Recent advancements in technology have paved the way for organizations to utilize artificial intelligence in conducting interviews. This innovative approach allows for spoken responses to be automatically transcribed using Machine Learning, also known as automated speech recognition (ASR). However, concerns have been raised regarding the accuracy of ASR recordings and whether they exhibit consistent precision across all demographic groups. A study conducted by Hickman et al. (2024) aims to delve deeper into these issues.
Automated Transcription Analysis
The research involved analyzing recorded interviews from over 1,000 participants who responded to 3-6 interview questions, generating more than 140 hours of video content. These responses were transcribed using various platforms such as Amazon Transcribe, IBM, and OpenAI’s Whisper, and then compared to transcripts and evaluations produced by human transcribers.
OpenAI’s Whisper emerged as the platform providing the most accurate transcriptions, closely resembling those produced by human transcribers. On the other hand, IBM and Amazon exhibited lower levels of accuracy. ASR inaccuracies were more prevalent when transcribing responses from individuals whose first language was not English (ESL), non-white participants, and male interviewees. However, these transcription errors did not systematically lead to biased assessments of applicants’ competencies and qualifications.
Practical Implications and Recommendations
While transcription errors were more pronounced among ESL, black, and male speakers, this did not necessarily diminish the predictive capabilities of this technology in the employee selection process. Nevertheless, it is crucial for practitioners to continue monitoring ASRs to ensure that these tools facilitate fair recruitment decisions and do not inadvertently disadvantage specific demographic groups.
For more information, refer to the study by Hickman, L., Langer, M., Saef, RM, & Tay, L. (2024) published in the Journal of Applied Psychology. Stay informed and stay ahead!
Image Credit: Unsplash+