Abstract
In this visual inquiry, we examine how various ethical theories are visualized through generative AI tools of text-to-image (T2I) models. We find that these models, themselves, being subject to data augmentation with input and output from users, data amalgamation, and interactions, are continuously updated and emancipated in the fractured and ever-changing lens akin to a kaleidoscope. To account for this change, and to understand how ethical theories are embedded within these systems, we investigate ethics that is categorized in definition and in practice, formalized as normative and meta-ethics respectively. We conducted semi-structured interviews with ten experts in ethics to formulate notions of ethical theories and generate imagery through text-to-image (T2I) models. We then showed our gallery of images to experts and found eight themes that highlight how the context of the image and metaphors of interpretation influence the theories at hand. We discuss the implications of our work for the critical examination of generative AI models, design considerations for knowledge dissemination in these models, and the subjective lens through which perception molds foundational understanding.
Presenters
Alayt Abraham IssakStudent, Ph.D. in Interdisciplinary Design and Media, College of Arts, Media and Design (CAMD), Northeastern University, Massachusetts, United States
Details
Presentation Type
Paper Presentation in a Themed Session
Theme
KEYWORDS
Visual Perception, Text-To-Image Models, Research Through Design, Ethical Theories
