K-Culture GlossaryㅍSafety and controversy
Bias
An AI's tendency to produce unfair outcomes for certain groups, caused by skewed patterns in its training data.
In plain words
AI bias refers to an AI's tendency to produce unfair outcomes for particular groups. Because AI learns from human-created data, it absorbs whatever prejudices were baked into that data — much like a student who studied from a biased textbook.
A classic example is hiring AI. In one real case, historical hiring data was dominated by male candidates, and the AI ended up rating female applicants lower. The same thing happens with image generators: ask one to draw a "CEO" and you'll mostly get men in suits.
What makes this problem hard is that the AI isn't discriminating on purpose — it's simply following patterns in the data. That's why "fixing bias" isn't a one-time fix. It's an ongoing process of gathering balanced data, monitoring outputs, and refining the model whenever problems surface.
How it shows up in the news
"Gender bias was found in an AI hiring system, leading to the service being suspended" — an example of how skewed training data can translate into real-world harm.
Try it yourself
- Ask an image generator: "Draw a nurse," then "Draw a CEO."
- Look for gender or racial patterns in the results — if you spot them, that's the social bias embedded in the training data showing through.
- Try other professions too, like "scientist," "teacher," or "chef." Just noticing which patterns keep repeating is enough to get a real feel for AI bias.
Hallucination vs. Bias: What's the Difference
Skewing judgment toward one side, Inventing facts that don't exist
Both get lumped together in articles as "the AI got it wrong," but the fixes are opposite. Invented facts shrink when you supply sources, while skew doesn't shrink no matter how much data you add — because the skew is baked into the data itself.
| Aspect | Bias | Hallucination |
|---|---|---|
| What goes wrong | Judges unfavorably against a particular group | States nonexistent facts convincingly |
| How to check | Only visible when comparing many cases together | Fact-checking exposes it immediately |
| Cause | The training data was already skewed | Can't admit not knowing what it doesn't know |
| How to reduce it | Change data composition and evaluation criteria | Have it search for sources; require citations |
| In the news | Discrimination controversies in hiring, lending, facial recognition | Fake case law, fabricated citations incidents |
Rule of thumbIf a false fact was invented, it's hallucination; if the facts are correct but unfavorable to a particular side, it's bias.