We’re Doing AI Phrenology Again

Every few months, another highly educated academic asks: What if I tried to destroy 18th century race science, but with AI?

D Last entry The AI ​​phrenology portfolio comes from a group of economics professors who say they have developed a method to algorithmically analyze a photo of a person’s face to calculate their personality and predict their educational and career outcomes.

Other recent academic forays into AI phrenology—such as algorithms that predict a person’s likelihood of committing a crime based on their sexuality or facial features— extensively criticized And debunked. The findings also found that commercial AI tools claim to measure personality traits Very unreliable.

Nevertheless, Marius Guenzel and Shimon Kogan of the Wharton School at the University of Pennsylvania; Marina Nissner, Indiana University; And Kelly Shue, from Yale University, concluded that a snapshot of a person’s face can determine their personality. They have received funding for their research from several AI and finance research funds at Wharton and have presented their findings at financial technology conferences and universities around the world, according to their paper.

The authors collected the LinkedIn profile pictures of 96,000 MBA program graduates and ran them through a facial analysis algorithm that allegedly measured how the person would score on the Big Five personality test, which rates people on their perceived openness, conscientiousness, extroversion, agreeableness, and agreeableness. nervousness

They then measured the correlation between these extracted personality scores and the prestige of the MBA program they completed and their ultimate compensation in the workforce (estimated by a proprietary model that analyzes LinkedIn data).

Based on this analysis, the authors concluded that personality plays a “significant role” in predicting whether a person will attend a school with a highly ranked MBA program and how much they will earn in their first job after graduation. For example, men in the top 20 percent of “desirable” personalities ranked 7.3 percent higher in MBA programs and had 8.4 percent higher estimated earnings than men in the bottom 20 percent of personality preferences. When the researchers controlled for factors such as a person’s race, age, and attractiveness (all of which were hypothesized), the effects became smaller.

Notably, the authors do not appear to have made any independent effort to establish that the Big Five personality scores their algorithm extracted from LinkedIn headshots were accurate. None of those whose profile pictures were analyzed took the Big Five personality test to confirm the algorithm’s decision.

The professors wrote that their findings highlight “the important role of non-cognitive skills in shaping career outcomes” and that using AI to analyze faces instead of administering personality tests to humans, “presents new avenues for academic inquiry … [and invites] Further exploration of the ethical, practical, and strategic considerations underlying the use of such technologies.

At the same time, they wrote that the technique they demonstrated should not be used for labor market screening and that “extracting personality from faces represents statistical discrimination in its most basic form.”

In other words, scientists stopped thinking about whether they should, concluded that it was discriminatory, and then did it anyway.