Intersectional analysis of visual generative AI: the case of stable diffusion
2025 (English)In: AI & Society: Knowledge, Culture and Communication, ISSN 0951-5666, E-ISSN 1435-5655, Vol. 40, no 6, p. 4341-4362Article in journal (Refereed) Published
Abstract [en]
Since 2022, Visual Generative AI (vGenAI) tools have experienced rapid adoption and garnered widespread acclaim for their ability to produce high-quality images with convincing photorealistic representations. These technologies mirror society's prevailing visual politics in a mediated form, and actively contribute to the perpetuation of deeply ingrained assumptions, categories, values, and aesthetic representations. In this paper, we critically analyze Stable Diffusion (SD), a widely used open-source vGenAI tool, through visual and intersectional analysis. Our analysis covers; (1) the aesthetics of the AI-generated visual material, (2) the institutional contexts in which these images are situated and produced, and (3) the intersections between power systems such as racism, colonialism, and capitalism-which are both reflected and perpetuated through the visual aesthetics. Our visual analysis of 180 SD-generated images deliberately sought to produce representations along different lines of privilege and disadvantage-such as wealth/poverty or citizen/immigrant-drawing from feminist science and technology studies, visual media studies, and intersectional critical theory. We demonstrate how imagery produced through SD perpetuates pre-existing power systems such as sexism, racism, heteronormativity, and ableism, and assumes a default individual as white, able-bodied, and masculine-presenting. Furthermore, we problematize the hegemonic cultural values in the imagery that can be traced to the institutional context of these tools, particularly in the tendency towards Euro- and North America-centric cultural representations. Finally, we find that the power systems around SD result in the continual reproduction of harmful and violent imagery through technology, challenging the oft-underlying notion that vGenAI is culturally and aesthetically neutral. Based on the harms identified through our qualitative, interpretative analysis, we bring forth a reparative and social justice-oriented approach to vGenAI-including the need for acknowledging and rendering visible the cultural-aesthetic politics of this technology and engaging in reparative approaches that aim to symbolically and materially mend injustices enacted against social groups.
Place, publisher, year, edition, pages
SPRINGER , 2025. Vol. 40, no 6, p. 4341-4362
Keywords [en]
Critical AI; Stable diffusion; Generative AI; Algorithmic reparation; Restorative justice; Intersectionality
National Category
Human Computer Interaction
Identifiers
URN: urn:nbn:se:liu:diva-212540DOI: 10.1007/s00146-025-02207-yISI: 001446077100001Scopus ID: 2-s2.0-105001164175OAI: oai:DiVA.org:liu-212540DiVA, id: diva2:1947259
Note
Funding Agencies|Royal Institute of Technology; Wallenberg AI, Autonomous Systems and Software Program - Humanities and Society (WASP-HS) - Marianne and Marcus Wallenberg Foundation grant [QC 20230614]; European Union's Horizon 2020 Research and Innovation programme under the Marie Sklodowska-Curie grant [955422]
2025-03-252025-03-252025-12-01Bibliographically approved