Women think artificial intelligence is riskier than men, research shows

Artificial intelligence (AI) is changing the nature of how people work; However, people have different opinions about the risk and reward of using AI technology. According to a recent study conducted by Beatrice Magistro and her co-authors at Northeastern University, women are more likely than men to view AI (particularly generative AI) as risky. Using large survey samples from the US and Canada, this study also identifies risk exposure and tolerance as potential factors contributing to this gender gap.

Magistro and her team sought to determine why women tend to be more cautious than men when it comes to new technologies. Previous studies on automation indicated that a significant percentage of respondents (about half) expressed caution or concern about new technologies. This research argues that a person’s exposure to risk and tolerance for risk are essential components for understanding their feelings about the uncertainty of using new technologies.

As AI is adopted by more and more organizations, the potential benefits of AI, such as improved productivity and increased intelligence, are outweighed by significant risks, including job losses, data breaches and bias in the decision-making process. The consequences of these problems will not affect all people in the same way.

Reasons why women experience a different perspective

Women make up a large proportion of clerical, administrative and service industries and are generally seen as one of the most vulnerable to automation. Additionally, women are underrepresented in STEM (science, technology, engineering and math) careers. Consequently, this imbalance creates barriers for women in accessing jobs and leadership positions in AI technology.

Previous studies have also shown that algorithmic biases negatively impact women in hiring, lending and health care. Both patterns identified in the study (exposure and orientation) play a role in how women perceive the benefits and harms of AI.

“These two variables are the most influential. Risk exposure determines the extent to which someone can experience negative consequences, such as being unemployed due to AI, while risk orientation indicates how comfortable someone is with ambiguity,” Magistro explains to The Brighter Side of News. “Statistically, women have a lower risk tolerance than men. When women have high risk exposure but low risk tolerance, they tend to be more skeptical of AI,” she continued.

Methodology for conducting the study

The analysis was conducted using data from 3,049 respondents in Canada and the United States in November 2023 via an opt-in YouGov panel survey. Respondents were asked to indicate the extent to which they agreed that the potential risks of generative AI outweigh its benefits to themselves and their communities on a scale of 0 to 10.

To understand attitudes towards risk, respondents were presented with a gamble: they were given the option of receiving €1,000 with certainty, versus a 0.50 chance of winning €2,000. If someone chose the first option, this indicated more risk-averse behavior.

Assessing an individual’s exposure to the risks of AI is complicated due to the unknown nature of AI’s long-term effects. However, the study used broad measures related to post-secondary education as a proxy measure, stating that higher levels of education correlate with greater adaptability to changes arising from AI. In addition, the authors assessed measures of AI risk exposure by position in additional analyses.

The gender gap remains consistent

In this study, there was a consistent discrepancy in the ratings that men and women assigned to AI risk. The average score for female participants when asked whether the risks outweighed the benefits was 4.87, compared to 4.38 for men. This 11 percent difference is about the same as the gender gap found on other key economic issues, such as trade.

Women reported significantly higher overall risk aversion. This resulted in more women choosing the guaranteed $1,000 payment instead of gambling. This pattern was consistent across educational levels.

Exposure was also a factor for workers who had not completed college. Between the two education levels, women experienced higher levels of exposure to technological disruption than men, largely due to the types of occupations women traditionally pursue. This result was also supported by additional measures used in the study.

When risk and exposure interact

When examining the relationship between risk and exposure, the results were of particular interest when risk orientation and exposure were considered together. Among individuals within the same study group who were willing to take the lottery gamble, men and women had similar perceptions of AI-related risks. Only among those who showed a risk-averse attitude was a clear gender gap observed.

The researchers also found that respondents’ education level played a major role in their perception of AI. Respondents with lower levels of education were more likely to believe that the risks of AI outweighed its potential benefits. Both men and women with higher levels of education perceived lower levels of risk associated with AI, although women still perceived higher risk than men with the same level of education.

Risk-averse men showed slightly greater sensitivity to their own exposure levels compared to risk-averse women. However, overall, risk-averse women perceived higher levels of risk associated with using AI in their workplace.

Test uncertainty immediately

To further explore the relationship between risk and uncertainty, the researchers conducted a pre-recorded survey experiment. In this experiment, participants read about generative AI being considered for development by an employer. Each individual was presented with a possible outcome, ranging from a 100 percent chance of a net job gain to a 30 percent chance.

As predicted, as uncertainty increased, support for AI adoption declined more for women than for men. The average level of support among women for a net job gain of 30 percent was 2.63 out of 5, compared to an average level of support of 2.98 out of 5 for men.

When the potential for net job growth rose to a guaranteed 100 percent, the gender gap disappeared. Based on the research, women’s level of support is therefore generally more dependent on the certainty of net job gains than that of men. When the benefits of AI are clear, women, on average, support AI almost as strongly as men.

What people say about AI

The survey also collected data on what respondents saw as the biggest risks and key benefits of artificial intelligence through open-ended survey questions. A large number of respondents indicated that they did not know what benefits AI would bring to them.

A greater share of women than men cited distrust or skepticism about the benefits of AI, while a greater share of men cited increased efficiency, productivity and innovation in medical and scientific fields as the key benefits.

Both men and women identified job loss, skill loss and financial inequality as primary categories of AI-related risks. However, fewer men than women focused on the impact of AI on job security and loss of human interaction, while more women than men raised concerns about malicious use of AI, disinformation and loss of control over the technology.

Practical applications of the research

The results of this study suggest that perceptions of artificial intelligence are shaped by personal experiences and the ability to deal with uncertainty. As a result, policymakers should exercise caution when developing AI-related policies, as a one-size-fits-all approach could further entrench existing inequalities in technology access and use.

To address skepticism and avoid negative reactions, government officials should focus on job security, transparency, and clearly demonstrating the benefits of AI implementation.

Furthermore, a more inclusive approach to AI research and development can lead to less biased systems and fairer access to the opportunities created by artificial intelligence.

Research results are available online in the journal PNAS Nexus.

Related stories

Do you like feel-good stories like this? Receive the newsletter from The Brighter Side of News.