Improving Fairness in AI Interviews with Outlier Detection

Study Details

• Data Sample: 177,691 answers from Sapia.ai’s Chat Interview™ platform, encompassing roles like retail assistants, call centre agents, and cabin crew.
• Analysis Process:Used advanced AI techniques to spot and flag unusual answers, ensuring fairer evaluations by recognising and handling outliers.

Key Findings

• Increased Fairness: Outlier detection models can accurately identify unusual answers, ensuring evaluations focus on relevant responses only.
• Reduced Errors: Human evaluations showed that the model made fewer mistakes, proving its effectiveness in real-world situations.
• Improved Insight: The model is designed to pay attention to the context of questions, helping it better understand and evaluate interview answers.

Key Takeaway
Employing a question-aware outlier detection model in AI interviewers can significantly enhance the fairness of candidate evaluations. This method ensures that outlier answers, which may differ significantly from the majority, are identified and flagged for manual review, reducing bias and increasing trust in AI interviewers. 

Download the full paper here: