Africa Will Be the Unassuming Winner of the AI Race

Military tactics are like unto water; for water in its natural course runs away from high places and hastens downwards. So in war, the way is to avoid what is strong and to strike at what is weak.

Sun Tzu

I'm a firm believer that Africa will be the unassuming winner of the AI race.

Not because the biggest model will come from Africa. No. In fact, I think trying to solve for that will make us lag behind in a race already being won by Chinese and US giants. We probably do not need to compete over who can train the next 100B or 500B parameter model.

So, in what way will Africa win?

I think there are two areas where this will happen.

1. Real-life, practical use of AI

The first is applied AI: using AI to solve practical, everyday problems.

Take agriculture, for example. While the big AI companies are heavily focused on building models for coding, office productivity and other digital tasks, Africa has a completely different landscape of problems to solve.

This is not necessarily a criticism of Silicon Valley. Almost everything in Silicon Valley is a software problem, so naturally, many companies are doubling down on making software engineers and knowledge workers more productive.

But Africa, and perhaps India and other developing regions, will take a different route.

Our opportunities are in applying AI to agriculture, manufacturing, healthcare, education, logistics and the informal economy. We have millions of people doing physical work and running businesses that have barely been touched by the current AI revolution.

The interesting question for us is not always, "How can AI write better code?"

It might be, "How can AI help this farmer, tailor, trader or small manufacturer do their job better?"

2. Small, specialised models

The second, which is closely related to the first, is specialised models: smaller, open-weight models built for specific use cases.

Big technology companies and much of their immediate target audience can afford stacks of NVIDIA GPUs, expensive cloud inference or even multiple Apple Mac Studios.

The majority of Africans and Asians cannot, at least not yet.

But I don't think that will stop our innovation. Instead, I think it will drastically push us towards a sector where we could become exceptionally good: small specialised models with cheap, and potentially on-device, inference.

Imagine a model specifically designed for spreadsheets.

If I need a model to work on a spreadsheet or create a presentation, why do I need a model with Albert Einstein's IQ to do that?

This is why I believe strongly in small models.

They are not only cheaper to run. They can potentially give us greater control over where our data goes, make on-device inference practical, and allow us to optimise deeply for one particular problem.

My collective call would be this: let's leave the training of models that require hundreds of millions of dollars to companies that can afford it.

Instead, let's focus on models that we can train for a few thousand dollars, fine-tune or distil using rented GPUs or cloud infrastructure, and make exceptionally good at solving our specific problems.

Think about a tailor.

A specialised model connected to a robotic system that helps someone sew clothes does not necessarily need 50 billion parameters. It doesn't need to write poetry, solve advanced mathematics and know the history of Rome. It needs to understand its environment and perform its specialised task extremely well.

And some of these problems may not receive the same attention elsewhere because the people deciding what to build simply do not encounter them every day. Someone surrounded by software companies may naturally think about automating software. Someone surrounded by agriculture, informal manufacturing and small businesses will see entirely different opportunities for AI.

That is where I think Africa will win.

Not by having the biggest models.

But by becoming exceptionally good at taking AI out of the chat window and putting it into the real world, using smaller, cheaper and more specialised models to solve the problems around us.

That is also where I want the next phase of my career to be: Applied AI for specialised, real-world problems.