Full Breakdown
Gender Bias Persists in Virtual-Office AI Assistants
By Drooid · · How we work
Study Overview
Researchers placed 189 participants in a virtual-reality office and asked them to complete work-related tasks alongside human-like AI assistants that differed only in visual gender presentation. The male-presenting agent, named Johan, and the female-presenting agent, Johanna, were powered by identical underlying technology. After each task, participants received real money and were instructed to divide a monetary reward between themselves and the AI assistant, creating a tangible transaction that reflected perceived contribution.
Findings and Statistics
Despite identical capabilities, the male-presenting AI received higher monetary rewards. On average, Johanna was paid 10.25 percent less than Johan for the same work. Participants also rated the male agent as more human-like and reported greater trust in his output. Although many said they had no explicit preference for a male or female AI, their allocation behavior revealed a contrary pattern.
Researchers’ Interpretation
Dr Mary Hausfeld, Assistant Professor at the Kemmy Business School, University of Limerick, noted that the results demonstrate how design choices can activate gender assumptions that already exist in human workplaces. She argued that the technology itself is neutral, but visual cues such as gendered appearance can shape user behavior and reinforce existing inequalities. The study’s co-authors—including Isabelle Cuber and Tarek Alakmeh of the University of Zurich, Anand van Zelderen of SKEMA Business School, and Moritz Jenny, Jochen Menges and Thomas Fritz of the University of Zurich—concluded that careful consideration of AI characteristics is needed to avoid reproducing bias in emerging work environments.
Context and Implications
The experiment extends longstanding evidence of gender bias in human labor markets to interactions with artificial agents. As AI assistants become more common in corporate settings, the assumption of algorithmic neutrality may mask subtle design-driven disparities. The findings suggest that organizations should evaluate not only the functional performance of AI tools but also the visual and social cues they convey, to prevent inadvertent reinforcement of gender-based inequities.
Upcoming Presentation
The research will be presented at the 14th Nordic Conference on Human-Computer Interaction in Vaasa, Finland, from October 5 to 7, and will appear in the conference proceedings published by the Association for Computing Machinery.