Open Collections

UBC Research Data

Replication Data for: The erasure of intensive livestock farming in text-to-image generative AI Sheng, Kehan Sky; Tuyttens, Frank; von Keyserlingk, Marina

Description

Generative AI like ChatGPT has become increasingly integrated into people's daily lives as a vehicle for content creation and information access. While research has revealed how AI could perpetuate biases against marginalized human groups, AI’s impact on non-human animals remains understudied. We found that ChatGPT's text-to-image model (DALL-E 3) is biased toward romanticizing livestock farming as dairy cows grazing on pasture and pigs rooting in mud, even when explicitly prompted for realistic depictions. The public values naturalness and pasture access, but most farmed animals in industrialized countries are housed indoors and intensively. Notably, when we inhibited automatic prompt revision, images shifted to reflect modern farming more realistically: cows accessing feed through metal headlocks in intensive indoor systems, and pigs behind metal railings on concrete floors. While prompt revision was implemented to mitigate bias, it paradoxically erased intensive farming reality and conveyed disinformation that farmed animals live extensively.

Item Metadata

Title	Replication Data for: The erasure of intensive livestock farming in text-to-image generative AI
Creator	Sheng, Kehan Sky; Tuyttens, Frank; von Keyserlingk, Marina
Contributor	Sheng, Kehan Sky
Date Issued	2025-02-24
Description	Generative AI like ChatGPT has become increasingly integrated into people's daily lives as a vehicle for content creation and information access. While research has revealed how AI could perpetuate biases against marginalized human groups, AI’s impact on non-human animals remains understudied. We found that ChatGPT's text-to-image model (DALL-E 3) is biased toward romanticizing livestock farming as dairy cows grazing on pasture and pigs rooting in mud, even when explicitly prompted for realistic depictions. The public values naturalness and pasture access, but most farmed animals in industrialized countries are housed indoors and intensively. Notably, when we inhibited automatic prompt revision, images shifted to reflect modern farming more realistically: cows accessing feed through metal headlocks in intensive indoor systems, and pigs behind metal railings on concrete floors. While prompt revision was implemented to mitigate bias, it paradoxically erased intensive farming reality and conveyed disinformation that farmed animals live extensively.
Subject	Agricultural Sciences; Computer and Information Science; Artificial Intelligence; AI bias; AI ethics; stereotype reinforcement; animal welfare; farming representation bias; coded gaze
Type	Dataset
Date Available	2024-12-12
Provider	University of British Columbia Library
License	CC-BY 4.0
DOI	10.14288/1.0448119
URI	https://doi.org/10.5683/SP3/EAWR6D
Publisher DOI	https://doi.org/10.5683/SP3/EAWR6D
Rights URI	http://creativecommons.org/licenses/by/4.0
Aggregated Source Repository	Dataverse

Item Media

Item Citations and Data

Licence

CC-BY 4.0