Female AI assistants were paid less money than male ones despite identical skills. What explains this gender pay gap?

In a virtual workplace experiment that was conducted recently, female-presenting artificial intelligence (AI) assistants were given 10.25% less money than their male counterparts for performing the same tasks. The difference was there even though both assistants had exactly same programming and capabilities. This naturally raised the question of how deeply gender stereotypes can influence people’s decisions, even when they interact with machines.

The study, by researchers from the University of Limerick, the University of Zurich and SKEMA Business School, looked at whether the presence of AI assistants influenced how people valued their work. The findings imply that technology in the workplace does not necessarily remove the existing biases in human workplaces.

How scientists tested AI for gender bias

In the experiment, 189 participants were placed in a virtual reality office and asked to complete work tasks with different AI assistants. These tasks included a text-based chatbot, a desk robot and two human-like assistants named Johan and Johanna.


Johan was made to look like a man, Johanna had a female look.The two humanoid assistants were using the same technology and capabilities.

Upon completion of the tasks, participants were paid real money, which could be shared with the AI assistants as a form of compensation for the work.This helped researchers to assess whether an assistant's appearance had an impact on the perceived value of its contribution by the participants.

The results showed a clear difference as Johanna received 10.25% less money than Johan on average, despite the two assistants performing the same tasks. The study also found that participants perceived the male-presenting assistant as more human-like.

Not just limited to artificial intelligence

The experiment points to a problem that has long affected human workplaces also that people may judge similar work differently depending on whether they relate it with a man or a woman.

According to the International Labour Organization (ILO), women worldwide still earn about 20% less than men on average. Its analysis of the gender pay gap identifies several factors that contribute to the disparity, including women's under-representation in leadership, differences in working hours, career breaks, occupational segregation and discrimination.

The ILO also notes that jobs dominated by women tend to be undervalued in wage-setting. There are measurable differences in work and employment patterns that explain part of the pay gap, but not all of it.

The AI experiment presents a different setting in which unequal treatment was seen. Unlike comparisons between human employees, the two assistants had the same technical capabilities and performed identical tasks. Differences in their skills therefore could not explain the unequal rewards.

Human-like AI designs reinforce stereotypes?

The researchers found that participants generally preferred interacting with more human-like assistants. However, the difference in how Johan and Johanna were perceived seems to point out that making AI more human-like can also introduce other considerations in interactions.

Speaking about the research, Dr Mary Hausfeld of the University of Limerick's Kemmy Business School highlighted the importance of how AI agents are presented to users.

“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 said.

She also warned against allowing the design of AI assistants to reproduce existing inequalities as these systems become more common at work.

The study reported that some participants said gender had not influenced their decisions, even as the female-presenting assistant received less money. This highlights a possible gap between people's stated attitudes and their actual choices.

These results do not mean directly that all gendered AI designs will result in unequal treatment, and they do not also imply that the same pay gap will be seen in real workplaces.