Johan and Johanna, two AI assistants, did the same work on the same underlying technology. One was presented as a man, the other as a woman. People in a virtual reality office paid Johanna about 10% less than Johan, according to Forbes' report on a University of Limerick study.
The experiment involved 189 people who worked with the assistants and then split real money between themselves and the AI. The report, published on 5 October, put the gap at 10.25%. The research itself appeared three days earlier, in the proceedings of a Nordic human-computer interaction conference.
An Identical Assistant Was Valued Differently
The two assistants were built to differ only in how they looked and sounded, while the software behind them was meant to be the same. The result tested how people respond to presentation, not how well the AI performs.
The researchers' paper notes that AI assistants are increasingly described as digital coworkers rather than tools. They say gender cues and human-like presentation can activate the social expectations people already rely on when dealing with each other.
'We often think of AI as being neutral,' co-author Dr Mary Hausfeld said in the report, 'but the way we design and present these systems can activate those same assumptions and biases that exist in our interactions with other people'.
The Same AI Did the Same Work for Less Pay
Participants worked in a virtual office set up like an ordinary desk job, completing tasks with different AI assistants, including a male-presenting one and a female-presenting one. Each person worked with only one of the two.
After each task, participants could give the assistant part of a small cash reward and keep the rest, so paying the assistant cost them real money. Johanna received less than Johan, and participants also rated Johan as more human-like.
Human-like assistants were also paid more than a desk robot or a text-based chatbot. That points to a link between how human an assistant seems and how much people pay it.
Most Said Gender Wouldn't Matter, but the Payments Differed
Researchers interviewed 34 of the participants. Of those, 30 said they would trust and reward an assistant the same regardless of its gender. Only three reflected on whether bias might affect their own decisions.
The payments told a different story, with participants of both genders rewarding Johanna less than Johan. Because nobody worked with both assistants, the gap shows up across the groups, not within any one person.
The authors call this a blind spot. Participants stated a fairness principle but underestimated how far presentation shaped their judgements in the moment.
One Experiment Points to a Pattern It Cannot Yet Prove
The lower human-likeness rating complicates the picture, because the study cannot separate it from gender. Each gender was also represented by a single avatar and voice, so the result may reflect those specific designs.
The money was not a wage, either. Participants gave it towards the assistant's further development, so the authors read it as a sign of the value people placed on the assistant. The test also lasted about 30 minutes, so it may reflect first impressions more than long working relationships.
Gender in AI design has been questioned before. In 2021, UNESCO was already discussing gendered voice assistants in an article titled 'AI-enabled Voice Assistants: No longer female by default'. That article does not explain this result, but it shows the question is not new.
This University of Limerick experiment adds one data point: two assistants on the same model, paid about 10% apart.
