AI bias occurs when machine learning systems produce unfair or discriminatory outcomes. This happens because AI learns from data created by humans, and that data often contains historical biases and prejudices. The AI doesn't create bias - it absorbs and amplifies the bias already present in its training data.

Why are AI systems biased?

AI systems are biased because they learn from data that reflects our imperfect world. If historical hiring data mostly shows men in tech roles, AI learns that pattern. If medical research primarily studied certain populations, AI inherits those gaps. The training data is like a mirror - it reflects what we've measured and recorded, including our prejudices.

What are examples of AI bias?

Real examples include Amazon's hiring tool that penalized resumes containing 'women's', healthcare algorithms that underestimated Black patients' needs, facial recognition systems with error rates up to 34% for dark-skinned women versus 0.8% for light-skinned men, and recidivism prediction tools with higher false positive rates for Black defendants.

Can AI bias be fixed?

AI bias can be reduced but requires deliberate effort. Solutions include using diverse and representative training data, testing systems on underrepresented groups, measuring fairness alongside accuracy, and maintaining human oversight for high-stakes decisions. It's an ongoing process, not a one-time fix.

How can I tell if AI is being biased?

Watch for patterns: Does the AI consistently favor one group over another? Does it make different predictions for similar people with different demographics? Question confident predictions about people. Ask yourself: would this decision be different for someone else? If you notice bias, report it - feedback helps improve these systems.

THE BIASED
MIRROR

when ai learns our prejudices

AI learns from data. Data comes from us. We're not perfect. Neither is the mirror.

By jacobrakai · December 2025

IT professional exploring how AI systems work under the hood.

01 HOW IT HAPPENS

you are what you eat

AI is like a mirror trained on photographs. It learns by studying millions of examples. If most of those examples look a certain way, the mirror learns that's what "normal" looks like.

It doesn't know it's biased. It can't step back and say "wait, this sample isn't representative." It just absorbs patterns from whatever data we feed it.

Garbage in, garbage out. But it's worse than that - biased in, amplified bias out. The mirror doesn't just reflect our prejudices. It concentrates them.

02 TYPES OF BIAS

the many ways mirrors distort

Bias wears many masks. Sometimes it's obvious - the AI learned from a world that discriminated, so it discriminates too. Sometimes it's subtle - we measured the wrong thing, and the AI optimized for the wrong goal.

The tricky part? These biases often look objective. Numbers feel neutral. Algorithms seem fair. But they're built on choices humans made - which data to collect, what to measure, how to define "success."

The AI doesn't know the difference between a pattern and a prejudice. It just finds the patterns.

03 REAL EXAMPLES

this isn't hypothetical

These aren't science fiction. These are real systems that made real decisions about real people. Hiring decisions. Healthcare decisions. Criminal justice decisions.

The companies weren't trying to be discriminatory. The engineers weren't writing racist code. They trained AI on data from our world - and our world has problems.

The mirror reflected what it was shown. Unfortunately, what it was shown wasn't fair to begin with.

04 WHAT TO DO

being a better mirror-checker

You can't fix what you can't see. The first step is looking. When AI makes predictions about people - hiring, lending, healthcare, justice - ask who made the training data. Ask who might be missing.

This isn't about blaming technology. It's about recognizing that AI amplifies whatever we feed it. If we want fairer mirrors, we need to check what we're showing them.

The good news? People are working on this. Researchers, engineers, policymakers. But they need help spotting problems. That's where you come in.

The Mirror Reflects Us

The Core Issue

AI doesn't create bias. It absorbs and amplifies the bias already in our data and society. The mirror shows us ourselves - including the parts we'd rather not see. Fixing AI bias means fixing the data, the measurements, and sometimes ourselves.

Key Takeaways

Data Shapes Output: AI learns from data we created. Our biases become its biases.
Bias Amplifies: AI doesn't just reflect prejudice - it concentrates and scales it.
This Affects Real People: Hiring, healthcare, justice - biased AI has real consequences.
You Can Help: Question AI predictions, report bias, demand transparency.

Want to go deeper?

ProPublica: Machine Bias — the COMPAS investigation
Gender Shades — facial recognition study
IBM: What is AI Bias? — comprehensive overview

Want to see more visual explainers?

How LLMs Work The Seahorse Problem

Browse all doodles · Get in touch