The International journal of eating disorders · 2025
Generative AI and Eating Disorders: Exposing Stereotypes in Image Depictions and Setting a Research Agenda.
СтатьяНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Linardon Jake
Аннотация
OBJECTIVE: As generative AI becomes more widely adopted for mental health support and information, concerns exist about its potential to depict psychiatric illnesses in stereotypical ways. This Spotlight investigates how ChatGPT-5 visually represents eating disorders and whether prompt-based manipulations can generate more inclusive outputs, with the goal of laying a roadmap for future research to ensure these technologies are applied in safe, ethical, and clinically responsive ways.
METHODS: ChatGPT-5 was prompted to generate images depicting anorexia nervosa, bulimia nervosa, binge-eating disorder, and body image difficulties. A standard one-shot prompt asked the model to create an artistic, realistic, fictional image of the specified condition. The model was then re-tasked to generate multiple images per condition using two approaches: (1) standard prompts and (2) prompts explicitly instructing variation in age, gender, ethnicity, and body type. Outputs were visually compared to evaluate whether bias-reducing instructions produced more inclusive and diverse depictions.
RESULTS: Standard prompts produced images that reinforced conventional stereotypes, depicting eating disorders largely through young, White, female figures with either emaciated (for anorexia nervosa) or overweight (for binge-eating disorder and body image difficulties) body types. Minimal change was observed with single bias-reducing prompts. In contrast, generating multiple images with explicit diversity instructions produced noticeably more inclusive representations, featuring variation in age, gender and ethnicity, although some stereotypical features persisted.
CONCLUSION: ChatGPT-5's depictions of eating disorders rely on stereotypical templates by default. However, prompting the model to generate multiple images with explicit demographic diversity instructions improved representational inclusivity.
METHODS: ChatGPT-5 was prompted to generate images depicting anorexia nervosa, bulimia nervosa, binge-eating disorder, and body image difficulties. A standard one-shot prompt asked the model to create an artistic, realistic, fictional image of the specified condition. The model was then re-tasked to generate multiple images per condition using two approaches: (1) standard prompts and (2) prompts explicitly instructing variation in age, gender, ethnicity, and body type. Outputs were visually compared to evaluate whether bias-reducing instructions produced more inclusive and diverse depictions.
RESULTS: Standard prompts produced images that reinforced conventional stereotypes, depicting eating disorders largely through young, White, female figures with either emaciated (for anorexia nervosa) or overweight (for binge-eating disorder and body image difficulties) body types. Minimal change was observed with single bias-reducing prompts. In contrast, generating multiple images with explicit diversity instructions produced noticeably more inclusive representations, featuring variation in age, gender and ethnicity, although some stereotypical features persisted.
CONCLUSION: ChatGPT-5's depictions of eating disorders rely on stereotypical templates by default. However, prompting the model to generate multiple images with explicit demographic diversity instructions improved representational inclusivity.