Do responses generated by synthetic intelligence methods corresponding to ChatGPT mirror social prejudice? A KAIST analysis staff has quantitatively analyzed and recognized age-related stereotypes embedded in the responses of generative synthetic intelligence. The research sheds mild on the potential affect of hidden AI biases on social perceptions and suggests instructions for the event of extra inclusive AI.
KAIST, led by President Kwang Hyung Lee, introduced on the twenty eighth {that a} analysis staff led by Professor Moon Choi of the Graduate School of Science and Technology Policy quantitatively analyzed delicate stereotypes about older adults embedded in sentences generated by OpenAI’s generative AI mannequin ChatGPT-4o.
Generative AI is now broadly used in on a regular basis data search and decision-making processes, however considerations have additionally been raised that it could reproduce social biases contained in its coaching knowledge. While earlier research have primarily targeted on biases associated to gender or race, this research, performed by Ph.D. scholar Wan Hong as the primary creator, is critical in that it examined ageism from the attitude of synthetic intelligence at a time when the difficulty is turning into more and more necessary amid world inhabitants ageing. Ageism refers to discrimination towards, or damaging perceptions of, sure teams primarily based on age.
The analysis staff collected 900 textual content samples generated by GPT-4o utilizing impartial prompts that requested the mannequin to explain the traits of age teams from 10 to 90 in 10-year intervals. The staff then analyzed the responses utilizing the Stereotype Content Model, a significant principle in social psychology that explains perceptions of individuals or teams alongside two dimensions: heat and competence.
The evaluation discovered that older adults, outlined as these aged 60 and above, acquired excessive scores for “warmth,” a trait related to kindness, trustworthiness, and consideration. However, their scores for “competence,” which refers to potential, experience, and effectivity, tended to be comparatively decrease than these of youthful age teams.
The generated responses additionally tended to painting the human life course as divided into three teams: youth, overlaying these in their teenagers and 20s; center age, overlaying these in their 30s to 50s; and older maturity, overlaying these in their 60s and above. In specific, descriptions of individuals aged 70 and older repeatedly confirmed comparatively uniform traits.
The analysis staff additionally targeted on “assertiveness,” which refers back to the tendency to actively specific one’s opinions and act with confidence and initiative. The evaluation confirmed that the frequency of expressions associated to assertiveness decreased as age elevated. This means that ChatGPT-4o tends to painting older adults as clever and caring, whereas representing their company and energetic capacities as comparatively decrease.
This research is critical as a result of it quantitatively recognized delicate biases embedded in generative AI by combining social science principle with computational evaluation strategies. The findings present that generative AI tends to painting older adults as a “warm but less competent” group, a sample much like typical stereotypes of older adults repeatedly discovered in mass media.
This research is critical as a result of it quantitatively recognized delicate biases embedded in generative AI by combining social science principle with computational evaluation strategies. The findings present that generative AI tends to painting older adults as a “warm but less competent” group, a sample much like typical stereotypes of older adults repeatedly discovered in mass media.
“Bias in AI is not merely a technological issue, but a social one,” mentioned Professor Moon Choi. “To build inclusive artificial intelligence, people from diverse generations must participate in the development process.”
The research was performed with Ph.D. scholar Wan Hong of the Graduate School of Science and Technology Policy as the primary creator. The findings have been printed in the February 2026 particular subject of The Gerontologist, a number one worldwide journal in the sector of gerontology with an affect issue of 5.7.
※ Paper title: “An Exploratory Semantic Analysis of Age-Related Stereotypes in OpenAI’s GPT-4o Model”
※ DOI: https://doi.org/10.1093/geront/gnaf291
This analysis was supported by the National Research Foundation of Korea via the Mid-Career Research Program for Convergence between Science and Technology and the Humanities and Social Sciences.
※ Research staff homepage: https://aging.kaist.ac.kr