A study from the Tasmanian School of Business and Economics at the University of Tasmania shows that AI-generated content can keep harmful gender biases alive.

The researchers looked at AI-generated content discussing what makes a ‘good’ or ‘bad’ leader. They found that men were often portrayed as strong, brave, and skilled, while women were frequently seen as emotional and not very effective.

Biased leadership

“By exploring how AI tools generate responses to questions around good and bad leadership, we were able to show that AI reflects and perpetuates known gender biases,” the researchers explain.

“Any mention of women leaders was completely omitted in the initial data generated about leadership, with the AI tool providing zero examples of women leaders until it was specifically asked to generate content about women in leadership.”

The study found that when generative AI did provide examples of female leaders, they were far more likely to be shown to be bad leaders than men were. The study’s authors argue that these findings highlight the need for more oversight and investigation into AI tools as they become a regular part of our lives.

“Biases in AI models have far-reaching implications beyond just shaping the future of leadership. With the rapid adoption of AI across all sectors, we must ensure that potentially harmful biases relating to gender, race, ethnicity, age, disability, and sexuality aren’t preserved,” they conclude.

“Our research highlights the need for ongoing monitoring and evaluation of AI-generated content to ensure that it is not perpetuating harmful biases. We hope that our research will contribute to a broader conversation about the responsible use of AI in the workplace.”