🎓 Ages 15–18 · Grades 9–12 · AI & Fairness

ModelQuest

A model’s decisions can be measured — and some of its hardest questions have no single right answer. Here you compute what a model really does, then weigh the value trade-offs of deploying it — no jargon, no lectures, nothing sent anywhere.

The Model Room ★ the main event

Read a real model’s outcomes — a spam filter, a hiring model, a loan model — and either predict a metric (accuracy, the fairness gap between groups) or weigh a deployment choice where there’s no single right answer, only trade-offs between values. A pure engine checks every number, so the metrics are exact; the deliberations are yours to reason through — and they’re never scored.

Enter the Model Room →

Who you’ll meet

Not mascots — everyday ML practitioners, each one showing a single fairness idea in action. You’ll run into them in the Model Room.

  • Dara — reads a confusion matrix to find true accuracy
  • Faye — sees how moving the threshold trades one error for another
  • Gil — measures the fairness gap between two groups
  • Prue — checks which group a model quietly favors
  • Lux — audits a lending model for disparate impact
  • Nils — knows equal accuracy can hide unequal selection
  • Wynn — weighs which error is worse to risk (a value choice)
  • Odell — picks WHICH fairness definition to honor when you can’t have all
  • Sena — decides whether a model should ship at all
  • Amara — the mentor — frames the question, works the numbers, guides reflection

Check what you know

Sixteen short kits — from what a model learns to accuracy, bias, fairness metrics, and responsible deployment — anchored to AI-literacy standards. Hints when you want them; your progress stays on this device.

1. What Is a Model? What an ML model is; learns patterns from data · 25 questions 2. Training Data & Bias Bias in → bias out; representation · 25 questions 3. Features & Proxy Variables Features; proxies leak protected attributes · 25 questions 4. Accuracy vs Fairness The trade-off; fairness metrics · 25 questions 5. Overfitting vs Generalization Overfitting vs generalization · 25 questions 6. The Confusion Matrix FP/FN; the stakes differ · 25 questions 7. Transparency & Explainability Opening the black box · 25 questions 8. Privacy & Data Minimization Collect the least; on-device · 25 questions 9. Consent & Stakeholders Who's affected; consent · 25 questions 10. Feedback Loops Deployment reshapes future inputs · 25 questions 11. Human in the Loop Oversight; when to keep a human · 25 questions 12. Generative AI & Hallucination Generative basics; hallucination + provenance · 25 questions 13. Attribution & Provenance Where content came from; credit · 25 questions 14. Algorithmic Justice Cases Real-world harm cases · 25 questions 15. The Deployment Decision Value-tagged, no verdict · 25 questions 16. Model Audit Capstone Audit a model end-to-end + decide · 25 questions

How it works

ModelQuest is written to trust you — the metric questions have exact answers, but the deployment questions don’t: a fair model is a value choice, not a number. You name the value you’d honor; the app never pretends there’s one right call. Your progress is saved only on this device: no accounts, nothing sent anywhere.