Ethical challenges for computer vision in the social sciences

Abstract

The rising accessibility of large image datasets and pre-trained models allowed a rapid expansion of computer vision (CV) methods in social science research. While this methodological development opens new potential applications, this combination also raises a new set of ethical concerns that neither computer vision nor social science ethical studies fully cover. This paper provides an analysis of the ethics of CV methods in social science, articulated around two stages of a research project: The first part explores methodological concerns, starting with data collection, data and modeling bias, and data storage afterwards, with special attention to issues of consent and explainability. The second part investigates the following project stage of result dissemination, through publication strategies, reproducibility, and finally public engagement. Building on this analysis, we propose an actionable checklist of ten ethical recommendations for future CV projects in the social sciences. This contribution fills the gap at the crossroads between AI ethics and applied social science, bringing concrete guidance for future research in the interdisciplinary, ethically complex field.