Why fairness is a key step in the model build process

Fairness is a key step in the model build process.

Why does it matter? Sapia research shows that gendered language differences can be significantly reduced by mapping to job relevant features such as personality traits, competencies and language skills, and then applying algorithmic scoring. Despite this, assessment vendors still need to embed fairness as a distinct step in the model build process.

We believe we are the first AI company working in HR tech, globally, to do this. Fairness as a constraint means that if the model build violates the 4/5ths rule or has an effect size greater than 0.2 (highest scoring vs. lowest scoring demographic groups), the model build process fails and should not be deployed.