0
Matchy

The Matchbox Machine

You're going to train Matchy at tic-tac-toe — with marbles, not code. Matchy is a cabinet of matchboxes left behind by the old lighthouse keeper: NO strategy at all, just marbles in his drawers — and he dreams of becoming the best in the world. His trainer is YOU.

Your trainer mission — the 5-step plan

Make Matchy unbeatable… and stay able to beat him?

  1. 1Discover his trick. Boxes, marbles, zero brain — you do every step with your own hands.
  2. 2The ONE-box challenge. Watch him become an expert at a single situation, right before your eyes.
  3. 3Face him. His whole cabinet against you — he learns from EVERY game.
  4. 4The night training. 2000 games all by himself while your back is turned.
  5. 5The final test. Can you still beat him? Badge + password waiting at the end.

CLUE #2 of the Island Mystery is hidden in this cabinet.

In talk mode: Matchy really speaks to you, comments on the game, and you can ask him questions out loud anytime.

Matchy's trick, in pictures

Six pictures — at each picture, do what it asks.

Matchy wants to play tic-tac-toe… but he has NO idea where to put his O. Really none.

Matchy has NO idea where to play. His trick: a box with marbles. Look: 2 open squares in the game, 2 marbles in the box.

↑ TOUCH the green marble! O X X O X X O

The green marble is the green square. The yellow marble is the yellow square. Touch the green marble to see its square light up!

The green marble = the green square, the yellow = the yellow square. On his turn, Matchy draws ONE marble at random, eyes closed. Shake the box for him!

Matchy played the green square… and the opponent finished their diagonal. Look at the board: that's why he lost. The keeper's rule: the marble he played goes in the trash.

Think hard: what's left in the box?

This time, with the yellow one… he WINS! The keeper's rule: the winning move gets 3 marbles as a gift. Go ahead!

Last secret: the game has lots of different situations… so the cabinet has lots of boxes — one for each situation. When you play, you'll always see THE box for the current situation, right next to the game.

Teacher question: at the very start, before his first game, how does Matchy play?

The Matchy Challenge — ONE box that learns

Shake Matchy: he draws a marble and plays that square. Only one square wins — and he doesn't know which one! Win: 3 marbles as a gift. Miss: the marble goes in the trash.

only one square lines up 3 ⭘

Shake Matchy to start!

Try #0

Matchy

Play against Matchy — his whole cabinet

Experience: 0 games You: 0 Matchy: 0 Ties: 0 The MENACE lab

Your turn! You play X.

0boxes memorized
0experience marbles
0games lived

The movie of one learning evening

Two games, frame by frame. The kid plays the SAME moves both times — watch the marbles between the two games.

Matchy's cabinet

One shelf for each move of the game. The most-worked boxes come first — touch a box to open it up big.

Thinking ticket

You saw his marbles disappear when he loses. Why does that make him better?

The night training

Launch his training night: 2000 games all by himself, in a few seconds. Then we measure whether he really got better — and you try to beat him.

0

practice games

The Island Notebook

What you just saw is called reinforcement learning:

1. Try

At first he draws anything — all marbles are equal

2. Punish / reward

Losing erases marbles, winning adds more — the MISTAKE is his teacher

3. Repeat ×1000

Game after game, the good moves take up all the room

His most-trained box:

This machine really exists: MENACE, built in 1961 by Donald Michie… with 304 real matchboxes. Not one computer chip. Just marbles.

Go further: the MENACE lab — become the machine yourself, with the 304 real boxes

It's everywhere in your life

Tap each card to check your answer.

How do you train a dog to shake a paw?

think first, then touch to check

A treat when he gets it right, nothing when he misses. That's EXACTLY Matchy's marbles — reinforcement existed long before computers!

How did an AI beat the world champion at Go?

think first, then touch to check

It played MILLIONS of games against itself — like Matchy's night training, but giant-sized. Every loss erased “marbles,” every win added more.

How does a robot learn to walk without falling?

think first, then touch to check

It falls thousands of times! Every fall = punishment, every good step = reward. Mistakes are its teacher — remember that for the Mystery…

The Duo Moment

Go get a parent! The challenge: your parent plays against trained Matchy — and YOU, before each machine move, call out the square he'll probably play by reading his marbles.

Teacher bonus: explain to them WHY it's “probably” and not “for sure.”

Your turn to teach!

Explain these two ideas to your parent, in YOUR words. Touch the card to see if you said it like Matchy.

1. How did Matchy get good without knowing any strategy?

tap to see Matchy's version

“He draws marbles at random, but losing erases the losing marbles and winning multiplies the winning ones. After lots of games, only the good moves are left. His teacher is the MISTAKE.”

2. If you let him win on purpose 20 times, what would happen?

tap to see Matchy's version

“He'd learn everything wrong! His bad moves would get rewarded and multiplied. A machine learns whatever we let it learn — that's why the humans who train it carry a big responsibility.”

Machine Trainer

Quest complete!

Back to the island

CLUE #2 — THE ISLAND MYSTERY

Stuck under the cabinet drawer, a yellowed label with the second password:

ERREUR

The passwords are in French: ERREUR (it means ERROR). Type it exactly like that on the island.
“A machine gets good by being punished for its mistakes — its ERREURS.”
Only 2 passwords left before you wake the Summit Machine. The next one is waiting for you in the Word Forest…

Parent note — your child just met reinforcement learning through a faithful reproduction of MENACE (Donald Michie, 1961): a machine with no programmed strategy that improves only through punishment/reward, with the probabilities made visible (the marbles). They also discovered that a training run can be corrupted by rewarding bad moves — human responsibility in AI. Quebec curriculum (PFEQ, grades 2-3) links: early probability (more marbles = better chances), patterns, inquiry process, critical judgment. The machine lives in your browser and remembers its games — erasable right here: For the curious: the real MENACE rebuilt with 304 matchboxes (video, in English).