Have you ever thought about whether gender bias exists not just in human interactions but also in the dynamics of human-AI interaction?
Cooperation is the process in which more than one individual works together to achieve common or complementary objectives. It is generally referred to as a way to maximize collective gains and reduce costs, which emphasizes the significance of collaboration in obtaining results that might not be possible alone.
From small to large-scale social matters, cooperation is fundamental. Yet, both individual and group interests contradict each other. Cooperation inherently necessitates people to leave their own personal goals behind and believe other people do not take advantage of their efforts. Despite barriers to cooperation like prioritizing self-interests first and favoritism—supporting one’s own group over other people, people prefer to work together because cooperation enables them to get mutually favorable results.
Cooperation with AI
As the notion of artificial intelligence has been gaining more popularity in time, the human-AI relationship, especially human cooperation with AI, has become one of the key topics in this field. Studies indicate that people are less keen to cooperate with AI than other humans, even though cultural and individual differences, such as openness to new technologies, affect this tendency. Also, people have a greater tendency to exploit AI for their selfish interests than they do to exploit other people.
It has been suggested that giving AI human-like characteristics could have a positive effect on people’s willingness to cooperate. Of course, there are cultural differences in this regard. Assigning a gender to an AI has been used for systems such as intelligent assistants or GPS navigators, in addition to emotions, voices, and appearances. There are many studies in the literature showing that the gender assigned to an AI affects people’s behavior. The desire to make a financial donation is one instance. Furthermore, compared to male AI, individuals believe that female AI is more human-like.
These findings shed light on many behavioral tendencies towards artificial intelligence. However, the impact of gender on human cooperation with AI in circumstances where individual and group goals conflict remains unknown.
To address this unclarity, Bazazi et al. (2023) conducted a study named “AI’s assigned gender affects human-AI cooperation,” exploring two essential subjects below:
- Do human-to-human and human-to-gendered AI agent cooperation share similarities?
- Do pre-existing gender biases also have an impact on human-AI interaction?
To study these phenomena, researchers employed the Prisoner’s Dilemma, which is a popular economic game. It is particularly used for investigating decision-making mechanisms under circumstances where personal and collective goals contradict each other. In the Prisoner’s Dilemma game, two individuals separately make decisions about whether to cooperate or defect, causing four different results.
- Mutual Cooperation (MC): Both players cooperate, achieving the best collective outcome.
- Mutual Defection (MD): Both defect, resulting in a worse result for both than mutual cooperation.
- Exploitation (E): One player defects while the other agrees to cooperate, the defector taking advantage at the cooperator’s expense.
- Unconditional Cooperation (IC): One player cooperates with another in spite of expecting the other player to defect.
In the research, the Prisoner’s Dilemma game was utilized to examine cooperation and defection behaviors in human-to-human and human-to-gender-labeled AI interactions. Participants preferred to either cooperate or defect and were asked to guess their partner’s choice.
Key findings include:
- In terms of cooperation, participants preferred to cooperate with women the most (58.6%) and with men the least (39.7%), whether partners were human or artificial intelligence.
- When motives for defection are examined, exploitation (self gain expense of the partners) was more prevalent to female AI partners. Moreover, there is a greater mistrust towards male humans, leading to defection.
- Regarding gender biases, gender labels considerably impact preferences to either defect or cooperate, highlighting that biases seen in human-to-human interactions also exist in human-AI interactions.
Takeaways
This study demonstrates the fact that gender substantially affects decisions of cooperation in both human-to-human and human-to-gendered-AI interactions. While people are less likely to cooperate with AI than humans, gender biases are still widespread. These results highlight the necessity of proper AI regulation and design in order to mitigate biases and enhance human-AI cooperation.
