Luger, Ewa

Speed, Chris

Bennett, S. J.

2023-11-13T16:23:44Z

2023-11-13T16:23:44Z

2023-11-13

Advances in Artificial Intelligence (AI) research and development have seen AI applied in various high-stakes domains such as healthcare and welfare. Furthermore, portrayals of AI are often characterised by narratives of perpetual progress and sleek optimisation, obscuring the intricate interactions of materiality and socio-political decision-making inherently embedded within wider systems of design and development. The resulting ethical and social concerns have prompted proposal of numerous frameworks, tools and guidelines for the ethical design and development of AI. However, translating these proposals into practice has proven challenging, and there is a paucity of research into the practical contexts shaping the ethico-onto-epistemology of AI practice. In this thesis I illustrate these contexts via the accounts of 24 AI practitioners, complemented by ethnographic observations from an industry research lab, examining the values which motivate practitioners, the constraints which shape their practice, and their approaches to ethics. Weaving through these discussions of practice, values, and the nature of responsibility, I examine how ambiguities pervade practice and shape the realities of ethical reflection and engagement at all stages of development. My findings uncover practitioner motivations linked with interconnected intellectual and moral values, how these related to intellectual conduct and culture within the field, and how practitioner heuristics for ethical decision-making are often relational and character-based in nature. This realization of values in practice is tempered by numerous constraints including hardware limitations, epistemic cultures, and ethical knowledge. Drawing upon the Ethics of Ambiguity, I discuss how the uncertainty, ambiguity and unequal access to resources shaping AI practice necessitate a process-focused ethics which pivots away from solutions, towards critical contextual reflexivity and awareness of how contexts impact realisation of values. To this end, I demonstrate how The Ethics of Ambiguity can offer a path forward for ethical AI practice. This vision of AI practice embraces ambiguities rather than attempting to segment and sideline them, focusing on how practitioner decisions (and their eventual outputs) impact others’ freedoms while acknowledging the multiplicity of values across socio- and geo-political contexts.

https://hdl.handle.net/1842/41155

http://dx.doi.org/10.7488/era/3891

en

The University of Edinburgh

Bennett, SJ., Claisse, C., Luger, E., and Durrant, A., 2023. Unpicking Epistemic Injustices in Digital Health: On Designing Data-Driven Technologies to Support the Self-Management of Long-Term Health Conditions. In Proceedings of the 2023 AAAI/ACM Conference on AI, Ethics, and Society.

AI ethics

Artificial Intelligence

ethics

STS

Science and Technology Studies

Transmuting values in artificial intelligence: investigating the motivations and contextual constraints shaping the ethics of artificial intelligence practitioners

Thesis or Dissertation

Doctoral

PhD Doctor of Philosophy