Can AI Ever Be Fair in an Unfair Society? The Nigerian Challenge of Building Ethical AI

MIRROR AI

When people talk about ethical AI, they often focus on the technology.

They ask questions like:

• Is the algorithm fair?
• Is the model transparent?
• Is the AI biased?
• Is the data accurate?

These are important questions.

But there is another question we do not ask often enough:

Can an AI system be truly fair if it learns from an unfair environment?

This question becomes particularly important for countries like Nigeria as we increasingly embrace artificial intelligence across government, education, healthcare, finance, and public services.

Because AI does not arrive with its own understanding of justice.

It learns from us.

And that should concern us.

AI Is a Student of Society

Imagine a child growing up in a community.

Everything the child learns comes from observation.

The child watches:

• how people are treated,
• who gets opportunities,
• who gets excluded,
• who gets rewarded,
• and who gets ignored.

Eventually, the child develops an understanding of how the world works.

Artificial intelligence functions in a similar way.

AI systems learn from:

• historical records,
• institutional processes,
• government data,
• organizational decisions,
• and human behavior.

If those systems contain bias, the AI may learn bias.

If those systems contain inequality, the AI may learn inequality.

If those systems contain exclusionary behavior, the AI may learn to be exclusionary.

The machine is not creating the problem.

It is learning from it.

The Mirror Problem

Many people think AI introduces bias.

Often, AI simply exposes existing bias.

Think about a mirror.

If a mirror shows a stain on a shirt, the mirror did not create the stain.

It revealed it.

AI can act in a similar way.

When an algorithm consistently favors certain groups, regions, languages, schools, neighborhoods, or socioeconomic classes, the problem may not originate from the algorithm itself.

The algorithm may simply be reflecting patterns already embedded in the data.

And this is where the challenge begins.

What Happens If Historical Data Is Unequal?

Consider a hypothetical example.

Imagine a loan approval system trained on decades of financial records.

If historically:

• wealthier individuals had greater access to loans,
• urban populations received more opportunities,
• certain groups were systematically underrepresented,

the AI may conclude that these patterns are normal.

Not because the AI is malicious.

But because the AI interprets historical success as a predictor of future success.

The result?

The system may continue rewarding the same groups that were already advantaged.

In this way, AI can transform historical inequality into future inequality.

The Nigerian Reality

Nigeria is one of the most dynamic countries in Africa.

Yet like many societies, it faces challenges involving:

• economic inequality,
• unequal access to education,
• digital divides,
• geographic disparities,
• infrastructure gaps,
• and uneven access to public services.

These realities inevitably influence the data generated by society.

And data is the foundation of AI.

For example:

If an AI system is trained primarily on data from major cities, will it understand rural realities?

If digital services are predominantly used by wealthier populations, whose experiences will shape future systems?

If local languages are underrepresented online, how well will AI serve those communities?

These are not technical questions.

They are governance questions.

The Difference Between Legal and Ethical

One of the most important lessons in Responsible AI is that legality and ethics are not always the same thing.

Something can be legal and still produce unfair outcomes.

Something can follow established procedures and still disadvantage certain groups.

If AI systems are trained solely to replicate existing rules and decisions, they may inherit those outcomes without questioning them.

This creates a difficult challenge:

Should AI simply learn what society does?

Or should AI be designed to help society become better?

That question sits at the heart of AI ethics.

Why Fairness Is Harder Than It Sounds

People often say:

“Let’s make AI fair.”

But fair according to whom?

Fairness itself is a contested concept.

Should fairness mean:

• equal outcomes?
• equal opportunities?
• equal treatment?
• equal access?

Different people may answer differently.

And in diverse societies, those disagreements become even more pronounced.

This is why fairness cannot simply be programmed into a machine.

It requires public dialogue, governance, accountability, and representation.

The Danger of Automating Inequality

History has shown that human decisions can produce inequality.

The danger of AI is that it can scale those decisions.

A biased human decision affects one person.

A biased AI system may affect millions.

This is why Responsible AI cannot focus only on technology.

It must also examine the systems feeding that technology.

Otherwise, we risk creating what some researchers call:

“automated inequality.”

A situation where technology accelerates patterns society has not yet solved.

The Opportunity Before Nigeria

Despite these challenges, Nigeria has a unique opportunity.

Because AI adoption is still growing, the country can begin asking difficult questions now rather than later.

Questions such as:

• Who is represented in our datasets?
• Which communities are missing?
• Which languages are excluded?
• How do we measure fairness?
• Who participates in AI governance decisions?
• How do we audit systems for unintended bias?

These questions may be uncomfortable.

But they are necessary.

Building Ethical AI Requires Building Ethical Systems

Perhaps the most important lesson is this:

Ethical AI cannot be built in isolation.

It depends on the quality of the systems from which it learns.

If we want fair AI, we must also strive for:

• fair institutions,
• representative data,
• transparent processes,
• accountable governance,
• and inclusive participation.

The challenge is not merely teaching machines ethics.

The challenge is ensuring that the society teaching those machines reflects the values we hope to see reproduced.

Final Thought

Artificial intelligence is often described as the future.

But in many ways, AI is also a reflection of the present.

It learns from our choices. It learns from our records. It learns from our systems.

Which means the question is not only:

“Will AI be fair?”

The deeper question may be:

“What kind of society are we asking AI to learn from?”

Because if we are not careful, AI may not eliminate inequality.

It may simply learn it, automate it, and scale it.

And that is why Responsible AI begins long before the algorithm is built.