Will AI in interviewing replace, or enhance humans?
As Artificial Intelligence (AI) develops more use cases for businesses, it is increasingly being used in the interviewing and assessment of candidates. It is important to understand and weigh up the pros and cons of any technology before using it, which raises the question as to whether AI will become the future of interviewing and replacing humans, or whether it is an example of technology that can be used as a powerful tool to enhance the human performance and decision making process.
AI already in use
AI has been developed and used throughout the hiring process already, with Amazon working on AI to sift CVs, HireVue claiming to be using facial recognition to assess candidates via one-way-video interviews and even fully robotic interviews by Tengai. However, the results have been mixed as Amazon had to stop development of what turned out to be a sexist AI, HireVue have a class action lawsuit against them with candidates unclear as to why they were not hired due to lack of transparency on how the technology worked and there have been further studies into bias that is affecting facial recognition technology.
This has created bad press and more questions as to whether AI can replace human decision making when it comes to interviewing and hiring. Perhaps this is misplaced, but this doubt has shown that either the technology is not suitable, not advanced enough or that the industry is not yet ready.
Different types of AI
Whilst there may be some bad press, there are also many examples of success not just in AI, but in using data throughout the recruitment process. Natural Language Processing (NLP) can be used to analyse the content, pulling out key words and phrases in order to determine how a candidate has performed against certain scoring, as well as sentiment analysis to give more context to a candidate’s answer. The important part with NLP, as well as other AI, is that the inputs are of good quality, the algorithm is developed without bias and the overlying questions and scoring criteria are fair. If this is done correctly, the output will be candidates that are the best for the job.
Just being able to say the right things is sometimes not enough to give a fully rounded picture of a candidate, so another layer that can be applied is voice recognition and analysis. By understanding the intent and sentiment in the voice, the level of engagement and negative or positive indicators can be measured and scored. This is especially valuable for roles that require a lot of engagement or human interaction, a sales role for instance. Companies such as Genius Voice are working towards this, helping devices to better understand humans, with many applications across the recruitment process. Voice recognition however, is at the mercy of the quality of audio, which can be difficult with older devices or if interviewing over the phone or video call. Once the input is sorted, the analysis and subsequent analysis provide another valuable layer in the assessment of candidates.
The final layer is facial recognition, which in theory is able to decipher and interpret facial expressions and movements as part of the assessment of candidates, through video recordings. Although NLP and voice recognition are quite advanced in this area, facial recognition is still quite nascent and unproven in recruitment to some extent. The HireVue situation is a good example of this as they claim to have over 25,000 data points to use, but they have not provided more insight into how this translates into a hiring decision. However, as this technology evolves, with more confidence and accuracy, this may become an important data source to create a more thorough assessment of a candidate.
The potential pitfalls
Considering the amount of development in this technology, why are more companies not using this for their interviewing and hiring, especially as data analysis and AI is being used in other areas.
Accessibility and availability are two barriers to using AI; not many technologies are actually capturing the data from interviews, and even then there is potential difficulty in sifting through the large amount of data to surface the important things that have been said. But there is always opportunity in data and by condensing and sorting the content into more manageable and relevant chunks, value can be driven by the analysis of this key content. This is where AI is most useful as it is all about the outputs from the analysis; it has to add value for it to be adopted.
Although AI can be used to reduce bias, there is still a potential risk that bias can exist within the technology itself. When developing the algorithms and using historical data, bias can be inadvertently built in from the start. The aforementioned example with Amazon is evidence of this, when bias is not considered, or the data being used is not diverse or of good quality, then an algorithm and the analysis being performed will perpetuate these biases. The HireVue example also highlights that this is not just about bias, but also about the right people being hired. By ensuring high quality data inputs, clear and measurable outputs and thorough, continual testing and review, this risk can be mitigated.
As the use of AI becomes more widespread, different types of companies from various industries will adopt it, each with their own scenarios and method of assessment. AI providers need to ensure that they are able to respond to these different use cases so that their software is capable of being used in a variety of situations, what works for one user may not work for another. By using adaptable models, dependent on the inputs being provided, providers are able to use AI across industries and users.
So why use it
Whilst there may be pitfalls to be aware of, for some companies the potential benefits of using any or a combination of AI methodologies can be great. Accuracy, time saving and reducing bias are all key areas that can be improved upon in interviewing and AI can affect all three.
Accuracy; candidates should be progressed based on their skills, experience and suitability to the job, which obviously relies on the assessment framework and scoring being well suited and appropriate to the job being hired for.
Through the capture, analysis and presentation of data to the interviewer, a more thorough evaluation of the candidate can be performed. If using a combination of NLP, voice and facial recognition, in addition to the normal method of interviewing and assessing, the interviewer has many more data points from which to draw an assessment of the candidate. By using data to drive decision making, a more informed decision can be made on whether or not to hire candidates.
However, the analysis should still be providing suggested scores, rather than making the decision entirely. The human element still needs to be involved as there are many complexities and nuances involved in assessment so the human interaction and decision is still important in selecting the right candidates.
Time saving; by suggesting scores to an interviewer, time savings can be realised as the interviewer is able to review and amend as necessary. At first this may not be substantial, but as the confidence in the algorithm grows, then the savings will increase. Capturing the data from the interview, analysing it and feeding it back to the interviewer also reduces the need for note taking and makes revisiting questions and answers much quicker and easier. Other assessors are also able to view the interview, reducing the need for multiple interviewers in one interview, or multiple rounds of the same interview with different people.
Reducing bias; similar to accuracy, by using the candidate’s content to suggest scores the emphasis is put on the candidate’s performance and not their demographic. Bias, be it conscious or unconscious, is a big issue in interviewing and hiring, but it is difficult to identify and reduce. By using data as a means to assess candidates, this can highlight cases where bias may have occurred and it can help the interviewer to make a better and less biased hiring decision.
All of this analysis and technology is very advanced, but in the context of interviewing, AI should still be seen as an enhancer rather than a replacer of humans.
Should you use it
The decision as to whether someone should be hired or not is too nuanced and complex for just an algorithm to be relied upon to make that decision. Perhaps some rounds where specific skills are being tested it could be, but as more jobs are becoming about human interaction, the human element is still vital.
What can and should happen is that interviews become more specific and focused, as technology is used for the more generic rounds. The candidates being put forward for these interviews will be higher quality and more suited to the job, meaning that interviewers can spend more time on interviewing and assessing the best candidates. Technology and AI should then be used to provide the interviewer with as much data and information as possible to allow them to make more informed decisions, enhancing their roles, rather than replacing them.