To what extent biases can be introduced to our model, and then how we identify and resolve them?
Author: Recruitment Smart
Posted On: March 11, 2021
Post Comments: 0
To create a truly bespoke model we train our algorithm on the organisation’s internal database/historical data, however we only focus on extracting the skills needed for performing a role and exclude other characteristics such as gender, race, nationality, religion etc. Our algorithm creates a detailed skill map looking at previous successful hires and uses that skill map to assess candidates for similar job opening prospectively. To further ensure that the bias can be contained we have enabled 3 formulations-
Bias Checker- We will look into the job description/skill map and then assess the historical data to see what candidates were shortlisted and selected and candidates who were rejected. Our algorithm will evaluate all the previous candidates considered for a job and come out with a fitment score. It would then do a comparative map to segregate and study outliers(candidates with high fitment scores who weren’t shortlisted and candidates with poor fitment score who were shortlisted/selected). the system would then look at how large this outlier group is and then study if there are any biases prevalent in terms of selection choices exercised by the recruiters(gender, race, religion etc.). This outlier report would then be presented for the recruiter’s perusal along with the details of the type of bias detected
Blind Hiring- We have innovated a cutting edge technology i.e. Blind Hiring to check prospective bias. Through this we can conceal personal identifiers such as name, phone number, age, location, previous employer etc. This gives the tool a handle to address a wide variety of biases substantially including gender, race, religion etc. with varying confidence. The tool then assess the skills of the candidate to make recommendation and only after recruiters have made a decision on the candidature(shortlist/reject etc.) does the tool reveal the name/contact details of the candidate
JD Templatization- We have a feature through which we remove gender bias at the inception of the process i.e. creating standardised job description with verbiage which is inclusive and lists the skills in a gender neutral manner to encourage female candidates to apply for the job.