Business school professors trained an AI to judge workers’ personalities based on their faces

The Gilded Age has entered the chat.

I’m on a 20+ city book tour for my new novel Picks and Shovels. Catch me TONIGHT (Feb 17) for an event in MENLO PARK with Charlie Jane Anders, and TOMORROW (Feb 18) for an event in BURBANK with Wil Wheaton. More tour dates here.

Theory-free inference is a hell of a drug. For years, Big Data advocates — the larval form of today’s AI weirdos — have insisted that if you have enough data, you can infer causal relationships between complex phenomena without ever having to understand how x causes y, and thus, we can slay the dread “correlation is not causation” beast.

This is cousin to Milton Friedman’s famous economic catechism:

Truly important and significant hypotheses will be found to have “assumptions” that are wildly inaccurate descriptive representations of reality, and, in general, the more significant the theory, the more unrealistic the assumptions (in this sense)

https://en.wikipedia.org/wiki/Essays_in_Positive_Economics

AI turns out to be a great tool for creating plausible statistical correlates of imaginary phenomena. Remember the guy who claimed to have invented Machine Learning Gaydar by analyzing the faces of gay people and comparing them to straight people? Same dude later claimed to have invented an AI that could guess, from your face, whether you were a Republican or a Democrat:

https://pluralistic.net/2021/01/15/hoover-calling/#phrenology

This is just AI Phrenology, a continuation of the “scientific racism” movement that was invented to provide a justification for colonialism, slavery, genocide and eugenics. It imagines that there are invisible genetic traits that determine things like your ability to be a good boss, or whether you will cheat on your partner, or whether you are destined to be rich. It’s a kind of cod-scientific astrology, where you get to declare yourself to have been born with “good blood” that destined you to rule over others.

Amazingly, this “scientific” philosophy has somehow managed to thrive after the rise of computational genomics, the science that analyzes population-scale genetic surveys to identify whether there is any genetic basis for the idea of “races” (and other cherished distinctions of the “human diversity” movement) have been shown to have no discernible basis in, you know, genetics.

As Adam Rutherford — a superb science communication and accomplished computational genomist — writes in his 2020 book How To Argue With a Racist, nearly everyone on Earth is descended from the same tiny group of survivors of a couple of severe genetic bottlenecks, the exception being Africa, where there is far more genetic diversity than in the rest of the world:

https://pluralistic.net/2020/08/16/combat-wheelchairs/#race-realism

A Swede and an Australian Aboriginal person are more closely related than two members of the different groups of San people. If genes were the dispositive factors in human personality and accomplishment, we’d expect to see far more variance in the outcomes of African people than we do between, say, Inuit people and Italians. And yet, somehow people who all live in the same society, facing the same structural challenges of post-colonialism, international looting, and a global IP regime that denies them the ability to manufacture their own medicines and fix their own equipment produces people with remarkably similar outcomes. Meanwhile, it’s surprisingly easy to predict the life outcomes of people from very different societies, based on those societies’ position in the global hierarchy.

Sure, genetics play a role in shaping our outcomes. We are built out of the interactions between our genome and the physical and social world around us. But all evidence points to the social and physical factors grossly outweighing the genes. Back to astrology: distant celestial objects inarguably interact with us at our births and through our lives. Some infinitesimal tidal stress is exerted upon the Earth by other planets; photons streaming from faraway, long-dead stars shower down upon us. But the gravity exerted by, say, Saturn, on your body as you pass through the birth canal is less than the force exerted by the paper covers the midwife wears over her shoes in the birthing room. Sure, those disposable covers are a lot less massy than Saturn — but they’re far closer, which matters when you’re talking about forces that attenuate at the square of distance.

Genes play an important role in the development of your brain and the systems that regulate it, like hormones and nerve signals. But that role is clearly swamped by the role that the physical and social environment play as you grow up. You don’t have “good blood” or “bad blood.”

But people who believe in — and benefit from — social hierarchy have always yearned for a freestanding, objective basis for the fact that they have more and everyone else has less. That’s the origin of “efficient markets hypothesis” (I’m rich because the market thinks I’m a good “capital allocater”), of “meritocracy” (if I’m rich, I must have merit), and “evolutionary psychology” (“Honey, it’s not my fault I fucked my grad students — blame the bonobos!”):

https://pluralistic.net/2020/08/30/arabian-babblers/#evopsych

Which brings me to this week’s caliper-wielding AI: the “Photo Big 5” AI that can look at your face and predict whether you’re going to be good at having an MBA:

https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5089827

This is the creation of four academics at elite institutions — however, their discipline isn’t genetics. They’re business school professors. They got a bunch of MBAs’ self-assessed results on surveys of “Big 5 personality types” — itself a kind of astrological exercise with barely more rigor than, say, Meyers-Briggs — and then fed these results, along with the subjects’ Linkedin profile photos and self-reported salaries and titles to an ML and produced — voila! — a machine that tells you whether you’ll be a good manager based on your face!

This is an objectively very funny exercise, like AI Gaydar for middle-managers. They resort to some hilarious obfuscation:

Photo Big 5 exhibits only modest correlations with cognitive measures like GPA and standardized test scores, yet offers comparable incremental predictive power for labor outcomes.

In other words, we created a new random-number generator that is as bad at predicting your life-chances as the SATs or your GPA, two extremely bad ways of predicting your life chances — except to the extent that both numbers can be inflated if you start with a bunch of money and hire elite test-prep consultants. Good thing personal appearance has no correlates with wealth and there’s no way to spend money to look more like a member of the elite? Naw, it must be the genetics underpinning the relationships between your “craniofacial features and behavior.”

It’s easy to see why AI is so tempting to people who want to incinerate any qualitative factors in a complex societal problem, transforming them into dubious quantitative residue that an algorithm can do math on:

https://locusmag.com/2021/05/cory-doctorow-qualia/

It’s junk science at scale, with a business model. The purpose of this automated eugenics is the same as every “rational” account of hierarchy in human history: to retroactively justify winners, and to condemn losers before the game even starts.

If you’d like an essay-formatted version of this post to read or share, here’s a link to it on pluralistic.net, my surveillance-free, ad-free, tracker-free blog:

https://pluralistic.net/2025/02/17/caliper-ai/#racism-machine

Image:
Cryteria (modified)
https://commons.wikimedia.org/wiki/File:HAL9000.svg

CC BY 3.0
https://creativecommons.org/licenses/by/3.0/deed.en