Why the future of machine learning research lies in Africa

Over the past few years, machine learning applications have blossomed and the community has grown rapidly. In Africa, people are applying this technology to challenges such as food security and healthcare.

We met Charity Wayua in a hotel lobby in Tangier, Morocco. When she talked about her tortuous journey to Tangier for the technology and innovation conference in the city, the cheerful woman laughed at herself. Her journey started from Nairobi, Kenya, where she leads one of IBM's two major African research centers. In order to get to Tangier, she had to fly to Dubai first, stay briefly, return to Casablanca, and then take a car to her destination, which took another three and a half hours. The flight that would have been a seven- or eight-hour flight with a direct flight turned into a nearly 24-hour journey. She said this was nothing new to her.

The difficulty of traveling to Africa is not the only difficulty facing the African research community. In fact, leaving the region is also difficult, which makes scientists from the region rarely participate in international conferences and exchanges. While these problems have a negative impact on every field of research, they are more destructive to AI research. Travel difficulties mean that African scientists have repeatedly missed various international conferences due to visa issues. They also have difficulty attending the world's largest AI events held in the United States and Canada. The lack of communication can easily cause researchers in the region to fall behind their peers in other parts of the world.

Despite these disadvantages, Africa's machine learning community has flourished in the past few years. In 2013, a group of local industry practitioners and researchers launched an annual workshop to share resources and exchange ideas with each other - Data Science Africa. In 2017, another group founded the Deep Learning Indaba organization, which has now established chapters in 27 of the 54 countries on the African continent. In addition, as the demand for AI increases, many other university courses and educational programs dedicated to machine learning education have also rapidly developed. Africa

's efforts in the field of artificial intelligence have attracted the attention of the international community. At the end of 2013, IBM Research opened its first African office in Nairobi, and later opened a second African office in Johannesburg, South Africa. Earlier this year, Google opened a new artificial intelligence lab in Accra, Ghana, and next year, ICLR, a major artificial intelligence research conference, will be held in Addis Ababa, Ethiopia.

For many years, artificial intelligence research has been concentrated in a limited number of fields, and its applications lack diversity and are often disconnected from the real world. The expansion of artificial intelligence in Africa can bring about a positive change. Currently, many academic institutions and corporate research labs that dominate artificial intelligence research are concentrated in wealthy areas such as Silicon Valley and Beijing's Zhongguancun, where research objects are concentrated and innovation bubbles are serious. These research institutions limit artificial intelligence products to a limited range. In contrast, Africa may provide an environment for artificial intelligence to return to its original intention. Here, hunger is rampant, disease is rampant, and poverty is ubiquitous. Artificial intelligence technology can deal with such urgent global challenges.

Charity Wayua leads IBM's research team in Nairobi, Kenya

"For those who are brave enough to challenge themselves, this is the place to go," Wayua said.

Africa's innovation model
IBM Research's offices in Kenya and South Africa, as well as Google's artificial intelligence lab in Ghana, share the same mission as their parent companies: basic and cutting-edge research. They focus on issues such as increasing affordable health care for everyone, making financial services more inclusive, strengthening long-term food security, and streamlining government operations. These missions may seem no different from those of labs anywhere else in the world, but the external environment has made the goals of these missions slightly different.

“研究不能脱离它最终应用的环境，”谷歌加纳人工智能实验室主管Moustapha Cisse说。“非洲是一个在很多方面都存在独特挑战的环境，在这里，我们有机会探索其它地方的研究人员无法接触到的问题。”

比如，在加纳建立人工智能研究室之前，谷歌就已经开始与坦桑尼亚的农民合作，了解他们在维持粮食生产方面面临的一些困难。研究人员了解到，农作物的疾病会显著降低农产品的产量，因此，他们构建了一种机器学习模型，可以用它来诊断木薯植物早期的疾病，木薯植物是该地区重要的农作物之一。农民可以直接在手机上运行这个模型，而无需访问互联网，这使得他们可以在疾病早期就开始干预，从而挽救自己的农作物。

Wayua shared another example. In 2016, the Johannesburg team at IBM Research found that it took four years for cancer data to be notified to the country's health policy department after being diagnosed in the hospital. In the United States, this data collection and analysis process only takes two years. The researchers found that the time lag was partly due to the unstructured nature of hospital pathology reports. Human experts need to read each case and classify them into 42 cancer types based on the specific circumstances, but the unstructured nature of pathology reports makes this process very time-consuming. Therefore, the researchers began to study a machine learning model that can automatically label pathology reports. They successfully developed a prototype system in two years and are now working to optimize it so that it can be applied in practice.

"Technology is only one side of the solution," Wayua said. "The other side is how to understand the problems we face and objectively define them in a scientific and engineering way."

Once a research project moves to the practical stage that can be used in the real world, it will immediately face another difficult problem: how to get support from the target users. "Building relationships with users and the real world is really important in driving change," said Wayua. It’s easy to design a perfect system without thinking about how to collect data, but that perfection means nothing if no one wants to use it. “It’s those relationships you build with your users that help you understand why something you implemented didn’t really work,” she adds.

Responding to user needs can also help drive fundamental advances in technology. For example, Google’s Ghana AI lab is working to improve natural language understanding for Africa’s roughly 2,000 languages. “Africa is the most linguistically diverse region in the world today,” Cisse says. “There’s a lot we can learn from it and build on it.”

Cisse
and Wayua have similar career trajectories. They both left Africa for higher education, hoping to make the most of their skills to make an impact in Africa. Cisse previously worked at Facebook in Europe and had been waiting for the right opportunity to return to Africa to serve his family.

Now, both are putting a lot of energy into developing more local educational opportunities for young people interested in artificial intelligence. Cisse founded and mentors a master's program in machine intelligence, an intensive one-year course that, in addition to local courses in Africa, attracts some of the best AI researchers in the world. Wayua's lab hires high-achieving undergraduates to work with its full-time staff and supports them in the online master's program in computer science offered by Georgia Tech. "The

most important resource for doing research is talent, and you'll find more talent in Africa than anywhere else in the world," Cisse said, noting that there are particularly many young people in Africa. "There are so many scientific and technological resources here, and the question is how do you get these smart young people to learn really useful science so that they can change the continent and create their own bright future?"

When Cisse mentors his students in the master's program, he tells them that in five years they will be leaders in the field and good enough to teach others, and he is confident of this.

"The future of machine learning research is in Africa," he said, "whether people know it or not, this will become an inevitable fact."