Study Shows Humans Prefer Anthropomorphic AI Agents but Exhibit Gender Bias

A recent study by the University of Limerick involving 189 knowledge workers revealed that humans prefer anthropomorphic AI agents over text or robot-shaped ones. However, the study also uncovered a gender bias, where the female-coded AI assistant received significantly lower "compensation" than its male counterpart despite identical performance. This research highlights persistent human biases in human-computer interaction, showing that societal prejudices can easily transfer into how people evaluate and reward AI systems. As anthropomorphic AI agents become more prevalent in workplaces, addressing these embedded biases is crucial for fair system design and deployment. In the experiment, 189 participants collaborated with three virtual colleagues powered by the same underlying model, presented as text, a robot, or male/female human avatars. While the female avatar "Johanna" and male avatar "Johan" scored similarly in trust, Johanna received statistically lower resource allocation, even though most participants claimed gender did not influence their feedback.