We treat machines differently from people, and that makes perfect sense. Machines, after all, are tools: they aren’t sentient and conscious beings. But what about machines that are given human-like features – for example, gender? A new study, co-authored by Mi3 associated researcher Jurgis Karpus together with colleagues from Trinity College Dublin, finds that gender stereotypes don’t just affect how we deal with other humans – they also carry over into how we behave toward AI. Here, Jurgis explains more.
Cooperation is a cornerstone of social life. From everyday traffic to tackling global challenges such as climate change or pandemics, we rely on others to work together for mutual benefit. Yet cooperation is fragile. It can easily break down because it requires us to balance our own interests against those of the group – and to trust that others will do the same.
Previous research, including some of our own, has shown that people tend to cooperate less with artificial agents than with other humans. This is not because people distrust machines, but because they are more likely to prioritise their own interests when interacting with machines. At the same time, it is often suggested that giving AI human-like features might encourage more human-like behaviour toward it. In our new study, we asked what happens when one such feature – gender – is assigned to AI.
At first glance, the idea of machine gender may seem odd. What could it mean for an AI to have a gender? And yet, many navigation systems, chatbots, and voiced smart assistants are clearly presented as male, female, or something in between. We wanted to know how this affects people’s willingness to cooperate with artificial agents.
To find out, we ran an experiment with 402 participants who played a version of the Prisoner’s Dilemma game – a classic setup used to study cooperation. In each round, participants interacted with a different partner, labelled either as a human or an AI-powered bot. Each partner was also assigned a gender: male, female, non-binary, or no gender.
The results were both familiar and striking. People cooperated with gendered AI bots almost as much as they did with human partners. However, their motivations differed. When interacting with bots, participants were more likely to act opportunistically – choosing not to cooperate in order to maximise their own gains rather than because they distrusted their partner. This pattern mirrors earlier findings about human–AI interactions.
More striking were the effects of gender. Participants cooperated most with female partners and least with male ones. Lower cooperation with males was largely driven by lower trust that males would cooperate in return. When participants chose not to cooperate, their motive to exploit their partner for selfish gain was most pronounced when they interacted with females. Crucially, these gender biases appeared regardless of whether the partner was human or a bot.
The central takeaway is a dilemma. Adding human-like features such as gender to AI systems can increase people’s willingness to cooperate with them. But it also means that existing human biases – here, gender stereotypes – are transferred to our treatment of machines.
Anthropomorphising AI may increase people’s willingness to cooperate with it, but it risks reinforcing the very social biases many of us are trying to overcome.
Check out the original article and Jurgis Karpus’ profile and LinkedIn.
Reference:
Bazazi, S., Karpus, J., and Yasseri, T. (2025). AI’s assigned gender affects human-AI cooperation. iScience 28, 113905. https://doi.org/10.1016/j.isci.2025.113905.