03/08/2026
The foundational data used to train AI models may carry the imprint of societal biases, raising questions about whether these powerful technologies simply echo existing gender stereotypes. A new study co-authored by Professor Li Hongfei, Assistant Professor in the Department of Decisions, Operations and Technology at CUHK Business School, reveals that large language models exhibit a largely balanced perception of both masculine and feminine traits, until they were pushed to think technically. A systematic bias quietly returns once the model is tasked with formulaic investment calculations.
📌 When tasked with describing successful entrepreneurs or evaluating pitches, ChatGPT avoids gender stereotypes and even gives higher marks to collaborative, people-focused, and warm tones.
📌 When instructed to act as a venture capitalist using mathematical formulas to analyse investment opportunities, AI treats the task as a simple math exercise, bypassing its ethical guardrails and favouring risk-embracing, stereotypically masculine traits.
📌 As AI models are trained on extensive data, human-in-the-loop oversight is critical when using algorithmic systems to screen pitches, ensuring that technology promotes fairness rather than reinforcing outdated biases.
Read the full article: https://bit.ly/3TPosss
Follow us for more