Applying Ethical Principles to Decision Support

这个提议属于AI伦理中公平和多智能体系统（MAS）的一般范围。我们通过社会技术系统（STS）的概念来看待MAS，Murukannaiah和Singh（2020）解释说，它由多个人类-智能体二人组组成，其中一个社会层对技术层施加了监管。为了促进STS中公平的最终目标，Chopra和Singh（2018）认为，重要的是通过宏观伦理的透镜来看待STS，目标是STS的整体治理，而不是单个代理人的行为。在STS的治理中，价值观和规范的作用对于取得令人满意的结果至关重要。然而，当利益相关者对必须做出的决定有不同的偏好时，就会出现问题。这可能发生在普通场景中，例如物联网智能家居代理决定提前或推迟供暖，考虑相关利益相关者的不同偏好。此外，正如Dignum和Dignum（2020）所强调的那样，也可能存在价值观或规范冲突的情况。在促进公平治理的STS中，决策必须考虑到利益相关者的各种偏好以及现有的价值观和规范。这些决定还必须解决任何现有的冲突。因此，在支持以令人满意的方式解决这些日常困境的决定方面存在差距。我们建议通过应用伦理原则来解决这一差距，Saltz et al.（2019），Leben（2020）和Conitzer et al.（2017）的工作证实了这一点。为了通过应用道德原则来支持决策，以促进公平的更高目标，有各种问题需要解决。这些问题包括人工智能伦理文献中存在哪些伦理原则，哪些伦理原则需要进一步研究人工智能伦理，如何将伦理原则整合到推理中以管理STS，如何解决冲突的伦理原则，如何将抽象的伦理原则应用于特定的环境，以及我们如何利用这些信息来创建一致的响应。该项目属于EPSRC工程研究领域的福尔斯。

This proposal lies in the general scope of fairness and multiagent systems (MAS) in AI ethics. We view MAS through the concept of sociotechnical systems (STS), which Murukannaiah and Singh (2020) explain consist of multiple human-agent duos with a social tier that imposes regulations upon a technical tier. To promote an end goal of fairness in STS, Chopra and Singh (2018) argue that it is important to view STS through the lens of macro ethics, targeting the holistic governance of STS rather than the actions of single agents. In the governance of STS, the roles of values and norms are crucial in achieving satisfactory outcomes. However, issues arise when stakeholders have different preferences about a decision that must be made. This can occur in mundane scenarios, for instance whether an Internet of Things smart home agent decides to put the heating on earlier or later, considering the differing preferences of relevant stakeholders. In addition, as highlighted by Dignum and Dignum (2020), there may also be scenarios where values or norms conflict. In an STS that promotes fair governance, decisions must be made that consider the variety of stakeholder preferences as well as existing values and norms. These decisions must also resolve any existing conflicts. There is thus a gap in supporting decisions to resolve these everyday dilemmas in satisfactory ways. We suggest that this gap can be addressed through the application of ethical principles, which is corroborated by the works of Saltz et al. (2019), Leben (2020), and Conitzer et al. (2017). To support decision making through the application of ethical principles in order to promote the higher goal of fairness, there are various questions that should be addressed. These include which ethical principles exist in AI ethics literature, which ethical principles require further research in relation to AI ethics, how ethical principles can be integrated into reasoning to govern STS, how conflicting ethical principles can be resolved, how abstract ethical principles can be applied to specific contexts, and how we can utilise this information to create consistent responses. This project falls within the EPSRC Engineering research area.