Oscar Darko-Sarfo sounds smitten by his latest job as a barber — an enormous break for a 22-year-old who once doubted he would find work. He was born with a cleft palate, which impairs his ability to talk clearly; it’s not at all times easy for listeners to grasp him.
Darko-Sarfo is one in all 20 Ghanaians testing out artificial intelligence-enabled technology developed by Google to assist individuals with non-standard speech. After speaking in English right into a smartphone trained to recognise his speech patterns, a female, American-accented voice broadcasts: “I’m working as a barber. I wish to do Afros.”
The app, still in prototype, has improved Darko-Sarfo’s ability to speak, he says — and with it his confidence. In addition to acquiring a job, he has recently found a girlfriend.
Google’s automatic voice recognition technology, called Project Relate, is a tiny example of how AI might help tackle problems in what researchers confer with as “low-resource settings”.
AI is being marshalled in multiple fields on the African continent, which incorporates a number of the poorest countries on the planet: in Zambia, to assist improve medical diagnostics; in Kenya, to enable farmers to discover crop disease; and in Ethiopia, to tailor education materials to pupils’ needs.
In the 18 months since AI became front-page news, there was extensive discussion of how it should change society: its effect on jobs, on elections, on science, even what it means for the long run of humanity. That conversation has centred largely on the wealthy countries where it has been developed.
Darko-Sarfo’s case illustrates lots of the probabilities this latest technology offers for Africa and in developing nations more generally, but additionally the pitfalls. On one hand, an AI-powered device has helped him overcome his speech issues in a rustic with limited health resources and only a handful of speech therapists. On the opposite, Google’s technology isn’t yet available in his first language, Twi Asante. There is not any guarantee it should be widely adopted.
Gifty Ayoka, a Ghanaian speech therapist, whose NGO Talking Tipps is testing the technology along with University College London’s Global Disability Innovation Hub, is cautiously optimistic concerning the role AI can play in her country and others prefer it. “If we are able to truly embrace this, it should make things easier for people,” she says.
But she also worries that Ghana’s state and society won’t give you the option to properly absorb a technology developed abroad. “Without awareness and training, plus local language and cultural support, no application — nonetheless clever — goes to be useful,” she says.
Proponents argue that AI might help poorer societies “leapfrog” whole phases of development in the identical way that many countries, lacking landline infrastructure, enthusiastically adopted mobile phones within the early 2000s. Once handsets spread, latest innovations followed, allowing people to make use of their devices for all the pieces from financial transactions to paying for access to solar energy.
“Well-run digital systems make states more capable,” said Microsoft co-founder Bill Gates in an interview with the Financial Times last 12 months, by which he extolled the virtues of developing nations embracing the digital revolution.
To optimists, AI presents a once-in-a-generation opportunity to go one step further. Machine learning, they argue, can turbocharge the leapfrogging phenomenon by putting revolutionary tools into the hands of people, businesses and states.
“I consider within the transformational effect of tech,” says Yordanos Asmare, an Ethiopian who’s head of talent at A2SV (Africa to Silicon Valley), a US-based impact incubator that seeks to develop African AI capabilities.
But many, including Asmare, worry AI could have precisely the alternative impact, amplifying some great benefits of richer nations which have more computational power and computer engineers, and leaving other countries trailing.
Bright Simons, co-founder of Imani Centre for Policy and Education, an Accra-based think-tank, says Africans produce lower than 0.5 per cent of machine learning and large-language models. “We’re already starting out at a much lower base in comparison with the remaining of the world,” he says. “The AI leapfrogging theory completely dissolves, because in AI you’re already marginalised.”
James Manyika, Google’s senior vice-president for technology and society, who grew up in Zimbabwe, says that the risks of falling behind make it all of the more vital for the continent to forge ahead. “AI presents a possibility too significant to disregard for Africa,” he says.
In May, Microsoft president Brad Smith gave a speech in Nairobi to mark a $1bn investment in Kenya. In combination with G42, an Abu Dhabi AI company, Microsoft is constructing what it calls a “comprehensive package of digital investments” including a geothermal-powered data centre and an innovation lab.
AI, Smith told his audience, was a technology comparable to the printing press and electricity. And while Africa partly missed out on previous transformative technologies — 142 years after Thomas Edison’s invention, 43 per cent of Africans still lack access to electricity — the identical needn’t occur with AI, he argued. “One must bring three things together: human capital, financial capital and technological innovation.”
FT Edit
This article was featured in FT Edit, a each day choice of eight stories, hand-picked by editors to tell, encourage and delight. Explore FT Edit here ➼
While the challenges of making an AI “technology stack” — data centres, AI models and developers — were considerable, he said, they were achievable. And AI played to the strengths of a continent where the median age is nineteen, the youngest on the planet.
Google has also made an AI push within the region, opening an AI research centre in Kenya in 2022 so as to add to the one it built across in Ghana in 2018, which was the primary of its kind in Africa. Seeking a business opportunity, access to talent and a likelihood to influence how a continent that can have 2.5bn people by 2050 engages with the technology, it has also committed to spend $1bn on digital infrastructure. Investments include a cable linking several east and central African countries to Australia and one other from Portugal to South Africa.
At Google’s AI research centre in Accra, where Jason Hickey leads a team of software engineers recruited from across Africa, the mood is upbeat. “This is a transformative technology,” says Hickey. On the sixth floor of a rented office constructing, Google software engineers are working on AI-powered tools to, amongst other things, predict famines and map buildings, including informal settlements that will be invisible to satellites.
MohamedElfatih MohamedKhair, a software engineer from Ethiopia, explains how Google is constructing an AI model to watch flash floods in real time. Parts of Africa have a paucity of weather radars; to deal with the issue, Google has designed a model that mixes data from two varieties of satellites.
“We use AI to fill within the gaps. That’s our hack,” says Hickey. Far from being a bargain-basement version of better-resourced systems, he says, the AI model beats standard methods. “Yes, we would like to do it on the low-cost. But we’re actually capable of recover forecasts from AI.”
One area where experts see big potential advantages is education, each through interactive learning and, more broadly, by teaching in relevant languages. There are a minimum of 1,000 languages in Africa, yet, because of the legacy of colonial history and the languages by which educational materials are produced, many students learn of their second or third tongue, putting them at an enormous drawback.
“In Africa, people speak their language at home but learn all the pieces in English or French. All their images and books are western,” says Simons, the Ghanaian think-tanker. “So how can we use natural language processing to generate more local content, to create textbooks in local languages and to record oral history?”
A2SV, the impact incubator, has developed an AI-powered “personalised learning” tutoring programme called SkillBridge, which uses Amharic and Afan Oromo, the 2 most generally spoken languages in Ethiopia, to teach pupils taking the country’s university entrance exam. Of nearly 675,000 students who sat the test this 12 months, only 5.4 per cent passed.
“It raises plenty of questions around the standard of education, but additionally about what are you going to do with all these young people and concerning the productivity of the country,” says A2SV’s Asmare. The hope is that SkillBridge can improve results.
Healthcare is one other area with potential, argue AI enthusiasts. In many countries, ultrasound equipment is just too expensive and trained sonographers are in brief supply. Ninety-five per cent of pregnant women in Africa don’t have any access to scanning, in line with Angelica Willis, a Google AI researcher.
The company has developed an AI model to analyse data collected from cheaper portable ultrasound equipment deployed by novice operators. A study in Zambia found that AI could assess gestational age and foetal malpresentation from “blind sweep ultrasounds” to the identical standard as trained sonographers using standard equipment. If it were adopted more widely, this might save many lives.
While AI guarantees much, some fear that the technology — like many before it — couldn’t only reflect existing global inequalities, but exacerbate them. “We call it technology amplification theory,” says Catherine Holloway, a professor of interaction design and innovation at UCL.
Google’s Manyika, who can be co-chair of the UN’s high-level advisory body on AI, acknowledges the danger of what he calls an “AI divide”. “AI is computer-intensive and we all know that device and data costs (in Africa) are the very best on Earth,” he says.
Simons argues there are limits to tech’s ability to patch up dysfunctional governance systems. “For 80 per cent of problems, AI will probably be the cherry on the highest,” he says. But in countries that lack a functioning health system, decent roads and power, or a fairly accountable state, there are few shortcuts, he says. “Nobody needs cherries in the event that they don’t have already got the cake.”
Simons also points to familiar anxieties — the threat to data privacy and ownership and the chance that AI could possibly be exploited by bad actors, including states. In 2020, Ghana, regardless that it has a thriving democracy, purchased 10,000 security cameras from China as a part of a face-recognition system. However creatively civil society uses AI, Simons says, “tyranny (could possibly be made) much easier”.
Some campaigners are also concerned a couple of potential “data grab” by which just a few big, almost exclusively US, corporations are collecting data from Africa and constructing tools that they’ll sell elsewhere. Campaigners have pointed to what they are saying is the exploitation of African labour in data centres, where people working for as little as $2 an hour are training AI models for giant tech firms.
“It could possibly be Ghanaians feeding ChatGPT, for instance, and that data is within the US,” says Asmare, who joined A2SV after moving from Ethiopia to check at Stanford and work in Silicon Valley, and who now lives within the US. “Where does that data go? Are governments interested by that?”
There can be the difficulty of Africa’s astonishingly wealthy number of languages. Until now, most large language models have been built using English databases, with all of the built-in inequalities that means. Some African languages are primarily oral with little digital text, equivalent to Wikipedia entries, to feed into natural learning processing models.
SEE Africa, a Tanzanian NGO, has designed a voice-enabled interactive app to assist rural women navigate the business opportunities of growing a highly nutritious variety of sweet potato. It needed its app to work in Swahili. Yet although the language is spoken by some 200mn people across east Africa, there aren’t sufficient data sets for AI models to work on, says EM Lewis-Jong, director of Common Voice, an open-source resource funded by the Mozilla Foundation, a non-profit group working to keeping the web accessible.
“The Anglo-centrism of the web is a form of colonialism in one other form,” she says. “Heritage, culture, the actual features of your existence, are so deeply intertwined with language,” she says. Even something as apparently universal because the thumbs-up sign fails to cross linguistic boundaries: in several African languages, it will probably be an insult.
So Common Voice is harnessing AI to tackle the difficulty. Individuals or organisations equivalent to SEE Africa can upload as much as 1,000 hours of speech in their very own language (or combination of languages) on to its platform, either reading out sentences or obtaining copyright-free material from local radio stations and the like to teach natural language models.
“Speech technology has huge potential to be impactful in low-literacy communities,” Lewis-Jong says. “AI could genuinely make language a bridge and never a barrier.”
Equiano Institute, a South African AI think-tank, has ambitions to construct an African-owned and operated large language model. “Just as ChatGPT provides an LLM model, we would like to offer that solution in Africa,” says Daniel Akinmade Emejulu, who chairs Equiano’s advisory board.
Google, too, is constructing large data sets in several African languages from its Ghana research centre. “Earlier this 12 months, we added one other 110 languages, of which a couple of third were African,” says Google’s Manyika.
Despite the challenges, Mayinka says he stays convinced that AI offers a novel likelihood for poorer countries to “level up” by flattening access for people and businesses to tools and knowledge, and by providing states with the means to tackle big societal challenges.
“I fundamentally consider that AI represents a possibility not just for the world,” he says, “but for developing regions like Africa particularly.”
Big firms equivalent to Google, Microsoft and Amazon, which says it’s investing $1.7bn to expand cloud and AI services in Africa, will proceed to take a position within the continent — albeit with what are, by their standards, relatively modest amounts.
What impact AI has, nonetheless, says Simons of Imani Centre, won’t ever be a purely technological problem, but an issue of the way it is absorbed by states. Technology, he says, can only improve people’s lives if it will probably cross the “interface” between cyber space and real-life, from what he calls “the world of bits to the world of atoms”.
Hickey cites the instance of nations deploying Google’s AI weather forecasts. “Ultimately, we are able to provide the knowledge. The final step of engagement with farmers or people who find themselves affected by floods must be done through effective engagement with the community,” he says.
If governments in wealthy countries are struggling to grapple with the implications of AI, still less legislate for it, in developing societies the situation is even tougher. Simons says that unless states are using AI themselves, they can not hope to grasp or regulate it.
Ayoka, the banker turned speech therapist, is frustrated at what she sees as a scarcity of ambition from Ghana’s government despite its digital-first rhetoric. “All this tech is so advanced, but our government is sleeping.”
Many African governments talk a superb game about being open to AI, agrees Asmare at A2SV, but she would really like to see more strategic pondering. “My query is, do you could have the security guardrails to ensure your people aren’t exploited, and that these investments are in step with what the country and the people need?”
Google’s Manyika argues that Africans are best-placed to steer use of the technology of their countries. “What Africa does with AI must be Africa-led,” he says.
