AI Stereotypes Against Older Adults Detected in ChatGPT-4o

Study Reveals Age-Related Bias in AI-Generated Content

According to НВ — Техно: A research team led by Professor Moon Choi from the KAIST Graduate School of Science and Technology Policy has uncovered subtle stereotypes about older adults in texts produced by OpenAI's ChatGPT-4o. After analyzing 900 text samples, the team found that the model portrays people aged 60 and above as warm yet less competent-a pattern that mirrors common media stereotypes and risks fueling digital ageism. This finding is especially relevant as AI tools become increasingly embedded in everyday communication and decision-making.

How Age Groups Are Characterized

Doctoral candidate Wang Hong, the study's first author, examined how the model described traits across age groups ranging from 10 to 90 years old, in ten-year increments. Using the Stereotype Content Model, researchers found that individuals aged 60 and older scored high on warmth-encompassing kindness, trustworthiness, and attentiveness. However, they scored lower on competence, which includes ability, expertise, and efficiency, compared to younger groups such as teenagers and people in their twenties.

The study also revealed that descriptions of people aged 70 and above were relatively uniform in nature. ChatGPT-4o tends to depict older adults as wise and caring, but with comparatively lower agency and active capabilities.

“Bias in AI is not just a technological problem, but a social one,” said Professor Moon Choi.

He further emphasized that “to create inclusive AI, people from different generations must take part in the development process.”

The research team highlighted a serious risk of digital ageism arising from such stereotypical portrayals of older adults. These findings underscore the importance of approaching AI use critically and the need to incorporate age diversity into technology development.

The stereotypes identified could have significant societal consequences, shaping biased attitudes toward older adults in both media and technology. These results point to the necessity of actively involving representatives from various age groups in AI design to prevent similar biases in the future. Embracing diversity during the creation of new technologies can foster a more inclusive and equitable approach for all age categories.

As AI-generated content becomes more prevalent, understanding the underlying biases is crucial. For instance, a recent analysis revealed that AI writers often rely on limited vocabulary, with a specific name appearing disproportionately in their narratives. This highlights the challenges of diversity in AI language models and raises further questions about representation. To explore this issue in depth, you can read more about the findings on AI writers' language limitations.