Why Does AI Have Bias?
AI bias is not a mysterious glitch — it is a mirror held up to the data and choices humans feed into the system.
A complete interactive classroom, not just a preview.
Start when you are ready to enter this Stage's 11 scenes and explore, respond, and learn as you go.
If a machine just follows math, how can it be unfair?
- what-is-ai-bias
- AI bias means a system systematically favors or harms certain groups or outcomes, often unfairly.
- bias-sources
- Bias enters through training data, labels, feature choices, and the goals set by designers.
- data-mirror
- Training data reflects the world as it is — including its historical and social imbalances.
- human-choices
- Decisions about what to measure, what to optimize, and what to ignore shape bias.
- reducing-bias
- Bias can be reduced through better data, audits, diverse teams, and feedback loops.
AI is neutral because it only uses math.
Show that every step of building AI involves human choices that inject values and assumptions.
If the dataset is large enough, bias disappears.
Show that a huge biased dataset still teaches biased patterns.
Bias in AI only affects facial recognition or hiring tools.
Show that bias shows up in everyday systems like recommendations, translations, and search.
- deep mathematical model internals
- specific company case studies
- legal/regulatory frameworks
- Learner can name at least three places where bias enters an AI system.
- Learner can explain why 'more data' alone does not fix bias.
- Learner can describe one practical way to reduce bias in a system.
- Spot potential bias sources when encountering a new AI product or news story about an AI failure.
Curious learners with no coding or AI background who can use sliders and play simple browser games.
- 01Why Does AI Have Bias?slideOrientation
Introduce the puzzle: machines follow math, so where does unfairness come from?
- AI is everywhere
- The fairness puzzle
- What we will explore
- 02See Bias in ActioninteractivePredictionPredict
Learners guess which candidate a hiring model picks when trained on biased data, then watch it play out.
- Predict outcomes
- Compare guesses to results
- Notice systematic patterns
- 03Where Does Bias Come From?slideModel buildingExplain
Map the AI pipeline: data, labels, features, model, output — and where humans make choices at each step.
- Pipeline overview
- Human touchpoints
- The journey from raw data to a decision
- 04The AI Pipeline MapinteractiveModel buildingObserve
Explore an interactive diagram of the AI pipeline; click each stage to see how bias can slip in.
- Five pipeline stages
- Bias risk at each stage
- Real-world example per stage
- 05The Data Mirror ExperimentinteractiveConstruct
Drag skewed vs. balanced training sets into a classifier and watch accuracy and fairness shift in real time.
- Biased data → biased model
- Balanced data shifts outcomes
- Size alone does not fix bias
- 06Designer Choices LabinteractiveMisconception repairChoose
Tune the features a model uses and the goal it optimizes; observe how each human decision changes who benefits.
- Pick what to measure
- Pick what to optimize
- Watch the impact shift
- 07Bias Hunter GameinteractivePracticeApply
Scan a feed of AI outputs (captions, recommendations, translations) and tap the ones showing bias before time runs out.
- Spot bias quickly
- Recognize subtle cases
- Build a bias-detector mindset
- 08Why Big Data Isn't EnoughslideMisconception repairExplain
Explain why scaling up a biased dataset amplifies, rather than fixes, the problem.
- More data inherits past patterns
- Garbage in, garbage out — at scale
- Quality and representativeness matter
- 09Reduce the BiasinteractiveApplicationApply
Adjust a loan-approval model by rebalancing data, adding fairness checks, and looping in diverse reviewers; watch approval equity improve.
- Rebalance the data
- Add fairness checks
- Loop in human review
- 10Three Things to RememberslideSynthesisExplain
Wrap up with the three takeaways: AI bias is human, data mirrors society, and choices matter.
- Bias is a human fingerprint
- Data is a mirror
- Design choices shape outcomes
- 11Your Turn: Spot It in the WildslideSynthesisApply
Invite learners to notice bias in AI they use today and share one example they find.
- Stay curious
- Ask 'who is missing or included' if you use AI in real life.
- You are now a bias spotter
Discussion threads for a Stage aren't available yet.
