In 2018, Google launched its first African AI Centre in Accra, Ghana with the aim of fostering the continent’s growing capabilities in the field.
It also aimed to deliver innovative solutions that will benefit both Africa and the world, working with local universities and research centres as well as working with policymakers on the potential use cases of AI in Africa.
Since its launch, the team at the Accra-based research centre has engaged in significant AI work that includes mapping Africa’s built environment using satellite imagery and machine learning.
This, Google said, has allowed quadrupling the number of African buildings on Google Maps, bringing the number to over 250 million.
“Our mission with this first African AI research centre is to push the boundaries of AI while solving pressing problems affecting millions of people both locally and globally,” the tech giant said.
“This includes mapping buildings in remote locations to provide better electricity.”
Below are six AI projects that the Accra research centre is working on and how they are hoping to make a difference.
Mapping buildings
To map buildings in remote locations can be difficult even with satellite imagery.
“When these buildings go unmapped, it can make things like planning infrastructure difficult,” Google said.
The Accra research centre has launched the Open Buildings dataset project that combines AI with satellite imagery to pinpoint the location of buildings.
It seeks to help governments and nonprofit organisations understand the needs of residents and offer assistance.
“In Uganda, for example, the nonprofit Sunbird AI is using the dataset and working with the Ministry of Energy in Lamwo district to study villages’ electrification needs, and plan potential solutions, such as prioritising electricity in important areas like commercial centres,” Google noted.
“We’re continuing to expand our Open Buildings dataset to see how it can help communities in more areas.”
In addition to various countries in Africa, the dataset now covers 16 countries in Southeast Asia, including Bangladesh and Thailand.
Forecasting floods
The United Nations has reported that half of the world’s least-developed countries lack adequate early warning systems for disasters, including floods.
“In West and Central Africa specifically, where flooding can be severe, early warning systems could enable better preparation and potential evacuation,” Google said.
The company is using AI models to predict when and where riverine floods will occur in 80 countries worldwide, including 23 in Africa.
Google’s Flood Hub platform displays forecasts up to seven days in advance, with detailed inundation maps that show different water levels predicted in different areas.
In this way, people know what to expect where they live.
Predicting locusts
The Google AI Centre has collaborated with AI-product-focused company InstaDeep as well as the UN’s Food and Agriculture Organisation (FAO) to better detect locust outbreaks and enable farmers to implement control measures.
The AI centre team is currently working on building a model that forecasts locust breeding grounds using historical data from the FAO and environmental variables like rainfall and temperature.
Improving maternal health outcomes with ultrasound
In recent years, sensor technology has evolved to make ultrasound devices significantly more portable and affordable as they can be crucial for identifying potential complications during pregnancy.
Globally, the AI centre has been working on building AI models that can read ultrasound images and provide important information to healthcare workers.
In Kenya, for instance, the centre has partnered with Jacaranda Health to help improve ultrasound AI technology, with a focus on using handheld ultrasound devices that don’t need to be attached to larger machines.
This, Google said, can help people who aren’t trained to operate traditional ultrasound machines to acquire and interpret ultrasound images and triage high-risk patients, simply by sweeping the handheld probe across the mother’s belly.
Helping people with non-standard speech make their voices heard
The centre has built Project Relate, an Android app that uses AI research, to help people with non-standard speech communicate more easily.
“After recording 500 phrases, users receive a personalized speech recognition model,” Google said.
Now available for user testing in Ghana, it can transcribe speech into text, use a synthesized voice to repeat what the speaker has said and engage Google Assistant to complete tasks, such as asking for directions or playing a song.
Teaching children how to read
According to Google, due in part to the effects of COVID-19, it’s estimated that about two-thirds of 10-year-olds globally are unable to read and understand a simple story.
To address the tech giant has developed Read Along, its AI-based reading tutor app and website, which is helping to increase child literacy.
Diya, the in-app reading buddy, listens to the speaker reading aloud, offering support when they struggle, and rewards them when they do well.
Over the past three years, more than 30 million kids have read more than 120 million stories on Read Along.
Google noted that progress helps children, but it also affects their families.
“For example, one of our Lagos users, William, began using the app when he was 10 years old,” the company said.
“He went from being able to read for three minutes at a stretch to reading for 90 minutes at a time. I am more confident about William’s future because he can read well,” said William’s mom, Martha, “Not just reading well, he now loves to read.”
