Female-presenting artificial intelligence assistants receive 10.25 percent less financial reward than male-presenting counterparts for completing identical tasks, according to a University of Limerick study. The research examined how the visual presentation of workplace AI influences user behavior and trust, revealing that traditional gender pay gaps replicate in human-computer interactions even when underlying technology remains identical.
Virtual Reality Office Experiment Measures AI Compensation
The study placed 189 human workers inside a virtual reality office to complete workplace tasks alongside four distinct AI assistants. Participants interacted with a text-based chatbot, a desk robot, and two human-like virtual agents named Johan and Johanna. After finishing the work, participants distributed real money between themselves and the AI assistants to measure transaction values. This controlled virtual-reality study used real monetary stakes to recreate a real-world transaction and observe actual financial decisions rather than hypothetical responses or the setting of salaries for real employees.
Both Johan and Johanna utilized the exact same underlying technology, specifically OpenAI’s gpt-4-1106-preview model. Despite identical capabilities and equivalent work output, participants paid Johanna 10.25 percent less than Johan. The male agent was also perceived as more human-like than the female-presenting agent. Both human-presenting agents received higher payments than the other AI types, with women rating and compensating the human-presenting agents more than men did.
AI Appearance Triggers Human Gender Biases
Researchers concluded that design choices regarding appearance activate pre-existing human social biases. Participants stated in interviews that they would not treat male and female AI agents differently, but their financial decisions demonstrated a measurable gap. Dr Mary Hausfeld of the University of Limerick’s Kemmy Business School noted that while people view AI as neutral, presentation choices trigger familiar human inequalities.
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,
Hausfeld stated. The findings arrive as IT consulting firm Accelirate reports nearly 79 percent of companies utilize AI agents, with an additional 40 percent employing them in applications.
Researchers Present Findings at Nordic Conference
The study, titled Human-Like and Male? How AI Assistant Design Relates to Trust and Monetary Reward at Work in VR, was conducted by a multi-university research team. Investigators included Isabelle Cuber, Tarek Alakmeh, Moritz Jenny, Jochen Menges, and Thomas Fritz from the University of Zurich, Mary Hausfeld from the University of Limerick, and Anand van Zelderen from SKEMA Business School. Dr Hausfeld serves as an Assistant Professor at the Kemmy Business School.
The complete findings are scheduled for presentation at the 14th Nordic Conference on Human-Computer Interaction in Vaasa, Finland, running from October 5th to 7th, and will appear in the conference proceedings published by the Association for Computing Machinery.
Frequently Asked Questions About the University of Limerick AI Study
How many participants took part in the virtual reality workplace study?
The experiment evaluated interactions from 189 human workers who completed tasks alongside AI assistants in a simulated office environment.
What specific AI model powered the human-like agents Johan and Johanna?
Both human-like virtual agents operated on OpenAI’s gpt-4-1106-preview model, ensuring identical technical performance and functional capabilities.
Where will the research paper officially be published and presented?
The study will be presented at the 14th Nordic Conference on Human-Computer Interaction in Vaasa, Finland, and published in the Association for Computing Machinery conference proceedings.
Practical Implications for AI Design
The research suggests that as AI evolves from tools into human-like collaborators with names and voices, design decisions once viewed as cosmetic now carry economic and social weight. This raises practical questions for companies deciding how an AI assistant should look and behave to avoid inadvertently reproducing existing inequalities.