Bias has always existed in hiring decisions and the interviewing process. When decisions involve people it is nearly impossible to eliminate subjectivity. Unconscious bias and mirroring exist far more than we realize when interviewing people.
The challenge facing hiring teams is how do we standardize our decision-making process to reduce human bias and create more defensible, collaborative and transparent hiring decisions. And, how do we improve our efficiency in this process?
It is these two questions that have driven the demand for digital interviewing technology to support human assessment processes, particularly for high volume hiring environments. Further, with the globalization of workforces and the gig economy, it is important for recruiting teams to remain competitive by offering interviewing options that are convenient that candidates can access on their schedule, from any device.
Enter, digital video interviewing. Recruiting teams have quickly adopted one-way video interviewing as a way to meet candidates virtually but allow them to respond to questions on their schedule, from any device, by a stated deadline. This prevents recruiting teams from having to schedule a live video or phone interview session reducing time to fill by 35% or more. Even with one way interviewing recruiting teams are still required to review the videos but this shift from phone to video saves up to 25 minutes. For recruiting teams dealing with higher volumes of applicants, there is a demand for further atomization in the process. Enter AI. Your immediate reaction may be “but AI can introduce bias in the process and screen out good candidates”. Undoubtedly you have read about the different kinds of AI where AI has generated bias-based decisions. This typically happens when AI is trained on biased data or when AI utilizes facial or visual elements in decision making. But not all AI is the same.
As AI is being widely adopted in recruiting and assessment stages in the hiring process, it begs a closer look at what the AI is looking at when making these decisions. AI assessment comes in all shapes and sizes and looks at different types of data depending on the system and software being used.
Recently we have seen more information being published about facial recognition and microexpression analysis in digital interviewing. It’s important for companies and candidates using and participating in digital interviews to ask and understand how AI is making decisions and what data it is looking at. Is it custom trained data or generalized, global data? Is it looking at visual elements or just text? In the case of our software, videoBIO Recruiter, no visual aspects are analyzed by AI. Now you may be wondering, how is this possible if you are a video interviewing software company? It is the audio from the video that allows us to assess the language-only responses by extracting the audio from the video file or inviting candidates to respond in audio-only. It is this process that allows for the baseline for blind interviewing.
The promise of AI is to support hiring teams in providing supporting data for decision making by analyzing the interview and providing a score that recruiters can review to help inform their decisions. Some recruiting teams have embraced AI fully and moved quickly to automated decision making relying on the trained machine to make the decisions for them. Others have taken a more stepped approach to use AI as a data-supported human process where recruiters review the AI data and use this data to support human decision making.
Armed with more information about AI and the data it is trained on, it warrants another look at how this supporting data can be beneficial in the decision making process. Andrew McAfee, author of Machine, Platform, Crowd, and The Second Machine Age says, "As good as we are at [judgment], we're also biased and flawed and buggy…In this era of very, very powerful artificial intelligence and machine learning, computers are exercising what we used to call 'judgment' and doing it in superhuman ways."
Blind interviews eliminate any personally identifiable data. This new interview and hiring method has become a mainstay used by corporate giants such as Google. This can be achieved only if you do not collect the name, gender or any other demographic data on the candidate. In the case of video interviews, this is made possible with audio-based responses that transcribe to text in video interviews where no questions are asked that require personal data to be shared. Candidates are invited to participate in a video or audio interview. After the responses are submitted, the software extracts the audio from the video and through a speech to text recognition process, transcribes that audio file to a text transcript of the interview. The AI, having been trained to understand natural language and expressions in the spoken word will analyze the transcript and identify words, phrases, and expressions. The AI, in our case we use IBM Watson’s natural language understanding (NLU), understands hundreds of languages, dialects, and accents. Using NLU, these words get labeled and classified automatically and generate a score which is based on the type of assessment being conducted. For example, it could be looking for specific competencies or expressions of how things would be done in the case of behavioral interviews asking questions about a situation, strengths, work style or performance.
By eliminating the recruiter review of a video or audio response, it can further reduce human bias. While many companies have moved to the step of automated, AI-based decisions, we recommend that companies implement this in a stepped approach to ensure that the trained machine is returning scores accurately. AI can be implemented as supporting data to provide recruiters with more information to help them make decisions. Armed with this information they can review the AI score and the transcript and make a decision without any external, visual, gender or name-based bias.
Going forward, humans and machines need to work together to embrace the best of digitization, data insights, atomization efficiencies, and human decision making. There are many ways that humans can and should remain involved in an AI-supported process that ensures oversight and the best possible decision outcome. Blind interviewing is an essential step towards diverse hiring measures and bias-proofing the decision-making process. While blind interviewing may not be possible at all steps in the process, it is helpful to start small by introducing one blind interviewing or screening step into your process to check and verify that your decisions are being made fairly and with objectivity.
Mark Sethi, Digital Interview Specialist