How can we quantify diversity?

Emma Lejal Glaude © Alain Herzog / EPFL 2022

Emma Lejal Glaude © Alain Herzog / EPFL 2022

Emma Lejal Glaude, an EPFL master’s graduate in Data Science, has gone on to tackle fairness in artificial intelligence and machine learning at the heart of Switzerland’s largest telecommunication operator, Swisscom.

For Emma Lejal Glaude, the need for fairness in machine learning is a no-brainer. “Think of the classic case that you learn with open datasets - classifying flowers. If you mistake all of the roses with lilies, they won’t get offended but as soon as you touch human beings you need to be careful.”

Lejal Glaude first studied communications systems and went on to complete a Master in Data Science in the School of Computer and Communication Sciences (IC), graduating from EPFL in 2019. She undertook her master’s thesis at Swisscom just as the conversation about fairness in machine learning was beginning and then became the first Data, Analytics and AI Engineer in the Human Resources Department.

The initiative at Swisscom around fairness in artificial intelligence began as a research project and was then applied in the Human Resources Department in an area with the most traction – recruitment. “We were trying to monitor the recruitment decisions made by the recruiter and line manager to see whether we could uncover unconscious biases. For example, does a recruiter or manager have a strong bias for people that resemble them? We used categorized and aggregated data from the candidates and colleagues to answer this question with analytics.” Lejal Glaude said.

With the demographic data, she also focused her work on measuring the diversity of existing teams as well as candidates. The metric she derived, that computes the probability of picking two individuals belonging to different subgroups, is inspired by Simpson’s Index of Diversity, a way to measure the diversity of species in a community. The higher the probability the more diverse your team is. However, the team size and number of subgroups influence the maximum value. To resolve this, she standardized the value of the theoretically most diverse team possible. This formula allows a comparison of diversity levels in a granular way, across team size quantities of subgroups.

If we consider a team of 11 people and define hair color as Brown, Red and Blonde, we will get the following results:

She is very proud of enabling the HR department to experiment with its data and consider new ways of leveraging it. “It’s really a dream job and I feel like HR Analytics can contribute to solving business problems!”

Lejal Glaude always loved solving problems and was sure from a young age that she wanted to be an engineer, although didn’t know in which field. It was EPFL’s Internet Analytics course that gave her the revelation that she wanted to focus on data science, realizing that it is such a powerful tool.

“The more you learn about machine learning, the more you see what it's capable of doing or not doing and the more you understand, or try to disprove, the fears people have around artificial intelligence and how it is developing. We know that machine learning algorithms will either emphasize or remove the bad aspects of human beings because they aren’t sentient, they just use our data and there is bias in data that records human decisions and behaviors. We don't have a choice but to deal with our historic heritage and I believe that with this knowledge comes the responsibility to say that we know there is bias but, mathematically, we can do something about it.”

Author: Tanya Petersen

Source: People

This content is distributed under a Creative Commons CC BY-SA 4.0 license. You may freely reproduce the text, videos and images it contains, provided that you indicate the author’s name and place no restrictions on the subsequent use of the content. If you would like to reproduce an illustration that does not contain the CC BY-SA notice, you must obtain approval from the author.