How to Eliminate Hiring Bias

By Pragati Verma, Contributor

When candidates shortlisted for the strategic digital coordinator position at Sweden’s Upplands-Bro Municipality showed up for their job interview in June, a 16-inch tall robot head called Tengai warmly greeted them.

Sitting at eye-level on a table across the candidates, it blinked lightly and began the interview with a smile. It posed questions such as, “Can you describe a situation where you were faced with a problem that you needed to solve on your own?” Tengai could listen, speak, and react to candidates’ answers while maintaining eye contact with them, thanks to its real-time visual tracking system. Occasionally, it would tilt its head, say “hmm” or lift the corners of its mouth into a smile during the interaction.

Upplands-Bro Municipality is the first employer in the world to use a social artificial intelligence-powered robot in the hiring process,” says Elin Öberg Mårtenzon, chief innovation officer at TNG, the recruitment company behind Tengai. The idea, according to her, is to remove unconscious biases that managers and recruiters often bring to the hiring process.

“It takes only seven seconds for someone to make a first impression, and it’s often based on appearance or handshake style.”

—Elin Öberg Mårtenzon, chief innovation officer, TNG

“Recruiters can start judging candidates even before the interview starts,” she points out. “It takes only seven seconds for someone to make a first impression, and it’s often based on appearance or handshake style.” In contrast, Tengai makes the interview “objective and fair” by getting rid of unconscious biases that color the interview process. “It doesn’t care about things like age, gender, appearance, religion, dialect, or sexual preference, and yet it keeps the experience human,” she adds.

In contrast, Tengai makes the interview “objective and fair” by getting rid of unconscious biases that color the interview process.

To build Tengai, TNG partnered with a Stockholm-based social robotics and conversational AI startup Furhat Robotics last year. Furhat created a platform for social robots that can build conversations with humans by speaking, listening and showing emotions. The platform includes the three-dimensional bust with a projection of a human-like face as well as the operating system with intelligent natural language processing tools to understand and participate in a conversation. Furhat didn’t build the platform for a specific use case, but rather offers developer tools for third parties to create their own applications.

TNG designed Tengai on top of Furhat’s social robotics platform. It began by creating an HR application, including question trees and social skills, to conduct an unbiased competence-based interview process. TNG also trained the robot using machine learning tools and data collected from multiple test interviews conducted by a diverse pool.

Today, Tengai works just like a human recruiter and starts the interview with an introduction, followed by questions to measure soft skills, personality traits, and competency. But unlike humans, it does not make small talk before or after the interaction to avoid forming any unconscious biases about a candidate. For example, it might ask the interviewee to talk about a work situation when he or she found it difficult to work with colleagues in a team or project, but it would never chat about their appearance or sports teams they support. After the interview, the robot presents a summary and assessment based on objective data.

Taking Out Cognitive Biases

Taking bias out of the interview process is significant, if TNG’s 2018 survey of job seekers in Sweden is any indication. Nearly three-quarters (73 percent) of candidates reported they had been discriminated against while applying for a job based on their age, gender, ethnicity, handicap, sexual preferences, appearance, weight, or health.

For Mårtenzon, Tengai is the answer, as “it is unbiased in design and free of unconscious prejudice.” Any AI, however, is only as good as the data and people who train it. Amazon, for instance, recently abandoned an AI recruitment system because the system discriminated against women, particularly for technical roles.

How does Tengai do what Amazon couldn’t? The problem, she says, is that most AI systems make decisions by looking at historical data, which only perpetuates existing biases. To avoid such pitfalls, TNG made it a point to not train Tengai on historical data. “Instead, we train it on [competency-based] behavioral anchors for every specific competency we are trying to measure,” explains Mårtenzon. These behavioral anchors demonstrate a certain competency, depending on the job role. The process looks something like this: The hiring manager first identifies the most essential competencies needed for the role and then designs a few questions—typically to pose a situation that would require certain behaviors that demonstrate the competencies associated with that job. For example, if the role has a customer service focus, Tengai might say: “Tell about a time when you were not effective in meeting a customer’s needs and what steps you took to correct the situation.”

To ensure that no human biases creep in, TNG turned to a mix of recruiters and volunteers, with diverse races, religions and gender , to train Tengai. In addition, they ensured the robot’s algorithm doesn’t look for any data on age, gender, ethnicity, or religion during the interview.

Besides removing human bias, Tengai also helps job seekers get more candid during the interview. “The majority of the candidates say that they can answer Tengai’s questions more honestly and sincerely than with a human recruiter. Tengai seemed to make them more comfortable, and they opened up and talked about difficult issues,” she says. Interviewees don’t need to worry about the interviewer’s body language or “second-guess what the interviewer wants to hear.”

Pick Your Interviewer

Mårtenzon believes that a robot like Tengai is “already great for the initial interview phase.” Currently, it provides recruiters and hiring managers a summary of the interview with data on a candidate’s skills and personality traits so they can decide whether or not to move forward. In the next iteration, the robot will be able to independently decide if a candidate should move to the next stage of recruitment without requiring human analysis of the interview transcripts.

TNG has more plans in the works for Tengai. By early 2020, they expect to roll out an English-speaking version of the robot (Tengai currently only conducts interviews in Swedish). Eventually, they plan to offer customizable faces and voices, and allow job applicants to choose the robot interviewer’s gender, language, and appearance. “Imagine what that would do to the interview process,” adds Mårtenzon.