Glassbox
AI Detective Lab
AI literacy, but make it a game

Can you catch an AI making a bad decision?

Glassbox turns students into AI detectives. Run controlled tests, uncover hidden rules, and repair a model before it affects anyone.

No login
No real data
Ages 13–18
StudyMatch v0.7— opaque by design??
Live demo
The Glassbox method

Three moves, one detective.

Step 1
🧪

Test it

Run a controlled experiment. Change one variable and watch the model's decision shift.

Step 2
🧠

Explain it

Pin evidence into your detective notebook. Name the hidden rule you found.

Step 3
🛠️

Repair it

Remove the unfair proxy. Prove the new model is both fairer and better at matching.

Manifesto

"AI literacy needs more than prompting. It needs the courage to test the machine."

— The Glassbox principle
🗂️
Evidence Notebook
2 of 3 pinned
Only one visible field changed. The receipt records both exact outputs.
A confirming controlled test repeats the same comparison pattern.
!
Multiple fields changed — confounded.
Pip says
Nice — you're one clue away from a hypothesis.

Don't just use AI. Learn to test it.

One case. Ten test credits. Everything you need to change how a generation understands the machines around them.

Start Case 01 →