Gender Bias in Artificial Intelligence-Written Letters of Reference

Area of Science:

• Medical Education
• Artificial Intelligence in Healthcare
• Linguistics

Background:

• Letters of reference (LORs) are crucial for postgraduate residency applications.
• Human-written LORs often contain implicit gender bias, disadvantaging women.
• Artificial intelligence (AI) tools like ChatGPT are increasingly used for drafting professional documents.

Purpose of the Study:

• To investigate whether AI-generated LORs exhibit gender bias.
• To compare gendered language in AI-generated LORs for male and female applicants.
• To identify factors influencing gender bias in AI-generated LORs.

Main Methods:

• An observational, multicenter study was conducted.
• Identical prompts for male and female Otolaryngology residency applicants were used to generate LORs via ChatGPT.
• A gender-bias calculator analyzed the proportion of male- versus female-associated words in the generated LORs.

Main Results:

• All ChatGPT-generated LORs demonstrated a bias toward male-associated words.
• No significant difference in male-biased words was found between LORs for male and female applicants.
• Significant gender bias differences were observed based on other variables like school and research activities.

Conclusions:

• AI-generated LORs consistently use male-associated language.
• ChatGPT LORs did not show gender bias based on applicant name or pronouns.
• Users must be aware of potential biases in AI-generated LORs, as factors beyond applicant gender can introduce gendered language.