AI’s Regimes of Representation: A Community-centered Study of Text-to-Image Models in South Asia

Existing socio-technical evaluations of AI have primarily centered on conversations with artists through interviews, with only a handful focusing on artists experimenting with tools directly or bringing artists into conversation with one.

“While these get closer to understanding AI as a socio-technical artifact, these studies have primarily centered on conversations with artists through interviews, with only a handful studies focusing on artists experimenting with tools directly or bringing artists into conversation with one [18, 19, 21-24].”

By elevating their expertise, the authors follow other forms of community-centered evaluations of AI and add to existing methods for evaluating the complex interplay between technology, creativity and socio-politics of the art world.

“By elevating their expertise we follow other forms of community-centered evaluations of AI [14, 34-36] and add to existing methods for evaluating the complex interplay between technology, creativity and socio-politics of the art world.”

The absence of underrepresented cultures from generative models is even harder to point out.

“The absence of underrepresented cultures from these models is even harder to point out (Qadri et al. 2023).”

Generative AI outputs tend to reproduce normative views and mirror societal biases for marginalized groups, including people with disabilities.

“Generative AI outputs tend to reproduce normative views and mirror societal biases for marginalized groups [15, 21, 36, 41, 42, 45, 53, 66, 72, 91], including PwD [10, 12, 39, 48, 54, 62, 77, 113].”

Existing research predominantly focuses on understanding, defining or evaluating representational 'harms' and often uses poor AI-generated images to elicit feedback from marginalized communities.

“Our approach to defining and evaluating 'good' representation differs from existing research that predominantly focuses on understanding, defining or evaluating representational 'harms' [24, 28, 38, 44, 67, 77, 91, 116] and often uses poor AI-generated images to elicit feedback from marginalized communities (e.g., [65, 77, 79]).”

In more community-centric data collection or AI evaluation work, active efforts are made by researchers to solicit broad participation or allow community members to self-select their participation, albeit with some limitations.

“In more community-centric data collection or AI evaluation work, active efforts are made by researchers to solicit broad participation (e.g., across geographical regions [52, 68, 91] and through local research partners [89]), or allow community members to selfselect their participation, albeit with some limitations (e.g., adults, higher-education levels or topic expertise, English-language ability”

Researchers have previously problematized anonymous crowdsourced data annotation approaches as being unspecific about who's worldviews and beliefs become manifest in the data.

“Researchers have previously problematized anonymous crowdsourced data annotation approaches [31, 53, 65] as being unspecific about who's worldviews and beliefs become manifest in the data [11, 17, 51, 78, 90, 91].”

The concern about generative AI's impacts on cultural expression is particularly urgent in non-Western contexts, where the dominance of Western aesthetics and values in AI development raises fears of cultural erasure and marginalization.

“This concern is particularly urgent in non-Western contexts, where the dominance of Western aesthetics and values in AI development raises fears of cultural erasure and marginalization [1, 76, 89].”