Chatspace data scientist Girija Shingte provides a roadmap on tackling bias in artificial intelligence.
A few months back I came across a video that started off with a riddle: A boy, who is about to interview in a big company, is in the car with his father.
The boy gets a call from the CEO of the company he is about to interview with, and when he answers the call, the CEO says “Good luck son, you’ve got this.” Participants were asked how this was possible? Do you have any guesses before reading further?
“Companies committed to diversity to gender, racial and ethnic diversity are more likely to have stronger financial returns above their competitors”
Some of them guessed the CEO could be the grandfather, or it could be a pre-recorded call from the father, or the guy has two fathers, and some even guessed that the boy’s name was ‘son’. But it turns out, all the participants were wrong, because the correct answer was that the CEO of the company was the boy’s mother.
Reinforcing the same ol’ same ol’
This is an example of the bias we humans are conditioned to. Usually, we don’t realise that these biases exist until we are at the receiving end. Now imagine these biases perpetuating in the data we use in our technologies.
Innovation in technology is exciting and fast paced, and as the past year has demonstrated, our dependency on technology has increased more than ever. Artificial intelligence (AI) in technologies enables us to make informed decisions like what movie on Netflix matches our interests or what’s the best route to drive home. But if the algorithms that are helping us make even more important decisions are trained using biased data, then are we not reinforcing social discriminations in our world today?
Take Amazon for instance. A few years back, their AI-based hiring tool was criticised due to its heavy bias against female applicants. The algorithm used nearly 10 years of data collected from previous job applicants (mostly men) and taught itself that male candidates were preferable. It penalised resumes that included the word “women’s” as in “women’s chess club captain.”
There is an enormous opportunity right now for businesses adopting AI, to critically think about the prejudices in data and their social implications and to find a way to fix them from the very beginning of the infrastructure design and development.
And if “because it’s the right thing to do” is not enough motivation, then leadership needs to consider the bottom line. A 2015 McKinsey study showed that companies committed to diversity to gender, racial and ethnic diversity are more likely to have stronger financial returns above their competitors.
My top 5 learnings
As a woman of colour in tech, I am determined to enhance my knowledge on ethical AI and implement reliable solutions that are free of biases and discriminations.
My own journey in AI began during the final year of my Bachelor’s degree back in my home country India when I was working on a capstone project that could determine a rare eye-disease in retinal images using artificial intelligence. Realizing the limitless potential of AI and its positive implications was a huge inspiration for me. After completing my Master’s in Data Analytics, I started working as a data scientist at Chatspace AI. Here, we are developing an AI-based project risk detector for Enterprise Service Delivery teams. As a team, we are always encouraged to voice our unique perspectives, to ensure we are building a product that is safe, inclusive and trust worthy.
For those of you who have embarked on a career in AI and like me, want to make technology more inclusive and ethical here are a few things I learned from my journey so far:
Seeking inspiration and guidance:
You can’t be what you can’t see. Before beginning my journey in AI, I looked for people who had similar objectives like myself and who could guide me at every step in designing my learning path.
I was fortunate enough to connect with Tushar Gawande, an alumnus from National University of Galway, Ireland, who generously helped me select the right modules for my Data Analytics course that were aligned with the industrial requirements, advised me on approaches to study and selecting the subject for my thesis, and also helped me prepare for my job interviews.
Having someone to look up to, with whom your objectives and values align, is extremely important.
Acquiring skills and experience
AI comprises of different disciplines such as machine learning, natural language processing, robotics, computer vision, etc. Depending on your area of interest, you can learn the skills either through courses and material available online or through a formal course.
One online course that I highly recommend is Jose Portilla’s “Python for Data Science and Machine Learning Bootcamp” available on Udemy. The concepts I learned in this course helped me throughout my Master’s course, interview preparations, and in my current work.
Building and maintaining a network
Build yourself a network. Since Covid-19, a lot of conferences and events are being conducted virtually, which is a great opportunity to build your network with people from all over the world. Connect with people over LinkedIn not just when you are looking for a job referral, but to maintain your network. Discuss about ethical AI and get a high-level understanding of the steps they and their organizations are taking towards standardizing and governing data to fight AI biases.
It’s important to have diversity in your team to highlight the inequities in data and algorithms from the very beginning. Also, being mindful of the recruitment process in your organization and being vocal about what you think is unfair is important.
I recently attended a webinar with data science team leads from IBM, Microsoft, and Facebook who all mentioned that they were extremely diligent about the jargon used in job descriptions.
Their objective was to ensure that a female applicant should never feel that the job is meant for a male applicant, which I thought was brilliant.
There have been instances where I have looked at a job description and assumed the employers would prefer a male candidate over a female.
Give back to the community
Share your work with the world and contribute to the development of open source packages as well. This way you are also giving back to the AI community.
Everytime I work on a mini-project, I share it on GitHub. It is a great way to present your work to the world. A few months back, Chatspace AI published an integration called “Covid-19 Public Virtual Assistant” that would auto-reload FAQs related to Covid-19 from reliable sources like WHO into a dialog flow agent. I was happy to be a contributor to this integration.
The AI community needs a large, inclusive, and a diverse support system to bring about a significant change and minimise the skewness in data. Inspiring, encouraging, and mentoring newcomers to take up AI, and helping them understand the repercussions of biased AI and why we need a change is extremely critical.
We need some overcorrection in the field of AI. Acknowledging that biases exist, understanding their negative implications, and fixing them are all vital so that people can use AI-powered technologies in a secure manner.
Girija Shingte is a data scientist at Chatspace, the Irish based AI for enterprise software company