IT Home On October 26, researchers at Limerick University in Ireland pointed out in a research paper published earlier this month that, in an experiment involving 189 knowledge workers, humans prefer anthropomorphic AI agents, and there is also a “gender” bias.
The tester works together with 3 AI colleagues based on the same underlying model in a virtual office. The three virtual colleagues appear through a text chat window, a virtual little robot, and a human-like avatar, with the avatar having two versions: a female “Johanna” and a male “Johan”. After the AI colleagues collaborate at work, the tester rates each AI assistant and allocates subsequent development resources based on their performance.
The results show that anthropomorphic images received higher trust, credit, and “compensation” than those that resemble robots more closely; there is no significant difference in trust and credit between “Johanna” and “Johan”. However, the feminine “Johanna” still received statistically significantly lower “compensation” than her male “colleagues”, although most of the participants interviewed did not believe that the virtual “gender” would affect their feedback.
Other notable situations include: the text chat window and the virtual robot are comparable in trust and credibility scores, the human-like rating of “Johann” is significantly lower than that of “Johan”, and the test subjects prefer “Johann” as an AI assistant in terms of appearance.
Reference