Skip to main content

Lund University Publications

LUND UNIVERSITY LIBRARIES

Age bias in artificial intelligence (AI): a visual properties analysis of AI images of older versus younger people

Allen, Laura ; Xu, Wenqian LU orcid ; Nishikitani, Mariko ; Atul Patil, Vaishnavi and Bradley, Dana (2023) p.986-986
Abstract
Concerns about age-related biases in Artificial Intelligence (AI)’s design, development, and data generation raise alarms for potential social inequalities. AI art-generators like MidJourney, known for creating distinctive and artistic images, have gained attention recently. However, the critical examination of their potential biases remains limited. Our study aims to investigate age-related bias in MidJourney by comparing how older and younger individuals are visually depicted. Using 19 keywords (as text prompts) tied to domains of potential social exclusion for older adults, we automated the collection of “old person” and “young person” images in MidJourney. The text-to-image prompts yielded 456 AI-generated images that were analyzed... (More)
Concerns about age-related biases in Artificial Intelligence (AI)’s design, development, and data generation raise alarms for potential social inequalities. AI art-generators like MidJourney, known for creating distinctive and artistic images, have gained attention recently. However, the critical examination of their potential biases remains limited. Our study aims to investigate age-related bias in MidJourney by comparing how older and younger individuals are visually depicted. Using 19 keywords (as text prompts) tied to domains of potential social exclusion for older adults, we automated the collection of “old person” and “young person” images in MidJourney. The text-to-image prompts yielded 456 AI-generated images that were analyzed with Amazon Rekognition image detection software to transform them to quantitative data. Data were compared statistically using STATA (version 16). The results show that the images of older people are significantly less bright (p< 0.001) and less sharp (p< 0.001) than the images of younger people. Additionally, images of older people have significantly more smile expressions (16%; p< 0.001) and are more likely to be wearing glasses (56%; p< 0.001) compared to images of younger people (6% and 12%, respectively). The findings reveal that MidJourney’s algorithm associates older people with a negative sense of antiquity, irrelevance, and dread. The findings also align with the Stereotype Content Model in which older people are seen as having warm personalities but not as competent or capable as younger people. We interpret these results as age bias that the AI art generator MidJourney has learned and in turn, creates, through image generation. (Less)
Please use this url to cite or link to this publication:
author
; ; ; and
organization
publishing date
type
Contribution to conference
publication status
published
subject
pages
986 - 986
DOI
10.1093/geroni/igad104.3168
project
Ageism in the Media
language
English
LU publication?
yes
id
40034fcd-21a4-4f24-91a0-e1135f45cdf3
date added to LUP
2023-12-24 13:43:45
date last changed
2024-02-06 13:31:12
@misc{40034fcd-21a4-4f24-91a0-e1135f45cdf3,
  abstract     = {{Concerns about age-related biases in Artificial Intelligence (AI)’s design, development, and data generation raise alarms for potential social inequalities. AI art-generators like MidJourney, known for creating distinctive and artistic images, have gained attention recently. However, the critical examination of their potential biases remains limited. Our study aims to investigate age-related bias in MidJourney by comparing how older and younger individuals are visually depicted. Using 19 keywords (as text prompts) tied to domains of potential social exclusion for older adults, we automated the collection of “old person” and “young person” images in MidJourney. The text-to-image prompts yielded 456 AI-generated images that were analyzed with Amazon Rekognition image detection software to transform them to quantitative data. Data were compared statistically using STATA (version 16). The results show that the images of older people are significantly less bright (p&lt; 0.001) and less sharp (p&lt; 0.001) than the images of younger people. Additionally, images of older people have significantly more smile expressions (16%; p&lt; 0.001) and are more likely to be wearing glasses (56%; p&lt; 0.001) compared to images of younger people (6% and 12%, respectively). The findings reveal that MidJourney’s algorithm associates older people with a negative sense of antiquity, irrelevance, and dread. The findings also align with the Stereotype Content Model in which older people are seen as having warm personalities but not as competent or capable as younger people. We interpret these results as age bias that the AI art generator MidJourney has learned and in turn, creates, through image generation.}},
  author       = {{Allen, Laura and Xu, Wenqian and Nishikitani, Mariko and Atul Patil, Vaishnavi and Bradley, Dana}},
  language     = {{eng}},
  month        = {{12}},
  pages        = {{986--986}},
  title        = {{Age bias in artificial intelligence (AI): a visual properties analysis of AI images of older versus younger people}},
  url          = {{http://dx.doi.org/10.1093/geroni/igad104.3168}},
  doi          = {{10.1093/geroni/igad104.3168}},
  year         = {{2023}},
}