Forewarned is forearmed: Using AI to detect SDR tendencies as personnel selection tools

Abstract

Current practices of personnel selection often use questionnaires and interviews to assess candidates’ personality, but the effectiveness of both approaches can be hampered if social desirable responding (SDR) occurs. Detecting biases like SDR is important to ensure valid personnel selection for any organization, yet current instruments for assessing SDR are either inefficient or insufficient. In this paper, we propose a novel approach to appraise job applicants’ SDR tendency by employing Artificial Intelligence (AI)-based techniques. Our study extracts thousands of image and voice features from the video presentation of 91 simulated applicants to train two deep learning models for predicting their SDR tendency. The result shows that our two models, namely the Deep Image Model and Deep Voice Model, can predict SDR tendency with 82.55% and 88.89% accuracy rate, respectively. The Deep Voice Model moreover outperformed the baseline model built on a popular deep learning algorithm ResNet by 4.35%. These findings suggest that organizations can use AI driven technologies to assess job applicants’ SDR tendency during recruitment and improve the performance of their personnel selection.