A virtual-reality study suggests that people may value identical AI assistants differently when the systems are presented with masculine or feminine characteristics.
Same AI, different perception
Researchers from the University of Zurich, the University of Limerick and SKEMA Business School invited 189 knowledge workers into a simulated office environment. Participants collaborated with AI tools designed to support ideas around workplace wellbeing, inclusion and teamwork.
Among the tools were two human-like AI avatars: Johan, presented with a masculine voice and appearance, and Johanna, presented with feminine traits. Both assistants used the same underlying version of the GPT-4 language model and completed the same role.
After the sessions, participants evaluated each assistant's contribution, trustworthiness and human-like qualities. They could also allocate up to four Swiss francs as a reward for the AI's work, while retaining any unallocated amount.
Design details influence decisions
Johan received an average reward of 2.96 Swiss francs, compared with 2.54 francs for Johanna. The masculine-presenting avatar was also rated as more human-like on a seven-point scale.
The difference appeared despite participants interacting with AI systems that shared the same core technology. In follow-up interviews, most respondents said they believed an assistant's gender presentation should not influence trust or compensation.
The findings highlight a key dimension of human-AI interaction: people may carry familiar social assumptions into digital environments, even when they consciously aim to make neutral decisions.
A more thoughtful path for AI design
The researchers note that the experiment was conducted with a limited sample in a controlled virtual setting, rather than in real-world employment conditions. Voice, appearance and other design choices may also have shaped participants' impressions.
Published in the Proceedings of the 14th Nordic Conference on Human-Computer Interaction, the research offers a useful perspective for teams building AI agents for workplaces, education and customer services. Designing assistants with awareness of social cues could help create more balanced and inclusive digital experiences.
As AI becomes a more visible collaborator in daily life, understanding how people respond to its design may guide the development of systems that are trusted for their capabilities rather than their presentation.
