From L-R: Vector Faculty Member Toniann Pitassi, Vector Interns Irene Kameni, Kobby Panford-Quainoo, Fadel Thior, Ines Birimahire, Aisha Alaagib, and Vector Graduate student Elliot
By Ian Gormely
Eighteen months ago, Fadel Thior had never worked with AI algorithms. Today, he is part of a growing community of AI experts in Africa.
The massive levelling up came courtesy of a new AI master’s program on the continent. The Senegal native was part of the inaugural cohort of graduates from the African Master’s in Machine Intelligence program (AMMI) that brought some of Africa’s top AI talent, as well as Vector’s Research Director Richard Zemel, to the Quantum Leap Africa Center of Excellence at the African Institute for Mathematical Sciences (AIMS) in Kigali, Rwanda for a one-year intensive program. “You learn a lot in 10 minutes,” says Thior. “They tell you something and it can take a week to digest.”
Thior was a year into his master’s in computer science when his interest in AI, specifically natural language processing, was piqued by an intro to AI course. He and two friends, who also lacked any AI experience, began building a question and answer chatbot from scratch, teaching themselves as they went along. “We started at the beginning and it was very complicated,” he says. “We didn’t know any of the frameworks that people usually use. We just knew coding.”
AI has the opportunity to put Africa “at the forefront of the Fourth Industrial Revolution,” according to a recent article from Brookings. IT hubs, like Kenya and South Africa are branching out into AI says Zemel, who was a guest lecturer at AIMS, while new centres are being established in Ghana and Rwanda. “The group Black in AI does a terrific job organizing and mobilizing the efforts,” he says. “And there are some great get-togethers known as Indabas, which are a kind of mashup of a conference, expo, and hackathon.”
Yet challenges remain. The same Brookings story points to problems with digital connectivity, as well as a lack of regulatory systems and technical skills as barriers to broader AI expansion.
“There’s a lot of very talented people who are very excited about AI,” confirms Thior. “But they don’t have the opportunity to do it because we don’t have an AI program in Senegal or even Africa other than AMMI.”
Thior’s work with his friends was put on pause when he became one of just 31 students from over 1,000 applicants that were accepted to AMMI, whereMoustapha Cisse, head of Google AI in Accra, Ghanna, is director. There he met Zemel, along with Vector Faculty Member Toniann Pitassi and graduate students David Madras and Elliot Creager who were all guest lecturers at the program. “AMMI is a great initiative that’s bringing a new stream of African talent into the field,” says Zemel. “It is very difficult for Africans to apply and become part of the grad student pipeline. Vector, the University of Toronto and other Canadian universities can help by developing a more regular internship program.”
Impressed by the cohort of students he and Pitassi taught (he describes them as some of the “best and brightest recent graduates from across Africa”), Zemel offered research internships at Vector to six of the students, under the supervision of some of Vector’s Faculty Members. “The things that people work on for two or three months in university, we were doing in three weeks,” recalls Thior, who was paired with Anna Goldenberg. “All the lecturers were very impressed [by the students].”
Like Thior, Vector intern Aisha Alaagib came to AI in a circuitous manner. Her interests lie in cryptography and she became interested in applying machine learning (ML) to that field while doing her Master’s in mathematics, yet had no background in ML. She says her time at AMMI, where she became interested in the integration of computer vision and NLP, and Vector, where she has pursued work in multi-modal representation learning with Sanja Fidler, has opened her eyes to the technology’s transformative possibilities. “AI forces us to rethink everything about our life,” she says. “How does AI affect health care, privacy, our data?”
They also both point to the extended research community at Vector for both helping to expand their understanding of ML’s capabilities and for helping them acclimatize to their new environs. “What I got from Vector was not just learning machine learning or doing research,” says Alaagib. “There’s a community, for both work and for fun.”
With their internships at Vector now over, both Thior and Alaagib are looking to gain more experience abroad before heading home. Alaagib co-authored a paper that was accepted at last year’s NeurIPS and Thior already has another internship on deck, this one at Vector’s sibling Institute, Mila in Montreal. And both hope to give back to the burgeoning AI community in Africa either through returning to AIMS (which Alaagib will do as a tutor and researcher) or developing their own AI courses. “We have the mathematical background, we have the coding background,” says Fadel. “So we just need to be initiated.”