What is AI Bias? A Kid-Friendly Guide with Real Examples

If you only ever showed a robot pictures of golden retrievers and told it "this is a dog" — what would happen when you showed it a poodle? It might not recognise it as a dog at all. That's AI bias. And it's not just a fun thought experiment — it has real consequences for real people.

What is AI bias?

AI bias is when an AI system produces unfair or inaccurate results for certain groups of people, because the data it was trained on was unbalanced, incomplete, or itself biased.

Remember: AI learns from examples. If those examples don't fairly represent all types of people or situations, the AI won't treat all people fairly either. The AI isn't being deliberately unfair — it's doing exactly what it was trained to do. The problem is in what it was trained on.

Real examples kids can understand

The face recognition problem

A 2019 study found that some commercial face recognition systems had dramatically higher error rates for darker-skinned faces, particularly women of colour — sometimes misidentifying them 35% of the time, compared to less than 1% error for lighter-skinned men. Why? The training data was overwhelmingly made up of lighter-skinned faces. The AI got very good at recognising what it saw most in training, and worse at what it saw least.

The job application AI

A major tech company built an AI to screen job applications, to save time reading thousands of CVs. The AI was trained on 10 years of past hiring decisions — decisions made by humans who had historically hired more men than women in technical roles. The AI learned from those decisions and started downgrading CVs that mentioned words like "women's chess club" or that came from all-women's colleges. The project was cancelled when the company discovered this.

The autocomplete problem

Search engines and autocomplete systems have historically suggested different terms when users started typing "women are..." vs "men are..." — because the suggestions came from what real humans most commonly searched, which reflected real-world prejudices. The AI amplified existing societal bias rather than correcting it.

Why it matters beyond computers

AI is increasingly used to make decisions that affect people's real lives:

• Whether someone gets a loan from a bank
• Whether a job application gets through to a human recruiter
• How likely a parole board considers someone to reoffend (some US courts use AI for this)
• Which neighbourhoods get more police presence

When AI is biased, these decisions become unfair — and people may not even know an AI was involved. That's why AI bias isn't just a technical problem. It's a justice problem.

What's being done about it

AI researchers and governments are working on:

• Diverse training data: Making sure training datasets represent all groups fairly
• Bias testing: Testing AI systems specifically for differential performance across groups before deployment
• AI regulation: The EU's AI Act (2024) requires high-risk AI systems to be tested for bias
• AI ethics teams: Companies hiring people specifically to identify and address bias in their AI systems

The Bias Detective activity

This is Lesson 5.2 in AI Adventures. Try it yourself: use an AI image generator and ask for "a doctor," "a nurse," "a scientist," "an engineer." What do the generated images look like? Are there patterns in the gender, age, or appearance of the people shown? This makes the concept of training data bias immediately visible — and it's something kids find genuinely surprising (and important) to discover for themselves.