Hiring Diversity through AI: Do Algorithms Help or Hurt Applicants?

Inspired by @NovemberRain" post in the less-trafficked JOB forum.


In a Bloomberg article last week, JP Morgan announced that it will begin using online behavioral science-based games to recruit college students.

"Beginning this year, the bank will be using the games, instead of in person campus recruiting and traditional college campus visits, to gauge and measure the interest and acumen of college students who seek to apply for jobs at the bank. The games will be in addition to video interviews that the bank has started using in recent years".

J.P. Morgan wrote in a letter to colleges that the new system will help "provide greater consistency and equitability for all who applied".

So JPM has made a decision to implement AI algorithms to force hiring fairness in gender and ethnic diversity. I'm going to repost my comment but I might be blinded by the potential good these algorithms could do. Anyone care to discuss?

Personally I am against it as it dehumanizes the recruiting process. Sounds like an HR / SJW power play to save costs and take over the process from people who do the job for a living.

Will help schmoozers and well connected students even more! Those will be the ones with family connections, network of analysts working in industry already, able to fly out to the city to have coffee chats in person, etc. Also being "fair" now means only looking at your resume versus a holistic view of your character. Hope you're at a target with a top GPA!

Here's straight from JPM's website: "Its matching engine uses supervised machine learning and an auditing technique called AuditAI that creates gender and ethnic fairness in algorithms. Millions of candidates have interacted with the games and other employers have experienced dramatic improvements in the diversity of their entry-level hiring."

Diving deeper into the rabbit hole of pymetrics. So here's how I understand it works:

  1. A bunch (80-100) top investment banking performers take a standardized test.
  2. The computer will analyze the traits of the top performers, such as risk taking, multitasking, etc.
  3. The AI will see if these traits are also correlated with demographic trends and correct / penalize against that when applicants in the future take the test. For example, if the top performing white males JPM hired are risk takers and that predicts performance at JPM, in the future if a white male scores high on risk taking, which would normally predict high performance at JPM, the algorithm will lower the candidate's overall score to promote gender/ethnic fairness because that trait is seen to merely correlate with whiteness/maleness versus be actually predictive of performance. Attribution analysis paralysis abounds!

pymetrics Lead Data Scientist: "We look at what traits make that population [top performers] unique, and sometimes those traits might be predictive not of job performance but the homogeneity of the people who went through it," [pymetrics Lead Data Scientist] Baker said. "And so we use Audit AI to make sure that we don't overweight any traits that are actually more predictive of a certain demographic group."

Anyone still trust Silicon Valley to figure out how to hire investment bankers?


Comments (1)

Apr 1, 2019