In a virtual-reality workplace experiment, human participants allocated 10.25% less money to a female-presenting AI assistant than to a male-presenting one for the same work, according to a University of Limerick announcement. The assistants had the same underlying capabilities. This was a difference in participants’ reward decisions—not a wage gap between AI workers.
What did the study find?
The experiment involved 189 workers completing work-related tasks with AI assistants in a virtual-reality office. Participants were later given real money to divide between themselves and an assistant. The University of Limerick reports that Johanna, the female-presenting agent, received 10.25% less than Johan, the male-presenting agent, for the same work. The announcement says their underlying capabilities were the same.
As an Amazon Associate I earn from qualifying purchases.
The study, “Human-Like and Male? How AI Assistant Design Relates to Trust and Monetary Reward at Work in VR,” lists Isabelle Cuber, Tarek Alakmeh, Mary Hausfeld, Anand van Zelderen, Moritz Jenny, Jochen Menges and Thomas Fritz as authors. The university said it was to be presented at NordiCHI ’26 in Vaasa, Finland, on October 5–7, and published in the Proceedings of the ACM on Human-Computer Interaction. The paper’s DOI is 10.1145/3829807.3829910.
Free tools Windows power users keep installed
One-click scans. No signup required.
What does “paid less” mean here?
The assistants did not earn salaries. People made decisions about how to divide money after interacting with them. The 10.25% figure describes the reported difference in those allocations to two differently presented agents in this particular experiment; it is not a statistic about AI workers’ pay, nor a human gender pay-gap estimate.
The announcement says the experiment also included a text chatbot and a desk robot, but it does not provide comparable reward results for those assistants. It also says participants reported no preference for male or female attributes even though their reward behavior differed. That summary does not establish why the allocations differed.
Could perceived humanness explain the difference?
The university announcement says Johan was perceived as more human-like than Johanna. That may matter when interpreting the reward result, but the announcement alone does not show that perceived humanness caused the difference. It also does not provide enough detail to assess the study’s full methods, subgroup results, or statistical uncertainty. The result should therefore be read as a finding from one experiment, not as proof that people will reward female-presenting AI agents less in every setting.
Is this the same as AI’s effect on the human gender pay gap?
No. A separate 2026 study, “Pre-AI Sorting, Post-AI Inequality: Generative AI and the Gender Wage Gap,” examines how women and men were distributed across occupations predicted to be exposed to generative AI, then models hypothetical wage changes. It uses 2021 Swedish administrative wage and occupation data. Its authors describe the calculations as mechanical partial-equilibrium simulations—not causal estimates or predictions of actual wage outcomes. Those questions and methods are distinct from an experiment measuring how participants allocated money to AI assistants.
The distinction is useful: one study concerns human choices about rewards for AI personas; the other models possible labor-market effects on people. Neither result should be treated as direct evidence for the other.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.How far can the finding be generalized?
The reported comparison comes from 189 participants in a virtual-reality office task setting. The university announcement is the source for the reported result; the full paper’s detailed methods and results are not established by that announcement. More evidence would be needed to know whether the same pattern appears in other tasks, interfaces, groups of users, or real workplaces.
As Dr. Mary Hausfeld, an assistant professor at the University of Limerick’s Kemmy Business School, put it: “What is striking about our findings is that the technology behind these AI agents was exactly the same, but people did not treat them in the same way.”
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
