Academics have found that hiring managers are more likely to recruit people they think would make good friends, a phenomenon one academic has coined “cultural matching”. This finding is vindicated by further research that shows that hiring committees are more likely to choose candidates that reflect themselves, and so unconscious racism, sexism and ageism enter the process. To eliminate bias from the hiring process, experts suggest standardising interviews, among other strategies like blind CV screening.
With artificial emotional intelligence software Human, interviews can be standardised to the extent that the questions are delivered to all candidates in exactly the same text format and assessment of the candidate’s performance is taken out of human hands.
Yi Xu, CEO and founder of Human, explained the process to DataIQ - the recruiter sends a link to the candidate with the interview questions, typically between five and ten; candidates record a video of themselves answering the question; this is sent back to the recruiter who can review it at any time.
"The machine is able to read micro-expressions and translate that into emotion and personality scores.”
The Human software also analyses the video recording, using machine learning to read all the candidate’s facial expressions and decipher those into emotions and the characteristic traits, and sends those scores to the recruiter.
Xu said: “When we have a millisecond of movement in our face, we call that a micro-expression. The machine is able to read micro-expressions and translate that into emotion scores and personality scores.” (Viewers of the FOX TV drama series "Lie to Me" will have seen a similar process depicted using the Facial Action Coding System.)
“We manually label faces, then we train the machine to mimic those observations."
To train the machine, Xu said they looked at 500,000 different faces, including European-Americans, African-Americans and Asian-Americans. “We manually label those faces for emotions and personality. Then we train the machine to mimic psychologists’ observations, who probably represent the highest accuracy in terms of emotion detection,” she said.
The software detects eight emotions, including happiness, anger and surprise, as well as seven characteristics, such as honesty, curiosity and confidence. Xu explained that these are expressed the same way by everyone across the globe, though to different degrees of intensity. “The machine puts everyone on the same benchmark and brings time efficiency and cost efficiency to the screening process,” said Xu.
Human has already worked with the NHS in the UK and Major League Baseball in the US and has just announced a new partnership with recruitment software Workable.
The tool isclaimed to have the potential to save hiring managers a vast amount of time as hiring a candidate for a vacant position can be an arduous process. In the UK and Ireland, the average time to fill a position is 48 days.
This is time that Xu did not have to spare when she was recruiting, working as an investment banker on Wall Street and subsequently in the City of London where 15 to 20-hour days were common.
"No matter how you interview, it was a 50% success rate."
She said the least favourite part of her job was setting aside two hours for hiring interviews. And, even then, there would be a one in two chance she would have to repeat the process. “The worse part is, no matter how you interview, you just cannot pick the right candidate. It was a 50% success rate.”
Xu was sure that other people who had gone through the experience of interviews felt challenged by how much time it consumed and the low accuracy rate.
When she left the investment banking world, she went on to work for a television channel, presenting the finance and political news. It was there that began to think about how we as humans react to the facial expressions of others.
“I realised behind the screen, the audience judge you quite brutally because, if they don’t like you, they just switch the channel. My audience views went up and down everyday, so I wanted to quantify if there was a way to know how and why they judged me,” she said.
Despite automating the very human process of meeting and assessing (or judging) people, Xu said she hopes her software will help us have a better understanding of what makes us human.
“Fundamentally, we have been around for 5,000 years. Do we really know ourselves? Our innovation is a small step that can really help us to understand ourselves better.” She added: “For me, what’s important is to really understand who we are as humans and what’s intrinsic about us, is not the IQ part but the EQ part, the emotions and personality that is embedded within us.”