The Psychometric Heritage Trap
How 100-year-old intelligence tests still dictate the structures of modern hiring
When Alfred Binet put together the first practical intelligence test in 1905, his ambitions were modest. He simply wanted to help French schools identify children who might need a little extra educational assistance. Binet explicitly warned against treating his scale as an immutable measure of innate intelligence. But his warnings were largely ignored once the concept crossed the Atlantic.
In 1916, Lewis Terman of Stanford University adapted the work into the Stanford-Binet scale, introducing the intelligence quotient and a distinctly American appetite for social sorting. Terman believed the test could neatly identify future societal leaders and justify occupational stratification. The goal shifted from helping struggling students learn to ranking human beings for industrial efficiency.
A century later, this heritage quietly runs modern hiring. Contemporary pre-employment assessments and automated platforms rely heavily on those early psychometric principles, using standardized cognitive tests to filter candidates at massive scale. Algorithms chase statistical correlations between test proxies and job performance metrics, often prioritizing mathematical optimization over any qualitative understanding of a human being's actual capability.
The result is a remarkably narrow corridor of acceptable candidate profiles. Optimizing humans for algorithmic compatibility produces a predictable sort of homogeneity, turning job applications into an exercise in gaming diagnostics rather than showcasing genuine competence. Candidates learn to speak the language of the test, while non-traditional individuals who excel in messy, real-world environments get systematically screened out.
Industrial-organizational psychologists still defend standardized tests as the single best predictor of job performance, while critics argue they merely perpetuate legacy biases while failing to measure practical adaptability. Yet beneath the debate lies an unresolved machine learning puzzle: to what extent are modern hiring models discovering genuinely novel signals of competence, or are they simply overfitting to the biased ghost of the past?
Published