Virtual Employee Performs Same Job... But Earns Less

One hundred and eighty-eight people sat in a virtual office wearing virtual reality headsets and completed work tasks with the assistance of four AI-powered assistants: a text-based chatbot, a desktop robot, and two virtual human avatars—a man named “Johan” and a woman named “Johanna.”

The four assistants operated on the same language model developed by OpenAI. However, when participants rewarded each assistant with real money, Johanna received approximately 10% less than Johan, according to a study conducted by Ireland’s University of Limerick.

Both male and female participants behaved similarly: they rated the quality of the two avatars’ work with comparable scores and expressed slightly higher trust in Johan, although the difference was not statistically significant. Johan appeared more human-like to them than Johanna did.

The two human-like assistants garnered significantly higher trust and rewards than the chatbot and the desktop robot. Participants preferred either clearly human or clearly mechanical designs over the robot that fell somewhere in between.