The question of how we can use ethics and ethical frameworks to avert the negative consequences of AI through guidance on human behaviour and the design of technological systems has recently been receiving increasing attention. The appropriate response to an ethics of AI has certainly been contentious. For some years the wisdom of deontology and utilitarianism in the ethics of technology has been questioned. Today, a kind of AI ethics principlism has gained a degree of widespread acceptance, yet it still invites harsh rejections in recent scholarship. In this paper, we wish to explore the contribution to an ethics of AI made by a narrative philosophy and ethics of technology inspired by the ‘little ethics’ of Paul Ricoeur, and virtue ethics of Alasdair MacIntyre, most recently and promisingly built upon by Wessel Reijers and Mark Coeckelbergh. The objective of this paper is to examine the extent to which a narrative and virtue based ethics (or, VPD, i.e., virtuous practice design) might be a plausible candidate for the foundation of an ethics of AI, or rather ethical AI practice. This will be achieved by exploring the ways in which this approach can respond to some of the significant faults with or critiques of applied and principles and guidelines based ethical approaches to AI ethics.

Algorithms and AI tools are becoming increasingly influential artefacts in commercial and governance contexts. Algorithms and AI tools are not value neutral; to some extent they must be rendered knowable and known as objects, and in their implementation and deployment, to see clearly and understand their implications for moral values, and what actions can be undertaken to optimise them in their design and use towards ethical goals, or whether they are even suitable for particular goals. Transparency is a term with variable uses and interpretations, a problem which can challenge its use in design and policy. Here, we attempt to further clarify transparency. We argue that transparency is the state of affairs that obtains when relevant and understandable information about some X is available and accessible to some target audience (A), so that this information is sufficient for A for the purpose (P). Moreover, we connect this conceptualisation with transparency’s moral value, where P is to provide an account about X’s supportive or conflicting relationship with relevant values and goals. Such teleological ends in our context here can be the ability to account for the degree to which an algorithm, process or organisation respects certain values and is conducive to (social) goals.

This paper builds upon recent work in narrative theory and the philosophy of technology by examining the place of transparency and responsibility in discussions of AI, and what some of the implications of this might be for thinking ethically about AI and especially AI practices, that is, the structured social activities implicating and defining what AI is. In this paper, we aim to show how pursuing a narrative understanding of technology and AI can support knowledge of process and practice through transparency, as well help summon us to responsibility through visions of possibility and of actual harms arising from AI practices. We provide reflections on the relations between narrative, transparency and responsibility, building an argument that narratives (about AI, practices, and those persons implicated in its design, implementation, and deployment) support the kind of knowing and understanding that is the aim of transparency, and, moreover, that such knowledge supports responsibility in informing agents and activating responsibility through creating knowledge about something that can and should be responded to. Furthermore, we argue for considering an expansion of the kinds of practices that we might legitimately consider ‘AI practices’ given the diverse set of (often materially embedded) activities that sustain and are sustained by AI that link directly to its ethical acceptability and which are rendered transparent in the narrative mode. Finally, we argue for an expansion of narratives and narrative sources to be considered in questions of AI, understanding that transparency is multi-faceted and found in stories from diverse sources and people.

To make evaluations about the morally relevant impacts of algorithms, transparency is needed. This paper lays out discussion of algorithms and transparency in an explicitly moral analysis with a special focus on the domain of justice and security. The paper provides an account of the moral import of transparency, defined itself as an instrumental value denoting a state of affairs conducive to acquisition of knowledge about some X. A normative account of transparency is outlined relying on an intuitionist framework rooted in the works of Ross and Robert Audi. It will be argued that transparency can be derived as a subsidiary (prima facie) principle from other duties including beneficence and justice and that it is groundable in the value of knowledge. Building on this foundation, the paper examines transparency and duty conflict with a special focus on algorithms in justice and security, recognising that complete transparency can be impossible where duties conflict. It is argued that as a subsidiary (prima facie) principle, transparency is overridable but ineradicable, which is to say that sufficiently justifiable reasons for secrecy or opacity can licence limiting transparency, that is, there may be occasion where full transparency is not our final duty.

AI for good, and AI for sustainability projects, are being developed by often well-meaning innovators across the world, intending to support initiatives in sustainable development. Some such projects have been positioned within the framework of the Sustainable Development Goals. In this paper, we critically engage with this phenomenon using a virtue-based Ricoeurian narrative philosophy. Through conceptual analysis and normative argumentation, we make a theoretical contribution to scholarship on this topic. We argue that SDGs could be regarded as internationally agreed high-order end-norms that crystallise values considered to constitute the good life. We argue for exercising caution in applying SDGs as end-norms to AI projects as they are of a high order, are not directly action guiding, and threaten to sediment presentist dominant values. We argue for narrative hospitality between AI innovators and community stakeholders to guide reflection on specifying the SDGs to more situated contexts and to support the development of AI systems that may be more contextually appropriate, and more capable of supporting plural visions of the good life. Such narrative hospitality represents accommodation between parties, but refusal or resistance to engagement with AI innovators can be justified when community interests are not respected, and actors within AI industries can also be morally obliged to resist or refuse development or deployment of systems for contextually inappropriate environments that would be incompatible with a critical reading of the SDGs formed in narrative exchange with the other.