The project was ambitious but ended very badly. In 2018, Amazon engineers developed recruiting AI to accelerate and improve the hiring process. Apart from the fact that sexist prejudices were quickly discovered, which forced the Jeff Bezos company to put an end to this program.
More than two years later, the problems are still there, according to a study by researchers at the University of Melbourne. Scientists initially gave 40 recruiters real résumés for jobs at UniBank, the company that funded the experiment.
The fight against machine bias also includes educating people
Half of the selection board went through applications that indicated the candidate’s gender, while the other half received the same CVs with names exchanged. So as stated TheNextWebMark became Sarah and Rachel was replaced by John. The result is then final. Recruiters consistently preferred male first names to candidates’ résumés, although the latter had similar qualifications and experience as their female counterparts.
The scientists used this data to create a recruiting algorithm. The latter reproduced the prejudices of the people, even if the names of the candidates were deleted.
Based on this observation, the researchers believe that recruiting algorithms should be more transparent in order to reduce the risk of bias. They also believe that educational work must be done against human prejudice, otherwise the latter will ultimately have an impact on the machines.
The problem is definitely not new and AI specialists particularly denounce the racist abuses that can arise from the use of facial recognition technologies.