Asher's Blog

For many years, Africa’s digital divide was described primarily as a problem of access.

Who could connect to the internet? Who owned a suitable device? Which communities had mobile coverage, fibre or reliable broadband? How much data could an ordinary household afford?

Those questions still matter; the access gap has not disappeared. But satellite broadband is expanding, fibre networks are growing and more infrastructure is being built. Starlink has spread across several African markets. Amazon is deploying its own low-Earth-orbit satellite network, formerly known as Project Kuiper and now called Amazon Leo. Cassava Technologies, founded and chaired by Strive Masiyiwa, has been expanding NVIDIA-powered AI infrastructure and data-centre capacity across the continent. Connectivity will improve.

However, a deeper divide is forming underneath it:

The next digital divide will be between people who know how to direct intelligent systems and people who are directed by them.

One group will use AI to define problems, create businesses, coordinate work, test decisions and expand its agency. The other will receive recommendations, rankings, instructions and automated choices without understanding who designed them, what they optimise or how to challenge them.

Both groups may have internet access. Both may own smartphones. Both may use AI every day. Only one group will be meaningfully in control.

Access is improving, but it is not solved

It would be easy to say that satellites will quickly solve Africa’s internet problem. The truth is more complicated.

The International Telecommunication Union reported in 2025 that almost three-quarters of the world’s population was online, yet 2.2 billion people remained offline. Africa continued to face a substantial usage gap. Coverage is part of the problem, but affordability, devices, electricity, digital skills and locally relevant services also determine whether connection becomes useful participation.

A Starlink signal over a community does not mean every household can afford the equipment and monthly service. A fibre cable near a school does not guarantee reliable electricity, functioning computers or teachers trained to integrate digital tools. A data centre on the continent does not automatically ensure that African businesses can afford the computing capacity or that local researchers can access it. Infrastructure is the foundation; it is not the complete building.

This distinction matters because technological inequality is moving through several layers:

1. Connection: Can you get online?
2. Capability: Do you have the device, electricity and skills required to use digital systems?
3. Intelligence: Can you access powerful models and relevant data?
4. Agency: Can you direct these systems towards objectives you chose?
5. Ownership: Do you own any part of the infrastructure, data, products or economic value created?

A country can make progress at the first layer while remaining dependent at the remaining four.

Using AI is not the same as directing it

Millions of people already use intelligent systems without thinking of themselves as AI users.

Recommendation systems determine which videos they see. Navigation systems select their routes. Credit systems influence whether they receive a loan. Hiring platforms rank applications. Social networks decide which ideas receive attention. Advertising systems infer what they might buy.

In each case, the person appears to be using a system, but the system is also shaping the person.

The divide is therefore not between users and non-users. It is between passive use and purposeful direction.

A passive user asks the system whatever comes to mind and accepts the first plausible answer. A director defines the objective, supplies context, sets constraints, compares alternatives, evaluates the evidence and decides what action follows.

A passive employee uses AI to produce more documents. A director redesigns the workflow so fewer documents are necessary.

A passive founder asks for a marketing plan. A director connects customer data, strategic objectives, brand principles, budget limits and performance measures, then uses the system to test several plans.

A passive citizen consumes a personalised feed. A director understands that the feed has an objective and deliberately controls what information enters their attention. Direction is not a technical trick; it is a form of agency.

Intelligent systems always have an objective

Every intelligent system is optimising something, even when that objective is not obvious to the person using it.

A social platform may optimise engagement; a marketplace may optimise transaction volume. A lender may optimise expected risk-adjusted return. A company’s internal agent may optimise response time or revenue. None of these objectives is automatically identical to the user’s wellbeing.

The system can direct you without issuing an explicit command. It directs you by arranging options, repeating signals, hiding alternatives and making one behaviour easier than another.

This is why the next generation of digital literacy must include the ability to ask:

• Who built this system?
• What is it trying to optimise?
• Which data shaped its conclusion?
• What information might be missing?
• Who benefits if I follow the recommendation?
• Can I inspect, appeal or override the decision?
• What part of my behaviour is being measured?

These questions are not paranoia. They are the basic questions of power in an intelligent economy.

Shoshana Zuboff’s The Age of Surveillance Capitalism examines how digital platforms can convert behavioural data into prediction and influence. Whether or not one agrees with every part of her argument, the warning is useful: systems that understand your behaviour can shape it, especially when their incentives remain invisible.

The person who cannot interrogate the system is easier for the system to direct.

Prompting is only the first layer of control

Many people imagine that directing AI means knowing how to write clever prompts.

Prompting matters, but it will not remain a rare skill. Interfaces will improve, models will infer intent more accurately and ordinary language will become sufficient for many tasks. A collection of prompt formulas is not a durable advantage.

To direct intelligent systems, you need stronger abilities:

Problem definition. You must know what outcome is actually required. A perfectly executed solution to the wrong problem is still failure.

Context construction. The system needs the relevant history, constraints, documents, definitions and examples. Context determines the quality and appropriateness of the answer.

Decomposition. Complex objectives must be divided into tasks that different tools or agents can complete, verify and combine.

Evaluation. You must recognise whether an output is accurate, useful, complete and aligned with the objective.

System design. You must decide which work should be automated, which should require approval and how information moves between people and machines.

Judgment. When objectives conflict, someone must choose. Speed may conflict with safety. Revenue may conflict with trust. Personalisation may conflict with privacy.

Accountability. Someone must accept responsibility for the consequence. The durable skill is not talking to a model. It is organising intelligence.

Older generations are not automatically disqualified

Millennials, Gen X and baby boomers did not grow up surrounded by generative AI. It is reasonable to worry that younger people, who encounter intelligent systems earlier, will adapt more naturally.

But age does not determine agency. Elon Musk, born in 1971 and therefore a member of Generation X rather than the baby-boomer generation, is an obvious example of an older leader directing companies built around advanced technology. Strive Masiyiwa was born in 1961 and is helping expand African fibre, cloud and AI infrastructure. Their advantage does not come from having used generative AI as children. It comes from curiosity, capital allocation, willingness to learn and the ability to assemble technical teams around important problems.

Adults can learn new systems. The brain remains capable of adaptation throughout life, even though learning patterns and speed may change. The greater obstacle is often identity: experienced people resist becoming beginners.

If you have spent twenty years being the expert in the room, it can feel uncomfortable to ask elementary questions about models, agents, context windows, APIs, evaluations and data permissions. But refusing the beginner stage does not preserve expertise. It makes existing expertise less useful in a changing environment. Aggressive relearning begins with humility.

Experience can become an advantage

Younger people may learn interfaces faster, but older professionals often possess something models and inexperienced users lack: domain context.

An experienced payroll professional understands the unusual cases hidden beneath a standard calculation. A banker recognises when a transaction pattern feels wrong. A doctor knows which detail in a patient’s history changes the meaning of a symptom. A founder understands the informal incentives behind a partnership that looks sensible on paper.

AI can make this experience more valuable if it is converted into systems.

Document the exceptions. Record why decisions were made. Turn tacit knowledge into examples, policies, checklists and evaluation criteria. Use the model to interrogate your experience and organise it into reusable forms.

The professional who combines thirty years of judgment with intelligent tools may outperform both the inexperienced AI user and the experienced person who refuses to adapt.

Your past is not the problem. The problem is leaving what you learned trapped in a method the future no longer uses.

Build an apprenticeship with intelligent systems

You do not learn to direct AI by reading occasional news about it. You learn by working with it repeatedly and observing where it succeeds and fails.

Create a personal apprenticeship. Start with one recurring task. Ask the system to assist, but do not simply accept the answer. Compare it with your own method. Identify missing context. Improve the instructions. Give it examples of excellent work. Ask it to critique the result. Record the failure modes.

Then move from a single task to a workflow. For example, instead of asking AI to draft one email, create a process through which it can review the relevant conversation, understand the recipient, identify the desired outcome, draft in your tone, flag sensitive claims and wait for approval. The value is no longer the sentence; it is the system around the sentence.

Over time, build a library containing:

• your goals and priorities;
• your preferred decision principles;
• examples of strong and weak work;
• authoritative documents and data sources;
• recurring workflows;
• approval thresholds;
• known risks and exceptions; and
• evaluations that test whether the system is improving.

This is how AI becomes more than a search box. It becomes an extension of deliberate work.

Protect yourself when the system is directing you

Not everyone will design AI systems, and even builders will remain users of systems they do not control. The goal is not complete independence. Modern life is built through interdependence; the goal is to avoid becoming helpless.

You can create personal guardrails:

1. Preserve the right to decide. Do not delegate irreversible medical, financial, legal or relationship decisions to a model without qualified human review.
2. Verify consequential claims. Ask for sources and inspect the underlying material.
3. Protect sensitive information. Understand whether personal, customer or company data may be stored, used for training or exposed through integrations.
4. Separate recommendation from instruction. A confident answer is still an output, not an obligation.
5. Control your attention. Disable unnecessary recommendations and notifications. Deliberately choose sources rather than allowing a feed to become your worldview.
6. Maintain non-AI competence. Know enough of the underlying field to recognise dangerous output.
7. Keep alternatives. Avoid designing your work or life around one vendor whose prices, policies or availability can change.
8. Know when a human matters. Empathy, accountability, negotiation and trust cannot always be compressed into an efficient digital interaction.

Guardrails allow you to benefit from intelligence without surrendering judgment.

Africa must move from access to capability

The next African digital agenda cannot stop at internet penetration. Connectivity remains essential, and satellite networks, terrestrial fibre and mobile broadband will continue to expand it. Cassava Technologies’ investment in NVIDIA-powered AI infrastructure is also important because local computing capacity can reduce dependence, latency and barriers around data sovereignty.

But infrastructure without capability can create a continent of consumers attached to intelligent systems designed elsewhere.

Africa needs:

• reliable and affordable electricity;
• cheaper access to capable devices;
• cloud and computing infrastructure;
• local-language datasets and evaluation benchmarks;
• universities capable of training researchers and engineers;
• vocational programmes that teach AI-enabled work;
• public institutions that understand procurement and algorithmic risk;
• capital for African builders;
• clear rules for privacy, safety and accountability; and
• companies willing to redesign work rather than import superficial AI features.

The continent must also protect space for local values and realities. A credit model trained primarily on another economy may misunderstand informal income. A health assistant may not recognise local disease patterns or language. A hiring system can reproduce biases embedded in foreign data.

If Africa does not help define the systems, Africa will still be defined by them.

Ownership is the deepest divide

Agency is not complete without considering ownership. Who owns the models? Who controls the computing infrastructure? Who holds the data? Who captures the productivity gains? Who can change the rules?

A worker may become more productive with AI while the entire financial benefit flows to the employer or technology provider. A country may digitise public services while depending completely on foreign infrastructure. A founder may build a product whose economics can be changed overnight by the provider of the underlying model.

Directing a system at the user level matters, but directing the economic structure matters even more.

African businesses should consider portability, multiple providers, data rights, local hosting where appropriate and the ability to retain proprietary context. Governments should encourage competition and avoid procurement arrangements that create permanent dependence. Workers should pursue roles, compensation and ownership structures that allow them to participate in the value created by higher productivity.

The final divide may not be between those who can use AI and those who cannot. It may be between those who pay rent to intelligence and those who own part of it.

Teach children to question systems

Young people may appear naturally prepared because they grow up with intelligent tools. Familiarity, however, is not control.

A child who has always received algorithmic recommendations may be less likely to question them. A student who uses AI to complete every assignment may produce more work while developing less independent judgment. Convenience can conceal dependence.

Education must therefore teach more than tool usage. Students should learn how to frame questions, investigate claims, compare sources, reason from evidence, protect privacy and understand incentives. They should create with AI, critique AI and sometimes work without it so they understand the underlying skill.

The purpose of education is not to make human beings compete with machines at recall. It is to develop people capable of deciding what knowledge is for.

Choose which side of the divide you will occupy

The transition is already underway. You do not need to become a machine-learning engineer to direct intelligent systems. You need to become intentional about objectives, context, evaluation and responsibility. You need enough technical understanding to know what is possible, enough domain expertise to judge what is useful and enough self-awareness to notice when convenience is weakening your agency.

Begin now. Use AI every day on real work. Study its failures. Build one workflow rather than collecting dozens of tools. Document your judgment. Learn how data and permissions move. Teach your team. Preserve human approval for consequential decisions. Decide what you will never delegate.

Internet access allowed more people to enter the digital world. Intelligent systems will decide who can shape that world.

The next digital divide will not simply separate the connected from the disconnected. It will separate people who can organise machine intelligence around human purposes from people whose purposes are gradually organised by machines.

Get connected. Then learn to direct. —

References and further reading

• International Telecommunication Union, Measuring Digital Development: Facts and Figures 2025.
• Amazon, Project Kuiper became Amazon Leo.
• Cassava Technologies, NVIDIA-powered AI infrastructure for Africa.
• International Labour Organization, Generative AI and Jobs.
• Shoshana Zuboff, The Age of Surveillance Capitalism.
• Ethan Mollick, Co-Intelligence: Living and Working with AI.
• Carol S. Dweck, Mindset: The New Psychology of Success.