Ai's gendered language patterns influence professional communication

Researchers at Johns Hopkins University found that AI-generated workplace writing varies significantly based on the language patterns used in prompts. When prompted with language more commonly associated with women, such as hedging phrases and collective language, AI produced less formal and less complex responses. The study tested prompts for emails, job applications, and resignation letters across four models: GPT-4, Llama, Gemini, and Vibe. Katherine Van Koevering, a postdoctoral fellow at the Johns Hopkins Data Science and AI Institute, noted that the responses differed sharply, with some being so poor they seemed unlikely to be generated by the models. The effect was consistent across all four models, regardless of the name used in the prompt. The study, set to be presented at the Conference on Language Modeling in October, highlights concerns about unconscious language biases in AI, as conversational interactions with AI become more common. Van Koevering emphasized that companies need to address these biases rather than placing the responsibility on users.