A Harvard study highlights a strange gap in how men and women describe their knowledge of artificial intelligence. Men use AI more often and answer factual questions about it more accurately on average, but they also show more confidence when they are wrong.
More than half of the men surveyed said they recognized AI leaders who do not exist. The results suggest that knowing the names surrounding AI can become a performance of confidence, even when the underlying knowledge is incomplete.
Confidence does not always match knowledge
Harvard researchers asked respondents whether they knew several people connected with AI. Among male respondents, 73 percent said they knew who Sam Altman was. Altman is the OpenAI CEO.
A little over half, or 53 percent, said they knew of Mira Murati, Susan Milner, and Craig Dalton. Murati is a real person: she is the former CTO of OpenAI and the current CEO of Thinking Machine Labs. Milner and Dalton do not exist.
The answers from women showed a different pattern. Forty percent said they knew who Milner and Dalton were, while 38 percent said they recognized Murati. That gap matters because the fictional names were presented alongside a real AI leader, making confidence easy to mistake for familiarity.
“Sometimes, that veers into overconfidence—like claiming to recognize fictional AI executives,” Laura Manley, executive director of Harvard Kennedy School’s Shorenstein Center, said.
The broader survey results add an important wrinkle. Men are more overconfident in general when surveyed, yet they also use AI more frequently and are, on average, better able to answer factual questions about AI correctly. Men and women answer questions incorrectly at around the same rate.
The clearest difference comes in how respondents handle uncertainty. Women said they were unsure of an answer for an average of 3.3 questions out of 12, compared with 2.3 questions for men. Manley’s advice is simple: “We have to be able to be able to say ‘not sure’ more often.”
Who gets seen as an AI expert?
The survey results sit within a larger pattern in AI coverage. Across a year of media coverage, men appeared in headlines about AI more than 14 times as often as women. Three-quarters of experts quoted in that coverage were men, even though women made up roughly 29 percent of the AI talent pool.
That imbalance can shape who audiences remember as an authority. When men dominate headlines and expert quotes, people may encounter fewer women as visible leaders, researchers, and voices on the technology. The result is a public picture of AI that does not match the share of women in the AI talent pool.
Women in tech have a lot to say about their experiences, including automating their days with AI and confronting bro culture. Those experiences add something that a list of famous names cannot: a view of how AI is used in daily work and how workplace culture affects who feels heard.
The emotional responses in the survey also split along gender lines. Both men and women rated “concern” as their top emotional response to AI, but female respondents were more concerned, fearful, and worried than male respondents.
Men rated every positive feeling higher than women did, including curiosity, enthusiasm, and excitement. The contrast points to different experiences of the same technology: men reported more positive feelings and greater confidence, while women reported more concern and more uncertainty.
Why admitting uncertainty matters
AI knowledge is not only about recognizing executives or answering factual questions. The technology also raises questions about safety and misuse. AI models can be easily modified to assist terrorist and extremist activity, which makes clear judgment more important than sounding certain.
That is where the Harvard findings reach beyond a simple comparison between men and women. A person who admits uncertainty leaves room to check an answer, correct a mistake, or ask for more information. A person who claims to recognize fictional leaders does the opposite, presenting confidence without reliable knowledge.
The study does not show that men know nothing about AI. In fact, men answered factual questions more accurately on average and used AI more frequently. It shows that those strengths can exist alongside overconfidence, especially when people are asked to identify names or describe what they know.
The findings also do not treat women’s uncertainty as a lack of ability. Women and men answered questions incorrectly at around the same rate, even though women were more willing to say they were unsure. That distinction matters because uncertainty can be an honest account of knowledge, not evidence that someone has less of it.
As AI becomes a bigger part of work and public debate, the people who speak with confidence will continue to draw attention. The Harvard survey offers a useful test for that confidence: Can the person separate real knowledge from familiarity, acknowledge an unknown answer, and take AI’s risks as seriously as its promise?
