Leading a movement to strengthen machine learning in Africa
Avishkar Bhoopchand† a research engineer on the Game Theory and Multi-agent team, shares his journey to DeepMind and how he’s working to raise the profile of deep learning across Africa.
Find out more about Deep Learning Indaba 2022, the annual gathering of the African AI community – taking place in Tunisia this August.
What’s a typical day like at work?
As a research engineer and technical lead, no day is the same. I usually start my day by listening to a podcast or audiobook on my commute into the office. After breakfast, I focus on emails and admin before jumping into my first meeting. These vary from one-on-ones with team members and project updates to diversity, equity, and inclusion (DE&I) working groups.
I try to carve out time for my to do list in the afternoon. These tasks could involve preparing a presentation, reading research papers, writing or reviewing code, designing and running experiments, or analyzing results.
When working from home, my dog Finn keeps me busy! Teaching him is a lot like reinforcement learning (RL) – like how we train artificial agents at work. So, a lot of my time is spent thinking about deep learning or machine learning in one way or another.
How did you get interested in AI?
During a course on intelligent agents at the University of Cape Town, my lecturer demoed a six-legged robot that had learned to walk from scratch using RL. From that moment on, I couldn’t stop thinking about the possibility of using human and animal mechanisms to build systems capable of learning.
At the time, machine learning application and research wasn’t really a viable career option in South Africa. Like many of my fellow students, I ended up working in the finance industry as a software engineer. I learned a lot, especially around designing large scale, robust systems that meet user requirements. But after six years, I wanted something more.
Around then, deep learning started to take off. First I started doing online courses like Andrew Ng’s machine learning lectures on Coursera. Soon after, I was fortunate enough to get a scholarship to University College London, where I got my masters in computational statistics and machine learning.
What’s your involvement in the Deep Learning Indaba?
Beyond DeepMind, I’m also a proud organizer and steering committee member of the Deep Learning Indaba, a movement to strengthen machine learning and AI in Africa. It started in 2017 as a summer school in South Africa. We expected 30 or so students to get together to learn about machine learning – but to our surprise, we received over 700 applications! It was amazing to see, and it clearly showed the need for connection between researchers and practitioners in Africa.
Since then, the organization has grown into an annual celebration of African AI with over 600 attendees, and local IndabaX events held across nearly 30 African countries. We also have research grants, thesis awards, and complementary programs, including a mentorship program – which I started during the pandemic to keep the community engaged.
In 2017, there were zero publications with an African author, based at an African institution, presented at NeurIPS, the leading machine learning conference. AI researchers across the African continent were working in silos – some even had colleagues working on the same subject at another institution down the road and didn’t know. Through the Indaba, we’ve built a thriving community on the continent and our alumni have gone on to form new collaborations, publishing papers at NeurIPS and all of the major conferences.
Many members have gotten jobs at top tech companies, formed new startups on the continent, and launched other amazing grassroots AI projects in Africa. Although organizing the Indaba is a lot of hard work, it’s made worthwhile by seeing the achievements and growth of the community. I always leave our annual event feeling inspired and ready to take on the future.
What brought you to DeepMind?
DeepMind was my ultimate dream company to work for, but I didn’t think I stood a chance. From time-to-time, I’ve struggled with imposter syndrome – when surrounded by intelligent, capable people, it’s easy to compare oneself on a single axis and feel like an imposter. Luckily, my wonderful wife told me I had nothing to lose by applying, so I sent my CV and eventually got an offer for a research engineer role!
My previous experience in software engineering really helped me prepare for this role, as I could lean on my engineering skills for the day to day work while building my research skills. Not getting the dream job right away doesn’t mean the door’s closed on that career forever.
What projects are you most proud of?
I recently worked on a project about giving artificial agents the capability of real-time cultural transmission. Cultural transmission is a social skill that humans and certain animals possess, which gives us the ability to learn information from observing others. It’s the basis for cumulative cultural evolution and the process responsible for expanding our skills, tools, and knowledge across multiple generations.
In this project, we trained artificial agents in a 3D simulated environment to observe an expert performing a new task, then copy that pattern, and remember it. Now that we’ve shown that cultural transmission is possible in artificial agents, it may be possible to use cultural evolution to help generate artificial general intelligence (AGI).
This was the first time I worked on large-scale RL. This work combines machine learning and social science, and there was a lot for me to learn on the research side. At times, progress towards our goal was also slow but we got there in the end! But really, I’m most proud of the incredibly inclusive culture we had as a project team. Even when things were difficult, I knew I could rely on my colleagues for support.
Are you part of any peer groups at DeepMind?
I’ve been really involved with a number of diversity, equity, and inclusion (DE&I) initiatives. I’m a strong believer that DE&I in the workplace leads to better outcomes, and to build AI for all, we must have representation from a diverse set of voices.
I’m a facilitator for an internal workshop on the concept of Allyship, which is about using one’s position of privilege and power to challenge the status quo in support of people from marginalized groups. I’m involved in various working groups that aim to improve community inclusion amongst research engineers and diversity in hiring. I’m also a mentor in the DeepMind scholarship programme, which has partnerships in Africa and other parts of the world.
What impact are you hoping DeepMind’s work can have?
I’m particularly enthusiastic about the possibilities of AI making a positive impact on medicine, especially for better understanding and treating diseases. For example, mental health conditions like depression affect hundreds of millions of people worldwide, but we seem to have limited understanding of the causal mechanisms behind it, and therefore, limited treatment options. I hope that in the not too distant future, general AI systems can work in conjunction with human experts to unlock the secrets of our minds and help us understand and cure these diseases.