Researchers at the University of Limerick found that study participants paid female-presenting AI assistants 10.25% less than male-presenting counterparts for identical work. The experiment, which placed 189 people in a virtual reality office, reveals that human-like gender cues can trigger workplace biases, even when the underlying AI technology is identical.
The Virtual Reality Experiment at the University of Limerick
To determine if gender bias translates from human workplaces to digital ones, researchers including Dr. Mary Hausfeld of the University of Limerick’s Kemmy Business School conducted a controlled study. They placed 189 participants in a virtual reality office where they performed tasks alongside four types of AI assistants: a text-based chatbot, a desk robot, and two human-like agents named Johan (male) and Johanna (female).
The study, titled Human-Like and Male? How AI Assistant Design Relates to Trust and Monetary Reward at Work in VR,
used a real-money payout mechanism to incentivize participants. After completing work, individuals were asked to decide how much real money was to be awarded to the assistants for their contributions, a method the researchers used to examine how much people were willing to pay AI for its work.
The research was conducted by researchers from the University of Limerick, the University of Zurich, and SKEMA Business School.
Monetary Disparity and Perceptions of Humanity
Despite the fact that Johan and Johanna relied on the same underlying model—OpenAI’s gpt-4-1106-preview—the economic outcomes were starkly different. Johanna was paid 10.25% less than Johan for performing the exact same tasks. Participants consistently perceived the male-presenting agent as more human-like than the female-presenting one.
This result occurred even though participants claimed in interviews that an AI’s gender was irrelevant to their decision-making. We often think of AI as being neutral, but the way we design and present these systems can activate those same assumptions and biases that exist in our interactions with other people,
Dr. Hausfeld noted.
Participant Behavior Versus Stated Preferences
The research highlighted a significant gap between what participants reported and how they behaved. While many stated they preferred non-human AI, their financial decisions showed a clear bias toward human-like assistants, which were generally trusted and credited more than the robot-like versions.
Dr. Hausfeld warned that as AI agents become more embedded in knowledge work, designers must be cautious. As AI agents become more common in the workplace, we need to think carefully about the characteristics we give them and the behaviours those choices may encourage.
We don’t want to inadvertently reproduce existing inequalities in a new technological setting,
she said.
Implications for Enterprise AI Design
With IT consulting firm Accelirate reporting that nearly 79% of companies used AI agents this year, alongside an additional 40% of applications employing them, the study raises practical questions for organizations. When developers assign names, voices, and faces to AI, they are not merely making aesthetic choices; they are influencing how users trust and value the technology. The researchers conclude that the industry faces a challenge: whether to audit these design choices with the same rigor currently applied to the underlying AI models to prevent the automation of workplace pay gaps.
