Narcissism, Machiavellianism, and the people who lean hardest on AI

Skip to content

News

Latest

Deep tech

Sustainability

Ecosystems

Data and security

Fintech and ecommerce

Future of work

Conference media hub

More

Startups and technology

Investors and funding

Government and policy

Corporates and innovation

Podcast

In 1914, Sigmund Freud borrowed a figure from Ovid and handed psychology one of its most durable terms. In his essay On Narcissism, the myth of Narcissus, the youth who fell in love with his own reflection and wasted away in front of it, became a way of describing a self that turns its love inward.

The word outlived the theory that carried it. It hardened into diagnosis, entering the psychiatric manuals as narcissistic personality disorder by 1980, and then loosened again into everyday speech, where it now stands in for anyone we find insufferably self-regarding.

Modern personality psychology treats narcissism more coolly, as one dimension along which everyone varies, grouped with three equally unflattering others, Machiavellianism, psychopathy, and sadism, under the label the Dark Tetrad.

What Freud could not have foreseen is the interlocutor these traits would eventually find. A century after he named the condition, the self that loves its own reflection has been given a machine that reflects, flatters, and answers on command, and never once argues back.

A growing body of research suggests this is not a neutral development, and that some personalities are drawn into it further than others.

The 💜 of EU tech<br>The latest rumblings from the EU tech scene, a story from our wise ol' founder Boris, and some questionable AI art. It's free, every week, in your inbox. Sign up now!

Personality psychologists, following the work of Delroy Paulhus and colleagues, treat these four not as diagnoses but as continuous dimensions, present in trace amounts across the general population and measured by degree rather than category.

Narcissism brings grandiosity and a hunger for admiration; Machiavellianism a cool, instrumental view of other people; psychopathy low empathy paired with impulsivity; and sadism the disposition to take gratification in another’s discomfort.

The question a study can put to them is empirical, and slightly strange: does standing higher on any of these dimensions predict how a person uses a chatbot?

In a study published in BMC Psychology in June, Kağan Kırcaburun, of Düzce University in Turkey, tested how these traits relate to what the literature terms problematic generative AI use, a pattern modelled on established behavioural-addiction frameworks in which engagement becomes compulsive and begins to impair everyday functioning.

Drawing on 677 adult users recruited through Prolific, with a mean age of about 40 and an even split by sex, he estimated a path model connecting the dark traits to problematic use.

Narcissism was the most consistent predictor, showing both direct and indirect associations; sadism produced small but significant total effects; psychopathy behaved less predictably, its direction differing between men and women.

The paper’s more substantive contribution lies in its mediators, the variables through which a trait exerts its influence. Kırcaburun modelled two constructs as serial pathways between personality and problematic use.

The first, vertical individualism, is a value orientation, in the tradition of Harry Triandis’s cross-cultural work, that couples a focus on the autonomous self with a concern for status and standing above others.

The second, intolerance of uncertainty, is a dispositional tendency, now widely treated as a transdiagnostic “fundamental fear”, to experience ambiguity and not-knowing as distressing and to be resolved at once.

Across the full sample and within each sex, those two variables were the most stable correlates of problematic use.

Among the women, the association between narcissism and problematic use ran entirely through vertical individualism; among the men, individualism and intolerance of uncertainty carried it in part.

The traits, on this reading, act less directly than through the values and the affective intolerances that express them, which is a more interesting and more testable claim than the blunt assertion that difficult people misuse chatbots.

The appropriate caution here is considerable, and the author is candid about it. The design is cross-sectional, so the mediational language describes covariance structure rather than an established causal sequence; longitudinal or experimental data would be needed to license claims about mechanism over time.

All measures are self-report, which raises the familiar problem of common-method variance. Measurement invariance across sex was not fully supported, which is precisely why the sex-differentiated findings warrant restraint.

And the base rates of genuinely problematic use in the sample were low, a reminder that this is a study of subclinical...