When robots are recruiters
Add Axios as your preferred source to
see more of our stories on Google.
Illustration: Sarah Grillo/Axios
One of the fastest-growing workplace applications of artificial intelligence is in hiring, but imperfect algorithms leave qualified women and candidates of color out — and can ultimately build weaker teams.
Why it matters: Algorithms are most often used to make the initial applicant screening process — the resume review — more efficient. But their role cannot be underestimated, as around 95% of all job applicants are rejected based on resumes.
But, but, but: Using AI to recruit isn't inherently bad, says Danielle Li, a professor at MIT and c0-author of a new working paper on algorithmic hiring. It's about using the right kind of algorithm.
Li and her colleagues tested different types of algorithms using a Fortune 500 company's data. The AI's test was to select candidates for first-round interviews for jobs in consulting, financial analysis and data science — all of which pay well and have been criticized for lacking diversity.
One algorithm used a traditional machine learning approach, making predictions about the future based on data from the past.
- For example, if a company had hired several white, male computer science degree-holders from Stanford and those people had been relatively successful at the firm, the algorithm would demonstrate a strong preference for those applicants.
- "This approach works if you think history is complete," Li tells Axios. "You have to assume that the things that predicted quality in the past will predict quality in the future. But we know that that's not true."
Another algorithm used a much more exploratory approach, incorporating bonus points for applicants who had untraditional majors, came from different places or had unusual work histories. The algorithm was not instructed to prefer applicants of color or female applicants.
- This approach treated hiring as a dynamic problem, Li says, valuing giving applicants the opportunity to show their chops.
- This algorithm increased the share of Black and Hispanic candidates selected for first-round interviews from 10% (with human evaluators) to 23%. The share of women selected went from 35% to 39%.
The bottom line: "Lots of companies have taken interest in using AI tools in the recruiting process," Li says. "In that world, algorithms stand to have a big impact."
