Harvard Study Finds Men Overclaim AI Knowledge, Citing Nonexistent Experts
A new Harvard study found that a majority of men surveyed claimed familiarity with fabricated AI leaders, exposing a gap between perceived and actual expertise.
A Harvard study has found that more than half of the men it surveyed claimed to recognize AI industry leaders who do not exist. The research points to a measurable gap between how much people say they know about artificial intelligence and what they actually know, with men far more likely than women to overstate their familiarity with the field.
The design is a standard tool in survey research: plant invented names alongside real ones and see who claims to know them. In this case, the fake figures were presented as prominent AI leaders. Respondents who said they recognized the fictional names were, by definition, claiming knowledge they could not have had, which gives researchers a clean way to measure overclaiming without relying on self-assessment alone.
The pattern is not unique to AI. Earlier studies using the same technique have found similar results in fields such as finance and science, where confidence and social pressure reward visible expertise. What makes the AI result notable is timing: the technology has moved from research labs into everyday products in a few years, and a large share of the public is now expected to have opinions about models, benchmarks and the companies building them.
The gender split is the headline finding. Men were substantially more likely than women to say they recognized the invented names, even though neither group could have known them. That suggests the gap is not about access to information but about willingness to claim it, a distinction that matters for anyone trying to gauge public understanding of AI through surveys.
The result lands as employers, schools and policymakers try to measure AI literacy, often through self-reported surveys. If men systematically inflate what they know, those instruments will produce skewed pictures of workforce readiness and could misdirect training budgets toward people who already believe they are fluent. The study does not claim that all men overstate their knowledge, only that the tendency is widespread enough to distort aggregate data.
For teams hiring or building AI products, the practical lesson is to test skills rather than trust stated familiarity. What remains open is whether the overclaiming fades as AI tools become routine, or whether the pressure to appear knowledgeable keeps growing as the stakes around the technology rise.