HireVue, one of our best-of-breed product partners, has recently updated and improved their already-impressive AI-augmented virtual interviewing solution with cutting-edge innovations. In this article, we detail the key enhancements as well as benefits these bring for our clients.
In previous articles, we discussed the HireVue solution in greater detail, so will not repeat that here. But in summation, HireVue virtual interviews are AI-augmented video interviews that are observed, scored, and categorized using machine learning algorithms that utilize natural language processing (NLP) to mimic the way professional human raters would score a job interview.
In doing so, HireVue’s solutions add efficiency, accuracy, and mitigation of typical interpersonal biases that affect traditional interviews.
Innovation #1: Improved transcriptions
While HireVue’s transcription technologies were already able to convert spoken English into written equivalents, new technologies such as those used by Google and others have allowed for even more accurate transcriptions.
Recently, HireVue switched to using rev.ai technologies, which has lowered error rates into the single digits (previous industry benchmarks were around 20-25%).
In addition, these new technologies include a wider recognition of non-native English-speaking accents.
What that means is that HireVue interviews recorded by international or non-first language English speakers will be transcribed with far greater fidelity, despite a wide array of different possible inflections, accents, and word usage.
Innovation #2: NLP improvements
As mentioned before, at its core, HireVue virtual interview assessments leverage powerful natural language processing (NLP) technologies that allow for the contextual recognition of spoken words by the AI.
Based on state-of-the-art models developed by Google and other industry leaders, HireVue uses an improved BERT (Bidirectional Encoder Representations from Transformers) model, RoBERTa (Robust optimized BERT pretraining approach), which takes the base language recognition algorithm and enhances it with industry and task-specific training.
Models like BERT are based on large datasets of words that are paired with contextualized meaning and then used to either recognize natural language or to predict sentences based on prior words (as can be found in predictive text applications). Their main drawback is that they are meant to be applied generically to a wide array of spoken word contexts, from informal emails to formal theses.
By using a more focused pretraining regime like RoBERTa, HireVue data scientists can hone the basic BERT algorithm by training it to recognize interview and competency-specific words and sentences.
In this way, HireVue’s NLP technology is specifically sensitized to recognize words related to job success and workplace behavior as they are used in a typical job interview.
In addition, contextual elements of language (e.g. the different uses of the same word given a particular context, such as “bank” referring to both the institution as well as a riverbank) are taken into account to further reduce error.
The upshot of these improvements is that HireVue’s AI-enhanced interview solutions closely mimic the ratings of trained behavioral experts when scoring interviews for the presence of competencies like teamwork, willingness to learn, and adaptability.
Innovation #3: Removing non-content specific speech processing
A common objection to the use of Artificial Intelligence in scoring data such as interview responses is that algorithms may place inappropriate emphasis on irrelevant aspects of speech, such as inflection, pauses between words, or obscure body language and speech rhythms.
Indeed, early versions of NLP technologies used in interview scoring did consider such data points. But, in subsequent analyses, the HireVue data science team has found that adding such data contributes little to predictive power.
Consequently, HireVue virtual interview solutions no longer incorporate speech patterns, emotion recognition, or similar inputs when scoring interviewee responses.
What this means is that both employers and interview candidates can rest assured that only the content of their answers, and nothing else, will be considered by the AI when responses are scored.
Innovation #4: New fairness optimization
As discussed in previous articles, one of the key benefits of AI-enhanced assessments is that the scoring algorithms can effectively ignore group-based data such as gender and race when evaluating interviewee responses.
Taking this feature even further, HireVue has now introduced fairness considerations directly into model optimization at training time. What this means for the predictive power of the algorithm is that it is not only incentivized to predict job-related outcomes accurately but also simultaneously penalized if it detects any demographic group differences in assessment scores (e.g. men and women have meaningfully different scores).
In using this new approach, HireVue’s assessment process prevents bias from entering into the training model, even if human raters whose data serve as the basis of that training may well have been affected by traditional biases.
In doing so, HireVue is breaking new ground in the minimization and possible elimination of traditional in-group biases that have plagued interview methods across the decades.
By using sophisticated methods such as these, we may well be witness to an age of truly bias-free interviewing.
In releasing these four enhancements and innovations, HireVue has taken an already impressive and best-of-breed virtual interview solution to even greater heights.
In fact, initial results suggest that these improvements yield upwards of a 40% improvement in predictive power (i.e. the ability of the AI scoring to predict job-related competencies based on interview answers).
Given these impressive gains in predictive power and bias mitigation, we look forward to seeing future HireVue innovations and how these will help our clients make better talent decisions.
If you would like to have a conversation with us about how HireVue can help enhance your talent sourcing and selection processes, why not drop us a line at firstname.lastname@example.org?
Zuloaga, L. (2021). The latest leap in HireVue’s assessment technology. Retrieved from: https://www.hirevue.com/blog/hiring/the-latest-leap-in-hirevues-assessment-technology