Creating a Dynamic Archive of Responsible Ecosystems in the Context of Creative AI

在创造性人工智能应用的背景下，确定利益相关者和划定生态系统边界尤其具有挑战性，因为这些应用的用户（包括艺术家、公众、地方当局或国家机构）通常不参与其开发。然而，定义边界和涉众是很重要的，因为如果边界是窄的或宽的，则应用不同的RAI考虑；划定狭窄的边界可能会将相关利益攸关方排除在这些生态系统之外。广泛划定边界可能会使RAI的任何考虑变得过于复杂而无法应用或识别。该项目的一个关键目标是开发一个动态档案的结构，该结构可用于识别创造性人工智能背景下这些生态系统中的利益相关者。一个并行的目标是使用存档数据来突出每个项目中需要解决的RAI考虑因素，并比较这些AI生态系统的边界。这将对如何更好地理解人工智能生态系统本身的责任产生更深入的见解。该项目将通过使用在具体研究项目中已经收集的自下而上的证据，为创造性人工智能背景下绘制RAI生态系统奠定基础。我们将人工智能生态系统解释为相互关联的生态系统，由不同的个体参与者和群体组成，以复杂的方式相互作用，并与人工智能应用程序互动。当这些生态系统密切关注RAI英国研究所确定的责任挑战时，它们就是负责任的人工智能生态系统。该项目将分为两个阶段：1)对使用人工智能为残疾音乐家创作音乐的Digiscore （https://digiscore.github.io/pages/aboutus/Creative）和使用人工智能与宠物互动的CAT Royale （https://www.blasttheory.co.uk/projects/cat-royale）进行严格分析，2)对至少5个其他项目进行不那么密集的分析。由此将产生新的和创新的见解，即什么可能真正构成负责任的创造性人工智能，它的特征和功能，它的局限性和风险。例如，我们的分析将涉及检查在创造力、真实性和责任概念之间产生的伦理和道德紧张关系，并将探索不同类型的责任，例如道德、法律、角色或美德。创造性人工智能提供了一种扩展和丰富人类创造力的方式，人工智能成为一个良性的合作者，使人类能够打破既定实践的限制。从这个角度来看，人工智能的数据管理员成为人工智能生态系统中一个关键的创造性角色。该项目将提供一个范围界定研究，在这个高度竞争和新兴的领域，什么可能构成负责任的人工智能实践。这个特定的背景很重要，因为创意产业在经济和社会上都很重要，而创造力本身是由一系列伦理和认识论的考虑所支撑的，这些考虑直接影响到任何负责任的人工智能的概念：我们不能简单地引入为银行或金融服务等其他部门创造的责任方法。也许同样重要的是，时间已经不多了：斯坦福大学最近的一项研究发现，所有领域的领先人工智能模型都没有接近符合欧盟人工智能法案草案。

Identifying stakeholders and delineating boundaries of ecosystems is especially challenging in the context of Creative AI applications as users of these applications including artists, the public, local authorities or national institutions, do not typically participate in their development. However, defining boundaries and stakeholders is important because different RAI considerations apply if the boundaries are drawn narrowly or widely; drawing narrow boundaries risks excluding relevant stakeholders from these ecosystems. Drawing boundaries widely risks any considerations of RAI becoming too complex to apply or identify. A key objective of this project is to develop the structure of a dynamic archive that can be used to identify the stakeholders in these ecosystems in the context of Creative AI. A parallel objective is to use the archived data to highlight the RAI considerations that need to be addressed in each project and to compare the boundaries of these AI ecosystems. This will generate deeper insight into how responsibility within AI ecosystems itself can be better understood.This project will lay the foundation work for mapping RAI ecosystems in the context of Creative AI by using bottom-up evidence already collected in specific research projects. We interpret AI ecosystems as interlinked ecosystems consisting of different individual actors and groups interacting in complex ways with one another and with AI applications. These ecosystems are Responsible AI Ecosystems when they pay close attention to responsibility challenges as identified by RAI UK research. The project will have two stages: 1) a rigorous analysis of Digiscore (https://digiscore.github.io/pages/aboutus/Creative) which uses AI to create music for disabled musicians and CAT Royale (https://www.blasttheory.co.uk/projects/cat-royale) which uses AI to interact with pet animals and 2) a less intensive analysis of at least 5 additional projects. From this will emerge new and innovative insight into what might actually constitute responsible creative AI, its characteristics and features, its limitations and risks. For example, our analysis will involve examining the ethical and moral tension arising between the concepts of creativity, authenticity and responsibility, and will explore the different types of responsibility e.g. moral, legal, role or virtue attaching to this context.Creative AI offers a way of extending and enriching human-based creativity, where AI becomes a benign collaborator enabling humans to break the constraints of established practice. From this perspective, the data curator for the AI becomes a key creative role within the AI ecosystem. This project will provide a scoping study of what might constitute responsible AI practice in this highly contested and emergent area. This specific context is important because creative industries are important economically and societally while creativity itself is underpinned by a range of ethical and epistemological considerations which impinge directly on any notion of responsible AI: we cannot simply import approaches to responsibility created for other sectors such as banking or financial services. Perhaps equally importantly, time is running out to get this right: a recent study by Stanford University found that none of the leading AI models in all sectors came close to being compliant with the draft European Union AI Act.