University of Limerick study shows female AI agents received less pay
- Female AI assistants received 10.25% less pay than male avatars for completing identical tasks in a virtual-reality office study conducted by the University of Limerick.
- Researchers placed 189 human participants into a VR workplace simulation to test whether compensation varied based on the gendered appearance of digital assistants.
- During the study, human participants were given real money to divide between themselves and their AI assistants after finishing shared work tasks.
“We don’t want to inadvertently reproduce existing inequalities in a new technological setting.”
Female AI assistants received 10.25% less pay than male avatars for completing identical tasks in a virtual-reality office study conducted by the University of Limerick.
The University of Limerick VR Experiment
Researchers placed 189 human participants into a VR workplace simulation to test whether compensation varied based on the gendered appearance of digital assistants. Although the underlying technology, including large language models and guardrails, remained identical, participants consistently assigned higher monetary values to male-presenting agents than to female ones.
Johan Versus Johanna: Measuring the Disparity
During the study, human participants were given real money to divide between themselves and their AI assistants after finishing shared work tasks. The male assistant, Johan, received 10.25% more compensation than the female assistant, Johanna, for the exact same labor.
Dr. Mary Hausfeld explained that users generally assume AI is gender-neutral. However, the study demonstrated that participants did not treat the agents in the same way once distinct avatars were introduced, pointing to deeply ingrained human biases transferring directly into software environments.
Deconstructing Human Bias in Automated Tools
Researchers warned that assigning gendered avatars and personalities to digital assistants is not merely a decorative choice. Giving AI agentic workflows a gendered appearance triggers real-world stereotypes, which subsequently influences how humans value, compensate, and rely on automated tools.
Evaluating Characteristics for Future Integration
As digital assistants integrate further into professional settings, experts caution that organizations must carefully evaluate the characteristics assigned to these systems. The underlying LLM models, harnesses, and guardrails were entirely uniform, proving that the disparity stemmed solely from human perception of the avatars rather than technical output.