Published on
A new study on how people interact with AI agents in the workplace has found that those with a female appearance or presenting as female would get paid less than their male-presenting counterparts for the same work, revealing that gender can influence interactions with artificial intelligence.
The findings come just as AI agents are entering the mainstream, with almost 79% of companies reporting the use of AI agents this year and a further 40% of applications using them too, according to IT consulting firm Accelirate.
This means that if you are not already working with an AI agent, you probably will be soon.
The study, conducted by the University of Limerick and called ‘Human-Like and Male? How AI Assistant Design Relates to Trust and Monetary Reward at Work in VR’ aimed to see how humans interact with AI agent co-workers.
Researchers placed 189 human workers in a virtual reality office and tasked them with completing workplace tasks with four different AI assistants.
Among their options were a text-based chatbot, a desk robot, and two human-like agents presented with male and female appearances: Johan and Johanna.
All the assistants had the same background functionality and used the same underlying model, OpenAI’s gpt-4-1106-preview.
Human participants were then asked to rate the AI agents’ work and given real money to pay them for completing tasks, with the aim of recreating a real-world scenario to see how much people were willing to pay AI agents for their work.
Overall, participants rated both human-like agents Johan and Johanna similarly for their work, with Johan trusted marginally more but not by any statistically significant margin.
However, both men and women paid the female-presenting AI agent Johanna significantly less than Johan (10%), despite her having the same capabilities. Johan was also seen as more human-like than Johanna.
This is despite most participants telling researchers in interviews that they would not treat male and female-presenting AI agents differently.
Fortunately for Johan and Johanna, they were both paid more for their work than the other AI agents, with women in particular scoring and rewarding the human-presenting agents more than men did.
Perhaps unsurprisingly, participants also saw both human-presenting AI agents as much more human-like than the chatbot and robot.
Not only that, workers trusted and credited Johan and Johanna more than the other AI agents, with women in particular trusting them more than men.
“Having Johanna next to me made me more prone to credit her more again because I felt there was a person, a colleague sitting next to me that I had a conversation with and that was giving me her own input,” said one participant.
The study also found that people preferred AI agents that were either human-presenting or clearly non-human presenting, rejecting the robot design that appeared to be “in between” the two options.
Outside of the gender dynamics, those who took part in the study said that “AI assistants should integrate seamlessly into every knowledge work to be useful”.