Generative AI instruments are sometimes used to display screen and rank candidates, create resumes and canopy letters, and summarize a number of information concurrently. However AIs are solely pretty much as good as the information they’re educated on.
GPT-3.5 was educated on large quantities of broadly accessible info on-line, together with books, articles, and social media. Entry to this on-line information will inevitably replicate societal inequities and historic biases, as proven within the coaching information, which the AI bot inherits and replicates to some extent.
Nobody utilizing AI ought to assume these instruments are inherently goal as a result of they’re educated on massive quantities of knowledge from completely different sources. Whereas generative AI bots might be helpful, we should always not underestimate the chance of bias in an automatic hiring course of — and that actuality is essential for recruiters, HR professionals, and managers.
One other research discovered racial bias is current in facial-recognition applied sciences that present decrease accuracy charges for dark-skinned people. One thing so simple as information for demographic distributions in ZIP codes getting used to coach AI fashions, for instance, can lead to selections that disproportionately have an effect on folks from sure racial backgrounds.
