An employee at a tech start-up works on her computer at her office in Nairobi, Kenya on October 30, 2018. AI-related barriers are seen, starkly, in Africa.
| Photo Credit: REUTERS
Artificial intelligence (AI) is poised to revolutionise how we live today. The world over, we see examples of its transformative impact not only on healthcare, education, agriculture and other areas, but also in sparking creativity by making it easier and faster for people to bring their ideas to life.
Realising this new technology’s full potential, though, requires addressing several challenges causing significant barriers and imbalances across many parts of the world. We should know in Africa because many of those challenges are playing out here.
For example, it is widely known that datasets from the Global South are largely underrepresented in AI training datasets, leading to inefficient application of these AI tools in contexts they are not built to understand or represent.
There is also a linguistic imbalance in the development of AI solutions, with a significant focus on highly resourced languages like English.
Africa is home to about 2000 languages, with 75 of them having more than one million speakers each. Yet, this linguistic abundance is not reflected in the large language models (LLMs)—which are at the core of many AI systems—that are widely used. This creates a significant barrier in trying to build localised solutions.
Also Read | AI beds to remote ICUs: How startups are plugging India’s health gaps
Overcoming those issues is complicated by continent grappling with a glaring skills gap in AI expertise. A pronounced digital divide further compounds this gap. Therefore, we’re often beholden to tools built by and for Western contexts being applied in the Global South, further perpetuating bias and inefficiencies.
Infrastructure is also an issue. Computer and data storage are Africa’s most resource-intensive inputs for developing AI solutions.
The cost of using existing LLMs at scale is relatively high, especially for non-profits, and building and using self-built models would also require significant investment in computing resources. Limited computing power and resources hinder even the best-funded organisations from innovating.
There are important local initiatives, though.
Ushahidi, a global non-profit organisation, has been focused on surfacing community-generated data to drive impact across various sectors using open-source technology for the last 16 years in over 160 countries and engaging with disenfranchised communities to play active roles in realising the future we want to see.
Insights from local communities and their collective intelligence help to provide a richer understanding of some of the world’s most pressing issues. And, with AI, it will be no different.
Perpetuating bias
The tech community in Kenya has already leveraged generative AI as a tool for civic education. Several AI-powered tools have been created to help local communities interact with and understand the implications of proposed laws such as a controversial finance bill this year. One software engineer created a GPT tool based on the Kenyan auditor general’s reports to expose corruption scandals and interrogate Kenya’s public debt.
The local development research institute has been collecting data from local farmers and combining it with satellite imagery to predict and estimate yields and crop stress. It then delivers personalised recommendations to these farmers to enhance their productivity and mitigate risks against climate.
Ushahidi has partnered with Dataminr to leverage AI to improve data management process by launching models that facilitate automatic categorisation, geolocation, and data translation. That helped analyse millions of tweets during Kenya’s nationwide protests against a new finance bill in June.
Bridging the digital divide is a priority of the Sustainable Development Goals, and there is global recognition of the role that digital technology tools will play in creating a healthier, more inclusive, and sustainable planet by 2030.
Need to fill data gaps
As the world is taken by storm by emerging technologies such as AI, if we do not pay attention to filling data gaps, growing AI talent, and investing resources for computing in the global south, these AI systems will continue to perpetuate bias with disproportionate negative consequences on disenfranchised communities.
Also Read | AI’s dark secret: It is rolling back progress on equality
Ushahidi’s work has been a constant reminder of technology’s critical role in facilitating social impact. However, that social impact would not be possible without people leveraging technology in the first place. Projects such as Masakhane, Lelapa AI, and GhanaNLP continue to champion efforts to deploy datasets for African low-resource languages and integrate them into AI solutions.
Others, such as Data Science Africa, are providing quality training in machine learning, data science and more in a bid to encourage the development of African solutions to African problems.
It’s also promising to see several African countries collaborating with the public and private sectors to develop national AI strategies that could help tackle these challenges from a policy level, grounded in African values and lived realities.
At a world level, UN member states’ adoption of the Global Digital Compact speaks to global recognition and commitment to ensuring that digital technologies benefit humanity equitably. But there is much more to be done, so the work to ensure that AI is inclusive, responsible, trustworthy, and accessible to all must continue. We can’t afford to let anyone be left behind.
This article first appeared on Context, powered by the Thomson Reuters Foundation.
