据的工具What it requires in practice
Human rights《联合国工商业与人权指导原则》(2011年);欧洲委员会人工智能公约(2024年)Human rights due diligence before and during deployment
人工监督欧盟《人工智能法案》关于高风险 系统人工监督的条款;联合国教科文组织建议书(2021 年)A named person able to understand, override and stop the system
风险管理NIST AI RMF 1.0(2023)及其生成式人工智能概要(2024);ISO/IEC 42001:2023Documented mapping, measurement and management of AI risks
无障碍WCAG 2.2 (W3C, 2023)残障人士可以使用的界面,并与他们一起测试
生物安全与生物安保世卫组织指导框架(2022 年);《生物武器公约》Risk assessment of dual-use research before it starts
04GCRI’s part

在项目内部应用这些原则,并报告其不足之处。

GCRI applies these instruments to its own systems, including Foundry, and reports gaps it finds in its own practice. Lessons are designed for the Exponential Technology series; AI components used in the programme would carry model cards and dataset sheets under NS-12 of the proposed Nexus Standards. Anyone can report a concern about how a GCRI system treats people, without an account.

05In GCRI’s own work

这些原则在计划内部如何适用。

  1. 01

    证据标签

    Every result says whether it is observed, modelled, a scenario or expert judgement, so automated outputs cannot pass as observation.

  2. 02

    受控演练

    模拟预警绝不会送达真实受众,真实基础设施也绝不会成为未经批准的目标。

  3. 03

    Release is not deployment

    经过测试的配置并不等于获准投入运行;该决定属于运营方或主管机构。

  4. 04

    Correction and challenge

    任何人都可以对公开结论提出质疑,具体规定见 科学与编辑诚信.

06局限

Principles are not a certificate.

GCRI does not certify that a technology is responsible, safe or compliant with any law or standard. It publishes evaluations with their scope and limits. The Global Risks Forum (GRF) examines how institutions govern AI and digital systems under digital and AI governance. GCRI’s own data handling is under data and privacy responsibilities. See also technology assessment.

08问题

问题与解答。

GCRI 是否遵循特定的人工智能框架?
GCRI 在自身工作中应用本页所述原则,并将公认框架作为指名参考引用。在没有评估记录的情况下,它不声称符合任何框架。
Can GCRI review my organisation’s AI system?
Organisations can propose a scoped evaluation through the technology provider or operator routes. Any evaluation has a documented plan and is never presented as certification.
人工监督在实践中意味着什么?
That a named, accountable person makes consequential decisions, can understand the basis for a system’s output and can stop or override it.
GCRI 网站如何处理无障碍问题?
GCRI sites are designed to meet WCAG 2.2 level AA. Pages are tested, barriers found are fixed and anyone can report a barrier.
报告贡献

从本页贡献证据。

Reports 是联合体的季刊。您对该主题的了解可以成为下一期中的证据、案例或更正,署名归您,并在属于您本人作品时以 CC BY 4.0 发布。

GCRI
全球风险与创新中心(The Global Centre for Risk and Innovation)

带来一种方法、一个数据集或一个问题。

信息和咨询现已开放。每项咨询都会获得一个编号,并告知下一步;申请、付款、算力分配和注册只有在相应团队、条款和审查能力就绪时才会开放。