Two AI assistants, built on identical technology, completed identical work. The one presenting as female was paid 10.25 percent less.
That is the central finding of new research from the University of Limerick in Ireland. According to the researchers, the gender bias familiar from human workplaces carries straight over to the way people treat AI models, and female-presenting chatbots were paid less in their experiment.
The study appears in the journal Proceedings of the ACM on Human-Computer Interaction, and the researchers will also present it at an upcoming conference.
A raise for Johan, but not for Johanna
The research team placed 189 participants in a virtual reality office, where they carried out work-related tasks alongside several AI assistants. Some of the assistants presented as male and others as female. To give the experiment real stakes, participants received actual money and had to divide it between themselves and the AI assistants.
Johan, the male AI, finished well ahead of Johanna, his female counterpart. Although she did the same work, Johanna received 10.25 percent less. The study also found that participants viewed her as less human-like than Johan.
“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,” said Mary Hausfeld, an assistant professor at UL’s Kemmy Business School and a coauthor of the study.
A face won trust, and a bigger payout
Gender was only one of the factors under study. The researchers also wanted to learn how the way an AI is presented, or embodied, shapes how people perceive it and work alongside it. This question is becoming more pressing as these tools are increasingly pitched as “digital coworkers.”
Three representations were tested in the virtual office: a traditional chatbot, a desktop robot with no face and no movement, and a male or female avatar seated beside the participant.
The human-like assistants earned more trust than either the chatbot or the robot, and their pay reflected it. Participants handed the avatars more money and gave them more credit for their portion of the work.
Participants said it made no difference
The oddest result came from the participants themselves, who denied that the human-like AIs had swayed them. In interviews, many said they favored assistants that did not look human and that the gender of the AI assistant had no effect on them.
How they split the money tells another story. This mismatch between what people claimed and how they actually behaved is the study’s most valuable insight, because it shows the problem can’t be solved simply by asking users whether they are biased.
“We often think of AI as being neutral, but the way we design and present these systems can activate those same assumptions and biases that exist in our interactions with other people,” Hausfeld said. “We don’t want to inadvertently reproduce existing inequalities in a new technological setting.”
Avatar design carries real consequences
According to Hausfeld, companies developing AI agents must take far more care in how those agents are presented. How an agent looks, or how it is embodied, should not be considered a purely aesthetic decision.
No one pays a chatbot an actual salary, so the money in the experiment serves as a proxy. Still, that proxy captures something genuine: how much people trust a system and how much credit they assign it for its work. If a voice, a name or an avatar can shift those figures by 10 percent without anyone realizing it, then the team choosing that voice is also making a decision about trust, intentionally or not.
“Such choices may affect how much people trust AI, how they judge its contribution, and even how they financially reward it,” Hausfeld said.