▲ Research image (AI-generated)
The Korea Advanced Institute of Science and Technology (KAIST) announced that it has quantitatively analyzed the subtle stereotypes against the elderly embedded in text generated by OpenAI's ChatGPT-4o.
The findings come from a research team led by Professor Choi Moon-jung of the Graduate School of Science and Technology Policy. The team collected 900 texts generated by GPT-4o using neutral prompts that asked the AI to describe the characteristics of people in 10-year age increments, ranging from 10 to 90 years old. They then analyzed the data by applying the Stereotype Content Model, a prominent theory in social psychology.
The analysis confirmed that while the group aged 60 and older scored high in warmth—a trait representing social likability such as kindness and consideration—they tended to score relatively lower than younger age groups in competence, which refers to traits such as professionalism and efficiency.
Furthermore, the generated responses showed a tendency to categorize the human life cycle into three distinct groups: young (teens to 20s), middle-aged (30s to 50s), and elderly (60s and older).
The research team also focused on assertiveness, which indicates confidence and initiative.
The analysis revealed that the frequency of expressions indicating assertiveness tended to decrease as age increased.
The research team explained that this suggests ChatGPT-4o tends to portray the elderly as wise and benevolent figures while depicting them as relatively lower in agency or proactivity.
The team added that the results show generative AI tends to describe the elderly as a group that is warm but relatively low in competence, a pattern similar to the typical stereotypes of the elderly frequently seen in mass media.
The researchers explained that if such expressions are repeatedly exposed through conversational AI services, it could reinforce social prejudices against the elderly.
Furthermore, they raised the possibility that this phenomenon could lead to digital ageism, which hinders the digital participation of the elderly population.
"AI bias is not just a technical issue, but a social one," said Professor Choi Moon-jung. "To achieve inclusive AI, diverse generations must participate in the development process."
(Photo courtesy of KAIST, Yonhap News)
※
Copying, redistribution, and unauthorized use in AI training are strictly prohibited.