Trustworthy and Ethical Assurance of Digital Twins (TEA-DT)

近年来，在定义“负责任”的人工智能研究和创新方面付出了相当大的努力。虽然取得了切实的进展，但许多部门仍然缺乏操作和实施道德原则的工具和能力。此外，许多项目团队还发现，知道如何实现目标（如公平性或可解释性）并将它们已实现的情况传达给受影响用户的其他涉众是一项挑战。如果被忽视，这些差距可能会阻碍建立公众对人工智能技术信任的努力，或放大由偏见和不透明的社会技术系统造成的现有社会危害和不平等。数字孪生的可信赖和道德保证（TEA- dt）项目将开发一个现有的开源平台，称为可信赖和道德保证（TEA）平台，该平台旨在帮助用户应对上述挑战。TEA平台帮助用户和项目团队定义、操作和实施道德原则作为要确保的目标，并且还提供了沟通这些目标如何实现的方法。它通过指导个人和项目团队识别相关的声明和证据来证明他们所选择的道德原则，使用可嵌入整个项目生命周期的参与性方法来实现这一点。平台的输出——用户生成的保证案例——可以由不同的利益相关者共同设计和审查，通过开放、清晰和可访问的沟通培养信任。TEA平台由三个主要元素组成：1)一个在线工具，用于形成关于道德目标的合理论证；2)用户友好的指导，以培养团队和组织之间的批判性思维；3)一个支持性的社区基础设施，用于分享和讨论最佳实践。虽然该平台是为广泛的应用而设计的，但TEA-DT项目将特别关注数字双胞胎，即与物理副本密切相关的虚拟副本，以便能够访问数据和见解，从而改进和优化其现实版本的操作方式。更具体地说，项目团队将在健康、自然环境和基础设施三种不同背景下，对数字孪生的保障进行范围界定研究。尽管数字孪生体有望在这些领域带来巨大的社会效益，但它们越来越依赖于各种形式的人工智能，并经常在安全关键环境中运行，这意味着必须解决一些挑战，以确保它们的道德和值得信赖的发展。例如，在卫生领域，出现了有关数据隐私和所有权的问题；环境应用必须解决偏见和公平问题，这一问题因全球规模和不同法律而变得复杂；在基础设施方面，关于不确定性通信的技术挑战引起了对透明度和可解释性的额外需求。TEA-DT项目将与主要合作伙伴和利益相关者合作，开展范围界定研究，共同开发示范保障案例，并增强平台的功能，使其更加用户友好，并集成到工作流程中。通过致力于开放研究和社区建设原则，该项目旨在a)系统地分享最佳实践和标准，b)使道德原则的运作更容易获得和包容，以及c)将项目可持续地与现有网络和社区相结合。

In recent years, considerable effort has gone into defining "responsible" AI research and innovation. Though progress is tangible, many sectors still lack the tools and capabilities for operationalising and implementing ethical principles. Furthermore, many project teams also find it challenging to know how to achieve goals, such as fairness or explainability, and communicate that they have been realised to other stakeholders of affected users. If ignored, these gaps could hamper efforts to build public trust in AI technologies or amplify existing societal harms and inequalities caused by biased and non-transparent sociotechnical systems. The Trustworthy and Ethical Assurance for Digital Twins (TEA-DT) project will develop an existing open-source platform, known as the Trustworthy and Ethical Assurance (TEA) Platform, which has been designed to help users navigate the process of addressing the aforementioned challenges. The TEA platform helps users and project teams define, operationalise, and implement ethical principles as goals to be assured, and also provides means for communicating how these goals have been realised. It achieves this by guiding individuals and project teams to identify the relevant set of claims and evidence that justify their chosen ethical principles, using a participatory approach that can be embedded throughout a project's lifecycle. The output of the platform—a user-generated assurance case—can be co-designed and vetted by various stakeholders, fostering trust through open, clear, and accessible communication. The TEA platform consists of three main elements: 1) an online tool for crafting well-reasoned arguments about ethical goals, 2) user-friendly guidance to foster critical thinking among teams and organisations, and 3) a supportive community infrastructure for sharing and discussing best practices. Although the platform is designed for a wide range of applications, the TEA-DT project will specifically focus on digital twins—virtual duplicates that are closely coupled to their physical counterpart to enable access to data and insights that can improve and optimise the way their real-world versions operate. More specifically, the project team will carry out scoping research on the assurance of digital twins within three different contexts: health, natural environment, and infrastructure. Although digital twins promise vast societal benefit in these areas, the fact that they increasingly rely on various forms of AI and often operate in safety-critical settings, means that several challenges must be addressed to ensure their ethical and trustworthy development. For instance, in health, questions about data privacy and ownership arise; environmental applications must tackle bias and fairness issues, complicated by global scales and differing laws; and in infrastructure, technical challenges concerning uncertainty communication give rise to additional needs for transparency and explainability. In collaboration with key partners and stakeholders, the TEA-DT project will carry out scoping research to co-develop exemplary assurance cases and enhance the platform's features to make it more user-friendly and integrated into workflows. By committing to open research and community-building principles, the project aims to a) systematically share best practices and standards, b) make the operationalisation of ethical principles more accessible and inclusive, and c) integrate the project sustainably with existing networks and communities.