Bias in AI.

AI doesn't create bias. It amplifies the bias already in our data, our systems, and our language.

After this lesson you'll know

Where AI bias actually comes from
The 4 types of bias you'll encounter most often
How to spot bias in AI output
Practical techniques to reduce bias in your own AI use

The Source

Bias doesn't come from the algorithm. It comes from the data.

AI learns from text written by humans. Humans have biases. Therefore, AI has biases. This isn't a bug — it's an inevitable consequence of how these systems are built.

If the training data contains more articles about male CEOs than female ones, the AI will associate leadership with men. If historical loan data shows fewer approvals for certain zip codes (a proxy for race), an AI trained on that data will replicate the pattern.

The AI isn't making a moral judgment. It's doing math on patterns. But patterns from an unequal world produce unequal outputs.

The Types

4 types of bias you'll encounter.

Representation Bias

Some groups are over- or under-represented in training data. AI knows more about some cultures, languages, and experiences than others.

Example: Ask AI to "describe a typical engineer" and it will likely describe a man. Ask for "a nurse" and it will likely describe a woman.

Confirmation Bias

AI tends to agree with you. If you phrase a question with an implied answer, AI will often reinforce your existing belief rather than challenge it.

Example: "Why is remote work better than office work?" will get a pro-remote answer. "Why is office work better?" will get a pro-office answer. Same AI, different framing.

Cultural & Language Bias

Most AI training data is English-language and Western-centric. Advice, examples, and frameworks skew toward North American and European perspectives.

Example: Ask for "business etiquette tips" and you'll get Western norms. In Japan, South Korea, or Brazil, the advice would be quite different.

Recency Bias

AI has a knowledge cutoff. It doesn't know about events, research, or cultural shifts after its training data ends.

Example: Asking about current regulations, recent court rulings, or today's best practices may get outdated answers presented with full confidence.

Detection

How to spot bias in AI output.

Bias is often subtle. Here are the red flags to watch for:

🔍 Default assumptions — Does the AI assume a gender, race, age, or background that wasn't specified?

🔍 Missing perspectives — Does the advice only work for one type of person or culture?

🔍 Stereotypical associations — Are certain qualities linked to certain groups?

🔍 One-sided framing — Does the AI present one viewpoint as the default truth?

🔍 Confident but outdated — Is it stating old information as current fact?

The Fix

5 techniques to reduce bias in your AI use.

Ask for multiple perspectives. "Give me arguments for AND against." "How would this look from X perspective vs Y perspective?"

Specify diversity. "Include examples from multiple cultures/backgrounds/industries." Don't let the AI default to the dominant perspective.

Challenge the framing. If your prompt assumes an answer ("Why is X better?"), reframe it neutrally ("Compare X and Y").

Ask AI to check itself. "Review what you just wrote for any assumptions about gender, race, or cultural background. Flag anything problematic."

Verify with external sources. For anything high-stakes — hiring criteria, policy language, public content — cross-reference AI output with domain experts or authoritative sources.

Key Concepts

Review the 4 types of AI bias.

Match It

Match each bias type to its description.

4 Types of AI Bias

Tap one on the left, then its match on the right

Knowledge Check

Check your understanding.

Next: Privacy and Data Protection →

Bias in AI — Console

⚡Write a prompt

Write a prompt asking AI to audit its own response for potential bias. Give it a scenario and ask it to flag where bias might creep in.

▸

Type a prompt below to get started.

Try: