Harvard Study: Men More Likely to Claim AI Expertise, Even Knowing People Who Don't Exist

A Harvard University survey of more than 5,400 Americans has revealed a fascinating phenomenon in AI perception: men not only talk about AI more, but are also more likely to claim they know people who don't exist at all.

Study Design and Core Findings

Harvard researchers presented respondents with a list of names that included several fictitious individuals. The results showed that 73% of men claimed to know OpenAI CEO Sam Altman, while 53% of men claimed to know people who don't exist. A similar trend appeared among women—40% said they knew fictitious figures, and 38% knew Mira Murati (former OpenAI CTO, now head of Thinking Machine Labs), compared to 53% of men.

These numbers point to a clear pattern: when it comes to AI topics, men are more inclined to adopt an "I know" posture, even when that knowledge is fabricated.

Overclaiming is a classic paradigm in psychology for measuring the accuracy of self-assessed knowledge. The basic logic is this: mix several completely fictitious names into a list of real items, and if respondents claim to know these non-existent people or concepts, it indicates a systematic inflation of their self-reported knowledge. Since fictitious items cannot possibly be genuinely known, any affirmative response can be treated as a reliable signal of inflated confidence. The paradigm was developed by Delroy Paulhus and colleagues in the 1990s to study self-deception and social desirability bias, and has since been widely applied to financial literacy, climate change, health knowledge, and other domains. This Harvard study transfers it to the AI topic, and its cleverness lies in this: AI is currently in a period of rapid public perception-building, where many people are not actually familiar with industry figures but feel social pressure to "pretend they know the field." Fictitious names thus become a mirror that quantifies the tendency to avoid admitting ignorance.

The Mismatch Between Confidence and Knowledge

The study also uncovered a subtle divergence. Men use AI more frequently and score higher on knowledge-based questions, but the error rates for both genders are actually equal. The real difference lies in attitudes toward uncertainty—women on average admitted uncertainty on 3.3 questions, while men expressed hesitation on only 2.3.

In other words, men don't know more—they're just more reluctant to admit what they don't know. This overconfidence may stem from societal stereotypes that men "should understand technology," and may in turn reinforce those expectations.

Gender Imbalance in Media Coverage

This cognitive bias finds an echo in media reporting. According to the study, over an entire year of coverage, men were mentioned in headlines more than 14 times as often as women. Three-quarters of experts invited to comment on AI topics in media were men, while women actually make up about 29% of AI professionals.

There is a significant gap between the "AI expert" image presented by media and the actual demographic composition. When the public repeatedly sees male faces crowned with AI authority, this impression gradually solidifies into a default assumption that "AI is a male domain."

The gendered "expert pool" phenomenon is often referred to in communication studies as the "Matthew effect of authoritative citation." Editors under deadline pressure tend to call on existing contacts or lists, and traditional tech reporting networks have long been male-dominated, creating a self-reinforcing cycle: men are repeatedly cited, accumulating visibility, and that visibility makes them "safer" choices for future citations. Research shows that women make up about 29% of AI professionals, but their citation rate is far below that, indicating the gap stems not from a shortage of talent but from the inertia of sourcing habits. This imbalance in turn shapes public perception: the more monolithic the "AI faces" viewers see, the more easily they associate technical authority with masculinity, affecting girls' career imaginations and media's subsequent selection criteria.

Polarized Emotional Responses

On the emotional level regarding AI, both genders list "concern" as their primary reaction. But women exhibit more anxiety and fear, while men score higher on positive emotions like curiosity and excitement.

This emotional divergence may be causally linked to the confidence gap described earlier: the more confident someone is, the more likely they are to approach new technology with curiosity rather than fear; those lacking a voice are more likely to feel uneasy.

Gender differences in technology anxiety can be explained by "perceived control" theory. When facing AI, if a person believes they lack influence over the direction of technology, they are more likely to experience fear and avoidance; conversely, if social roles have given them the expectation of "understanding technology," they are more likely to experience uncertainty as challenge and curiosity. Women's underrepresentation in public AI discussions reduces their perceived voice, amplifying anxiety. At the same time, emotional responses themselves affect participation: anxious people speak up less, while curious people enter discussions more often, further tilting voices on media and social platforms toward the confident—creating a filtering mechanism that compounds with cognitive overconfidence.

What This Means

The value of this study is not in proving that "men understand AI better" or "women fear AI more," but in revealing a deeper issue: public discussion about AI may be largely dominated by overconfident voices.

When 53% of men are willing to claim they know people who don't exist, while women are more inclined to say "I'm not sure," how much of the "expert opinion" we hear in media is genuine expertise, and how much is merely confident expression?

Gender equality in AI may not just be about the proportion of professionals, but about who has the authority to define what it means to "understand AI."

One-line Summary

Harvard survey finds men are more likely to feign understanding of AI, and this confidence is dominating public discussion.

Paragraph Summary

A Harvard survey of over 5,400 Americans reveals significant overconfidence among men on AI topics: 73% claim to know Sam Altman, 53% claim to know fictitious figures, while women are more willing to admit uncertainty. Men use AI slightly more and score marginally higher on knowledge questions, but error rates are equal to women's—the difference lies in attitudes toward the unknown. Media coverage amplifies this bias further, with men mentioned in headlines 14 times as often as women and 75% of AI commentators being male. Emotionally, both genders list concern as their primary reaction, but women are more anxious and fearful while men are more curious and excited. The study raises a deeper question: AI public discourse may be dominated by overconfident voices, and gender equality is not just about workforce proportions but about who gets to define what it means to "understand AI."