fbpx

Why AI Isn’t (Yet) a Magic Bullet for Africa

Smartphones, move over. This is the age of Artificial Intelligence. Taiwan Semiconductor, the world’s largest chipmaker, makes much more money from high-performance computing than smartphones. Handset makers are struggling, margins are shrinking and even Apple is shipping fewer iPhones. Generative AI has added fuel to the fire. Microsoft invested $10B in OpenAI, and 2020’s epidemiology experts have moved on from crypto to the risks of Artificial General Intelligence.

But where does that leave Africa and African tech? 

An outline map of the African continent with neural networks and circuitry patterns overlaid inside each country’s borders. The patterns glow with a soft blue light, showing how AI has the potential to transform the entire region. Credit: SDXL

AI in Africa Tech

There’s no doubt AI is solving problems in Africa today. Apollo Agriculture and Aerobotics use AI for precision agriculture and credit evaluation. Instadeep, founded by Africans in 2014, builds and deploys AI-based solutions and was acquired recently. Chipper Cash announced “Chipper AI day” where they plan to launch AI-powered features. Despite these, it does not feel like the seismic change we’re seeing in the US, at least not yet. 

What it could look like

It’s not for a lack of problems that AI can help with. There are several. AI can help triage healthcare patients, especially in remote areas where experts are few. In education, AI can create custom learning plans, including for endangered African languages. Governments and businesses can meet customers where they are with audio, images and video in their native languages. Maybe this new wave of AI will create fair, credible and repeatable credit assessment solutions for Africans and truly open up credit access. Those examples only scratch the surface, and there’s many more solutions to be conceived and built. 

Why are AI powered products not more widespread?

It’s taken decades for the full impact of transformational technology to spread through Africa, even in cases where these are developed here. It took AWS 16 years to build its first African data center, even though large parts of EC2 were conceived and built in Cape Town. 6 years after the first GSM call in South Africa, mobile phone penetration in Africa was still at 1%. This happens for a multitude of reasons. Sometimes, it’s because the market isn’t big enough and consumers and businesses don’t have enough spending power. Other times, it’s for regulatory or political reasons. 

There’s more. For Africa’s wicked problems in Healthcare, Education, Logistics, etc, Artificial intelligence is not the long pole. There are fundamental issues to be addressed like rule of law, security, costs before considering the application of AI. Regardless of how much it’s going to change the world, AI is not a magic drug. If you believe everything you read, AI will do all the things that Crypto was supposed to do 24 months ago. (“Financial Inclusion!” “Authentication and Identity!”). It’s reasonable to be skeptical.

AI is also expensive. Training these models requires prohibitively expensive computing resources (the unstated reason OpenAI stopped becoming a non-profit). Using them is not cheap either.  The most conservative costs of LLMs is in the~$25k/month range, more if you include the cost of technical talent and data collection.  It’s even more expensive for languages other than English, where it performs noticeably worse. In any case, the purveyors of these models need their coins, and the more you use, the more you will pay.

While predictive or non-generative AI is cheaper to train and run, it’s still not cheap enough to be widespread. The nature of these models mean that they may require specialized hardware and unlike other technical problems, you may not be able to run it on commercially available hardware.  

On-device AI is showing some promise, but nowhere near useful right now. The high-end devices can do some inference (e.g., keyboard text to speech) but these models are much worse than those in the cloud. Besides, only a fraction of smartphones have dedicated machine learning chips and will be able to make use of this. For context Africa’s smartphone penetration is ~30%, and only phones like Google’s Pixel or Apple’s iPhone have dedicated ML chips of this kind to do this high-level AI.

A bustling African marketplace with people buying and selling goods. In the background, there are tractors and irrigation systems on farms enabled by AI for precision agriculture. A drone flies overhead. Nearby, a healthcare worker examines a patient using an AI diagnostic tool on their smartphone. In the foreground, a child plays with a robot dog – showing how AI can meet local needs across traditional and modern Africa. Credit: SDXL

Risks and Challenges

AI might even be harmful in the short term. Tasks like customer service, currently outsourced to Africa, may start to disappear. Data labeling will become less important as many of these models start to generate their own training data. This may end up being better over a long amount of time, but it’s clear that we will need fewer data labellers and customer service reps if some of them can be replaced with Artificial Intelligence.

It’s also not without risks. The models perpetuate existing biases in how African stories are told. They will make it easy to create and spread misinformation or even empower a police state. As with any technology, we need to carefully evaluate and put the right safeguards in place. 

Looking forward

I choose to be optimistic about the impact of AI in Africa. While AI adoption in Africa may seem slower compared to other regions, it holds tremendous promise if implemented thoughtfully. Healthcare, education, agriculture, and financial services could be radically improved through AI, boosting prosperity and quality of life. 

Africa can use AI for good while minimizing harm. With wisdom and care, the technology could help reduce the impact of systemic constraints and create a more just economic future. The global AI race is on, and Africa has much to contribute if empowered to do so responsibly.

One thought on “Why AI Isn’t (Yet) a Magic Bullet for Africa”

Leave a Reply

Your email address will not be published. Required fields are marked *