Explainable AI

Interviewer.AI - What’s the technology behind it?

Our Explainable AI Approach

At Interviewer.AI, we build Explainable AI [1] to help teams identify desirable talents in their talent acquisition processes. Having an Explainable AI framework allows us to provide insights on the key performance factors of candidates and minimize the risk of prejudice judgements that may arise from black-box AI approaches.
Our Explainable AI assesses interview candidates’ soft skills through evaluation of key success factors identified by I/O psychology heuristics and industry knowledge [2],[3] . From asynchronous video interview data provided by candidates, visual, audio and textual information can be extracted for computer vision, natural language processing, and audio analysis tasks for our AI to perform. These observable pieces of information are evaluated by our Narrow AIs [4] that were built to specifically measure these observable features of a candidate such as eye-contact, emotional state, energy level, etc. These observable features are then provided as data points to evaluate the candidates’ key success factors through a combination of I/O psychology heuristics and machine learning to arrive at the final interview score of the candidate.
Our AI models are trained to be blind towards age, genders, ethnicity for a fair and objective assessment approach. Our model training datasets are global and purposefully curated to avoid any undesirable historical biases in hiring [5],[6],[7] . We ensure that there are fair representation of age-groups, genders, and ethnicity for each label in our supervised training datasets.
When augmented with traditional talent acquisition human resource processes, our Explainable AI brings the value of objectiveness, scalability, and explainability to the professional human teams to boost the effectiveness and efficiency of the department.

Addressing Common Concerns

"What’s the accuracy of your AI model?"

Our accuracy is measured at the Observable Features level. On average, our models' test dataset accuracies are between 85-90% accurate. All models must have at least 70% test dataset accuracy to be considered for deployment.

"How do you test for “machine bias” considering even a unicorn like Hirevue had to shut down their model due to bias found during an audit?"

We avoid the black-box approach where an unexplainable black-box AI is used to make judgements which could be judging candidates on an unfair basis. Instead, we approach the task from an Explainable AI approach where Narrow AIs are built to evaluate defined observable features with high accuracy, which are then used as inputs to calculate the key success factor scores using explainable machine learning methods where input feature importance can be extracted.

"How do you replace the current traditional recruitment process that relies heavily on gut/instinct with your AI?"

We are not aiming to replace the current traditional recruitment process performed by human HR professionals. Instead, we aim to augment that process by integrating the scalability and objectivity of Explainable AI interview assessment to allow the HR recruitment process to increase its effectiveness and efficiency.

“What about the data sets? Does it encompass a global mix or is restricted to a regional sample?”

Yes, our dataset is global. Our models are trained across different demographics and regions. Our AI is trained to be objective and not-bias to any skin-color, age or gender. We curate our datasets to avoid any historic data biases and validate our model on current data. As such, our AI is trained to be blind to gender, ethnicity and age.

"Related to the first. If I am a non-native speaker of English, it might also affect my confidence level."

This challenge occurs both with AI interviews and traditional recruitment interviews by humans if the candidate is speaking in a non- native language. One advantage our AI has is that it can assess interviews in different languages, which allows the candidate to present in their native language if the human interviewer allows multi-lingual responses on our platform.


1Gunning, D., Stefik, M., Choi, J., Miller, T., Stumpf, S., & Yang, G. Z. (2019). XAI—Explainable artificial intelligence. Science Robotics, 4(37). 2 Robles, M. M. (2012). Executive perceptions of the top 10 soft skills needed in today’s workplace. Business communication quarterly, 75(4), 453-465. 3 Kačamakovic, M. K., & Lokaj, A. S. (2021). Requirements of Organization for Soft Skills as an Influencing Factor of Their Success. Academic Journal of Interdisciplinary Studies, 10(1), 295-295. 4 Dickson, B. (2017, May 12). What is Narrow, General and Super Artificial Intelligence. TechTalks. https://bdtechtalks.com/2017/05/12/what-is-narrow- general-and-super-artificial-intelligence/. 5 Correll, S. J., & Benard, S. (2006). Gender and racial bias in hiring. Memorandum report for University of Pennsylvania. 6 Petersen, T., & Togstad, T. (2006). Getting the offer: Sex discrimination in hiring. Research in Social Stratification and Mobility, 24(3), 239-257.. 7 Wilson, M., Parker, P., & Kan, J. (2007). Age biases in employment: Impact of talent shortages and age on hiring. University of Auckland Business Review, 9(1). Dickson, B. (2017, May 12).

Related Posts

Scroll to Top