Better Little People Pictures: Generative Creation of Demographically Diverse Anthropographics
Date
2024-05-11
Journal Title
Journal ISSN
Volume Title
Publisher
ACM
Abstract
We explore the potential of generative AI text-to-image models to help designers efficiently craft unique, representative, and demographically diverse anthropographics that visualize data about people. Currently, creating data-driven iconic images to represent individuals in a dataset often requires considerable design effort. Generative text-to-image models can streamline the process of creating these images, but risk perpetuating designer biases in addition to stereotypes latent in the models. In response, we outline a conceptual workflow for crafting anthropographic assets for visualizations, highlighting possible sources of risk and bias as well as opportunities for reflection and refinement by a human designer. Using an implementation of this workflow with Stable Diffusion and Google Colab, we illustrate a variety of new anthropographic designs that showcase the visual expressiveness and scalability of these generative approaches. Based on our experiments, we also identify challenges and research opportunities for new AI-enabled anthropographic visualization tools.
Description
Keywords
Citation
Dhawka, P., Perera, L., & Willett, W. (2024). Better little people pictures: Generative creation of demographically diverse anthropographics. Proceedings of the 2024 CHI Conference on Human Factors in Computing Systems.