Humanoide Robots “Tengai Assesses Everyone the Same Way”
How do you completely eliminate bias in recruiting situations? With robots - claims software developer Vanja Tufvesson. She has founded Pink Programming, which organizes Coding Camps for women-identifying and non- binary people, and has been developing one of the world’s first unbiased robot.
By Natascha Holstein and Svenja Hoffmann
Vanja Tufvesson, How did you come to the field of AI technology?
I have been interested in AI technology for a long time. That was before the word AI was as hyped as it is today. At university, I studied mathematics and then specialized in software because I liked the intersection between the two fields. What I found most interesting was the analyses of texts as well as search engines and clustering algorithms. After I had been working in various technology related fields, I realized that as much as I like solving complex problems, it is not crucial to use advanced technology to target real problems. Whether the technology is simple or advanced doesn’t really matter as long as you find the right solution.
How did you come up with the idea to found Pink Programming – one of the biggest non-profit organizations in Sweden organising coding events for women-identifying and non-binary communities – and why is it so important to have it?
Being a minority in the tech field myself in terms of gender, I wanted to make a change. One summer, I spoke to one of the only two female developers that I knew and had been studying with. My idea was to create a programming camp for women with only female mentors and fun social activities in-between. At that time there weren’t many other organizations fighting for the same cause and we received a lot of media attention. The camp got fully booked in a week and we had a hundred people on the waiting list. After the camp, we thought - we can't waste this big interest, but must do something about it.
The organization is important for all sorts of diversity, not just gender but age, background or interests. It can be challenging to work with people that are different from yourself, but it is also more fun. With different inputs you consider aspects from different angles, learn from each other and, most importantly the outcome is much better. All over the world, the stereotype – the common perception of who can be a developer – is shutting many doors, not only for women. You don’t have to have been programming since you were a kid to become a programmer. The field is very broad and I really want to emphasize – as long as you are interested, you can learn.
When did you first notice the problem of AI in science?
To me who studied the technology, I always had the opinion that AI is bias. When you want your computer to learn from previous behaviours that is what you program: find some common patterns, and then being able to place something else into the same category. Isn’t that the definition of bias? So, I was surprised that people even were surprised by this fact.
The important thing here is to realize that AI can be super powerful, but not all problems are suitable for an AI solution. It is important to ask yourself, if there are other powerful solutions that you could use. Usually, for example simple statistics can take you very far when it comes to analysing data.
AI is bias.
Our sister company TNG, which is the leading unbiased recruitment in Sweden, started a long time ago with a first important step towards unbiased recruitment: blind CVs, meaning omitting your name and photo in the application. Later on, their innovation lab considered how they could completely reduce bias. They concluded, the only way to do so is to remove humans from the process and have robots conduct the interviews. They paired up with a Stockholm robotics start-up, which created Tengai’s hardware, the actual machine. Since the interest from the costumers was bigger than expected the owners decided to detach Tengai from TNG and start their own company. That’s when I joined, a bit over a year ago, to lead the technical process of it.
You said, you are removing people from the procss. But when you have thousands of applicaitons, isn't it always people deciding who is invited to the interview in the first place?
We recommend the very first selection to be done by Tengai. She can’t read applications, at least not yet, but the application itself could have some screening questions to filter the applicants. Afterwards, when there are still a few hundred applicants, Tengai would make the first selection and recommend, e.g. the top five candidates to the recruiters. You could have ten robots doing the interviews at the same time and thus wouldn’t need many human resources.
How does Tengai asses whether someone is suitable for the job?
We’re using a framework that is commonly known in behavioural science, called “Big 5”. Those traits are: openness, conscientiousness, extraversion, agreeableness, neuroticism. Currently, Tengai only analyses what the candidate says since there is no research showing that the manner of speaking is relevant for your future work performance.
There is criticism saying that the Big 5 were established and are mostly used in western contexts. Thus, how can Tengai be unbiased if it uses this framework?
The robot wouldn’t know which country the candidate comes from, neither the name nor the gender or looks. It’s only the answer that matters. We’re not questioning proven research but use the latest technology, combine different frameworks and insert it into the robot. In the traditional interview, a recruiter goes into a room with the candidate, comes out and makes the decision. There is no data on what has been said in the room. Eventually, it depends on the recruiter’s mood whether the candidate gets the job. That is one of the biggest reasons why Tengai is unbiased because she treats and assesses everyone exactly the same way. The robot asks the same questions, it doesn’t get tired, it doesn’t diverge into other topics, like shared interests or going to the same school as kids.
She treats and assesses everyone exactly the same way.
We have different types of questions and we combine them. One typical yes-no question would be, “When you are at work, do you like to socialize with your colleagues?”. Then there are more complex questions that would result in an open answer by the applicant, for instance, “When you face an unfamiliar problem, how do you approach that?” After the candidate has answered, Tengai will further ask, “What were the steps?” or “What was the result?”. This is a common technique used by recruiters as well, called situation - behaviour - result.
We have concluded a validation study, which showed that it is quite hard to lie to Tengai and that she is better at noticing attempts of manipulation than any other personality test. We ask multiple questions during the assessment and if you lied, you would need to do so consistently. If you lied to only one question, we would notice that your answers don’t align.
How important is it to have a human face and not a computer in this recruiting situations?
It is important to have a human-like robot because you shouldn’t think that a machine interviews you. Tengai really interacts with you and even though you know she is a robot, you feel like you are talking to a person. You become more honest, but at the same time you don’t need to reflect on your words and way of communicating like you may usually do, like “Was that a good answer? Was I polite enough?”. It’s the first time that you can talk only about yourself in an interview and that’s okay.
Who decided on the facial features and could they be altered?
Technically they could be altered, but we have decided to always use the same face. However, that is a very valid question. The answer is not trivial because who decides on what the ideal face is? Would it be the projection of the candidate’s own face on the robot? Or should it be a celebrity? It is really difficult because in one way we want to have the same for all the candidates, but also different looks might give different impressions.
Who decides on what the ideal face for the robot is?
Which companies are the main target group for Tengai and who uses it already?
At the moment, about 10 to 15 companies use Tengai, currently only Sweden. For any company, recruitment processes can be very expensive, especially if you choose the wrong person. That in itself is a huge opportunity for savings to implement any technology that could help you in the process. If you only have a few candidates who are all very qualified for the job, you might just interview them yourself. But otherwise, it is ideal for high-volume recruitments since the robot can interview many people that way you get the most value out of it.
Do you think, in a few years every bigger company will have Tengai or another recruiting robot?
I think it is definitely the future. There are many aspects of Tengai that are part of the future. For example, we are getting more used to voice technology. For me, I am still a bit unfamiliar to controlling gadgets with my voice. However, the younger generation grew up with this knowledge and will not think it is weird to be interviewed by a robot. In my opinion, the digitalization not only of recruitment but everything we interact with will become natural.
Many other industries still have a lot to gain from technology, they have just begun their journey. Through Tengai, the recruiters have data and support for their decisions, this way they can work more effectively. We are not planning on replacing the recruiter, we are planning on helping them.