Assessing Implicit Gender Bias in Large Language Models for STEM Career Guidance Applications

As large language models (LLMs) become more ubiquitous, their utilization as trusted advisors for critical life decisions, including career exploration, will increase. While research has the embedding of various human biases within LLMs, further investigation of possible historical biases regarding female participation in STEM is needed. To address this gap, we assess evidence of potential embedded gender biases in modern LLMs through a structured analysis of responses to custom engineered prompts representing common inquires of career advice seekers. Responses are summarized using word clouds in order to generate summary visualizations capable of quickly identifying potential indicators of bias. We demonstrate that LLMs offer a unique perspective on gender dynamics in STEM, revealing progress and challenges while empowering individuals, especially underrepresented groups, to actively contribute to a more balanced and diverse STEM community.