Would you trust a robot to conduct a job interview? Talkpush, an American artificial intelligence firm, has designed a chatbot to do just that. Candidates needn’t worry about an untucked shirt or having something stuck in their teeth – but how well can a machine really evaluate a potential hire?
Talkpush’s chatbot asks applicants to record information about their personality and relevant work experience, as they might on a WhatsApp voice message.
The chatbot does an initial sift of candidates. It scans the recordings for key words or phrases and then, if the applicant meets a set of criteria, the system forwards the audio content onto a recruiter or hiring manager, who then decides whether or not to invite them for a face-to-face meeting. Talkpush, its website says, aims to “make hiring conversations faster”, and generally to streamline shortlisting processes, rooting out the under or ill-qualified and saving unnecessary travel on behalf of candidates.
The coronavirus pandemic, and the social distancing protocols it has created, has undoubtedly changed the recruitment landscape. Companies across all sectors have had to adjust to the new normal of the virtual office and mass home-working.
And, when it comes to HR, Talkpush is not alone in using AI to try and help that transition along. The programmatic job advertising software used by another American firm, Appcast, runs predictive recruitment algorithms to analyse market data to determine how job ads will perform.
The software decides when and where to place ads with, the Appcast website suggests, the ability to put them in front of “the most applicable of 200 million candidates.” Appcast has over 10,000 recruiter partners, including job sites Glassdoor and Adzuna, and can, the company says, place an ad based on where a certain kind of candidate is statistically most likely to see it.
Recruiters are also using AI to scan CVs. Sifting through CVs, the Canadian technology firm Ideal estimates, can take up to 24 hours per individual hire. Ideal’s cloud-based software allows employers to upload large numbers of CVs and cover letters, ranking them according to suitability. Ideal’s software looks for key words, length of time in previous employment, and other criteria that may have been specified by the employer using the technology.
Meanwhile, HireWand, a recruitment tech firm founded and based in India but also operating in the US, allows businesses to use an AI-powered headhunter of sorts. Collating candidate data from sources including career sites, publicly available LinkedIn profiles and CVs and applications directly sent to employers, HireWand builds a “talent pool” of potential employees. Similar to Ideal, the HireWand software screens and ranks candidates according to their suitability, but it goes one step further and contacts them on the employer’s behalf.
HireWand produces an automated message that it sends to candidates who match a certain profile, explaining why they are a good fit for a role available with the employer. It then gauges their interest in the position by asking a series of questions.
Inefficient processes are far from the only problem facing recruiters, however. The growing momentum of the Black Lives Matter movement in 2020 has highlighted structural inequalities and lack of diversity. HR departments have a big role to play in changing that picture when it comes to their organisations.
Unconscious bias – implicit, learned stereotypes that are able to influence behaviour – is the scourge of the hiring process. And within that, there are different types of bias. Confirmation bias is the tendency to interpret new evidence as a confirmation of a recruiter’s existing beliefs; personal similarity bias leads to recruiters gravitating towards candidates who remind them of themselves.
Could AI help address these issues? Dr Lewis Paton, a research fellow at the Department of Health Sciences at the University of York, points out that AI technologies are “only as effective as the data put into them.” Paton, who has previously worked on the application of AI to medical school selection, and studies aiming to diversify medical student backgrounds, explains that machines “learn from the data being fed to them, and if that data reflects human biases, then machines can learn them as well.”
Paton adds: “If there are systemic reasons as to why individuals from underrepresented backgrounds aren’t applying for certain jobs or university courses, then it makes very little difference whether it is a robot or a human making the decision on each application… If there is no diversity in the applicant pool, there will be no diversity in the selected applicants.”
James Targett, a consultant at Velocity Recruitment, a London-based firm specialising in construction, admits that unconscious bias is one of the key challenges of his profession. His firm, he says, conducts regular training to overcome this, but he notes that this is not standardised across the sector. “No amount of training can entirely remove it [bias],” he says, “but more regular sessions, or making it mandatory [across all companies in recruitment] would be a good start.”
Targett is not convinced by a computer’s capacity to make particularly insightful decisions on someone’s personality. “The human touch is still necessary when looking for personal characteristics… there are many qualities that a person cannot easily translate to a CV.” But he sees the merit in AI for streamlining some of the hiring process. “In some cases,” he explains, “we can have applicants for a job who don’t even meet the basic qualification requirements, so it would be good to be able to remove those from the outset. It would save time.”
For Paton, there is “possibly a line to be drawn” between what AI can currently do in the hiring process and where human reasoning skills will still be needed. “Having an application rejected because an automated AI system found a number of spelling mistakes in your CV,” he says, “is perhaps easier to understand than having an application rejected because an AI system, through a highly complex formula, predicted you wouldn’t be very good at a job.”
This article first appeared in the Spotlight supplement on AI and automation.