AI may be making women's emails and job applications sound less professional (www.businessinsider.com)

A study from Johns Hopkins University highlights a significant bias in how AI models generate workplace communication, revealing that emails and job applications prompted with feminine-coded language result in less formal, less complex responses. Researchers tested OpenAI's GPT-4, Meta's Llama, Google's Gemini, and Mistral's Vibe, discovering that prompts containing gender-associated language patterns led to divergent outputs. For instance, a male-coded prompt generated a professional, succinct reply, while a female-coded prompt elicited a more emotionally expressive and verbose response, illustrating a clear disparity in tone and style. This finding is crucial for the AI/ML community as it raises awareness about inherent biases in language models that may reinforce stereotypes, particularly regarding gender. The implications are significant, especially as AI becomes increasingly integrated into professional environments. The researchers emphasize the necessity for companies to address these biases in AI models rather than placing the onus on users to navigate the complexities of language. As interactions with AI evolve and become more conversational, it will be critical to ensure that these systems produce equitable and professional communication, regardless of gendered language nuances.