The Art of AI Phrenology 2.0
Every now and then, a well-educated scholar poses the question: Could I utilize discredited 18th-century racial science, but employ AI as my tool instead?
The most recent example of this AI phrenology venture comes from a team of economics professors who claim they've devised a method to dissect a single photo of a person's face, harnessing AI to calculate their personality and forecast their academic and professional achievements.
Similar academic ventures into AI phrenology, such as algorithms purporting to predict a person's sexuality or delinquency based on facial features, have been wholly criticized and debunked. Investigations have also exposed that commercial AI tools offline, meant to gauge personality traits, are wildly unreliable.
Ignoring these criticisms and warnings, Marius Guenzel of the University of Pennsylvania's Wharton School, Shimon Kogan of Indiana University, and Kelly Shue from Yale University believed that a snapshot could unveil a person's personality. They received funding for this research from several AI and finance research funds at Wharton and have shared their findings at financial technology conferences worldwide, as their paper implies.
This research team collected the LinkedIn profile photos of 96,000 MBA program graduates and fed them through an AI algorithm, supposedly capable of measuring the person's perceived personality traits according to the Big Five test: openness, conscientiousness, extraversion, agreeableness, and neuroticism.
Subsequently, they correlated these extracted personality scores with the prestige of the MBA program the individual completed and their estimated income in the job market (determined by a proprietary model that scrutinizes LinkedIn data).
Based on these calculations, the researchers concluded that personality exerts "significant influence" on determining if an individual attends a high-ranking MBA program and their income post-graduation. For instance, men in the 20% most 'desirable' personality bracket attended programs ranked 7.3% more highly and boasted estimated incomes 8.4% higher than those in the 20% least 'desirable' bracket. Upon controlling for factors like race, age, and attractiveness (all inferred), these effects lessened.
Notably, the researchers haven't attempted to establish the authenticity of the Big Five personality scores their algorithm allegedly derived from LinkedIn photos. None of the people whose profile pictures were analyzed took a personality test to verify the algorithm's findings.
The professors suggested that their findings emphasize the "vital role" of non-cognitive skills in shaping career outcomes and that using AI to decipher facial features rather than administering personality tests offers "new avenues for academic inquiry ... inviting further exploration into the ethical, practical, and strategic considerations involved in harnessing such technologies".
In contrast, they also recommended that the method not be employed for labor market screening, asserting that "personality extraction from faces represents statistical discrimination in its most basic form".
Effectively, scientists formed a conclusion, questioning whether they should apply this methodology only to discover it was discriminatory, and yet, proceeded to implement it anyway.
The use of AI to analyze facial features for predicting personality traits and future achievements in the field of academics and professions is a contemporary example of how technology, specifically artificial intelligence and artificial intelligence-powered tools, are being applied in unconventional ways. However, the reliability and ethics of these AI methods remain a subject of intense debate, as criticisms and debunkings of similar AI phrenology ventures have highlighted their inaccuracies and potential for discrimination.
