Google has launched a new centre in Accra, Ghana, that will allow African-based researchers to shape Artificial Intelligence (AI) to solve African problems.
“AI is a critical tool used today and used to accelerate all sorts of sciences in physics, chemistry and engineering. But most of the people working and advancing the science and developing it in the field are based mostly in Western countries. It’s important that such an important field [can address] a diversity of the problems that the world faces today, and Africa is accurately represented,” said Moustapha Cisse, Head of Google AI.
In just seven short years, Google's AI technology has progressed at breakneck speed. Take the example of identifying objects in images: In 2012, AI programmes made mistakes with 26% of the objects. These days, the AI error rate is on average 3% - which is better than humans (5%).
“We’ve managed to train machines to identify pixels, and they can transcribe audio and translate languages for us. We want all people in Africa to benefit and access this information,” said Cisse.
The new centre plans to help address problems that farmers experience. They plan to integrate technology already in existence which help Ugandan farmers identify sick Cassava plants by taking a photo of it. The app, which works offline and in rural areas, identifies sickness and can recommend treatments. Making apps like these accessible across the continent could help greatly with food security.
Other projects are in healthcare, and getting AI to count populations and predict population migration using satellite imagery.
But the new centre will have to address the way Africans access the internet: primarily through low-performance cellphones.
Which is why among the projects the centre is looking at is finding ways of dumbing down AI - which runs complex algorithms on massive data sets - so that it can run on basic cell phones.
Google also wants to make the AI available offline, to compensate for those who struggle to get their internet once or twice a day. Limited access to connectivity and high data costs is something that requires even further simplification of AI as most Africans use phones with 512mb of storage.
Another challenge: developing AI to understand the 2,000 African languages spoken on the continent.
“Africa is linguistically the most diverse place on this continent. It would be nice if people could communicate beyond these language barriers. Many Africans speak up to three languages, but if we could have machines that can speak hundreds of languages imagine who would be able to speak to the 2.4 billion people projected to be living on the continent in the next 30 years.”
Also from Business Insider South Africa: