令和6年度 学際共創プロジェクト【デジタルヒューマニティ部門】
Identifying and considering the social implications of private information in image datasets
Amelia KATIRAI: Research Center on Ethical, Legal, and Social Issues
Noa GARCIA DOCAMPO: Institute for Datability Science
Kazuki IDE: Center for Infectious Disease Education and Research
Yankun WU: Intelligence and Sensing Lab
Yuta NAKASHIMA: Institute for Datability Science
Atsuo KISHIMOTO: Research Center on Ethical, Legal, and Social Issues
Research background
The development of image generation models and their free availability to members of the public has made use of these technologies widespread, before their social and ethical issues have been adequately considered. Data for image generation models have been shown to contain medical images, in addition to other sensitive imagery, posing a privacy risk (1). Much of this data is scraped from online sources, including social media and other platforms where individuals may share their private images, without awareness that these images may be scraped and used for other purposes. Furthermore, when image generation models are trained, they risk outputting this private information to users of the models (2). However, there has been insufficient attention to image generation models, despite their particular sensitivity.
Research aims
The overarching aim of this project was to investigate privacy issues in image generation models, with three sub-aims:
1) To verify whether image-based private information exists in the datasets used for major image generators;
2) To identify the social and ethical implications of these issues;
3) To investigate the perspectives of a survey sample of members of the public on these issues.
本年度の成果について、詳しくは活動報告書（PDF）をご覧ください。