In the AI era, machines are making rapid progress, but humans? Well, maybe not so much. A new study from Ireland’s University of Limerick found that people paid female-presenting AI agents 10.25% less than male-presenting agents, yet both used identical technology and performed the same work.
The findings indicate that gender bias can influence how employees evaluate AI systems, and that agents’ human-like characteristics affected how much users trusted and rewarded them.
The study, titled “Human-Like and Male? How AI Assistant Design Relates to Trust and Monetary Reward at Work in VR,” involved 189 knowledge workers who completed tasks in a virtual reality office.
Researchers tested three AI assistant designs: a text-based chatbot, a robotic desk assistant and a human-like agent. The human-like version was presented as either Johan, a male agent, or Johanna, a female agent. All three designs used the same underlying language model.
Participants evaluated the assistants’ performance and human-like qualities. They also allocated monetary rewards of up to $4 per assistant, providing researchers with a measure of how perceptions affected actual payment decisions.
The results showed that Johan received significantly higher compensation than Johanna. Participants also rated Johan as more human-like, even though the two agents had identical capabilities.
AI Design Influences Workplace Decisions
Human-like assistants generally received higher trust ratings and greater monetary rewards than the robotic assistant. Companies using agents may gain greater employee engagement by giving AI agents human characteristics.
But doing so creates other issues. An AI agent’s appearance can influence workplace decisions independently of its technical performance, the study suggests. In particular, gendered designs can prompt assumptions that employees may not recognize in their own behavior.
The researchers also interviewed 34 participants, with 30 saying they would evaluate and reward AI assistants equally regardless of gender. Yet the experiment’s overall results showed a clear pay disparity between male- and female-presenting agents.
The interviews uncovered another contrast: some participants preferred female-presenting assistants, even though the female agent received lower compensation in the experiment.
The findings suggest that asking employees whether they hold gender biases may not accurately predict how they will interact with AI coworkers.
Implications for Enterprise Use
The research raises questions about how AI agents should be designed and introduced into business workflows.
Organizations are now using AI assistants for everything from document creation to workflow automation. As agents work on these tasks more collaboratively, employees will face decisions about how much to trust their recommendations and which tasks to delegate.
The University of Limerick study suggests that these decisions may be influenced by an agent’s presentation rather than its actual capabilities.
One possible approach is to use gender-neutral or less human-like interfaces. However, the research does not establish that either design would eliminate bias. Less human-like assistants received lower trust ratings and rewards, highlighting a potential tradeoff.
The study, to be sure, had limitations. It examined behavior in a controlled virtual reality setting, not a business setting involving extended worker relationships with AI systems. Its findings about pay levels reflect experimental rewards rather than actual salaries.
Still, the results highlight the fact that humans have biases, and creating a technically capable agent does not guarantee that employees will evaluate it objectively.
