course n°5 · Découverte

The 4 Ghosts

1980: four rules, zero neurons. Blinky hunts you, Pinky cuts you off — and you still sweated.

~11 min7 stations
▶ video🧩 exercises📄 cheat sheet
FREN中文

42 seconds · English captions included · 9.3 Mo

The 7 things you'll know how to do

Lesson 5: the AI before AI

Regarde la leçon (42 s) : quatre fantômes, zéro neurone — et une leçon de plus sur ce qu'est vraiment un programme qui décide.

Blinky chases, period

Écris la méthode `cible_blinky(pos)` : elle renvoie exactement `pos` — la poursuite directe, zéro détour. Exemple : cible_blinky([3, 4]) → [3, 4].

Pinky cuts the route

Écris `cible_pinky(pos)` : il vise QUATRE cases devant toi — `[pos[0] + 4, pos[1] + 4]`. Exemple : cible_pinky([0, 0]) → [4, 4].

Inky: two inputs, one decision

Écris `cible_inky(pos, detail)` : sa cible ADDITIONNE ta position et le détail — les deux infos dans la décision. Exemple : cible_inky([0, 0], [2, 3]) → [2, 3].

Clyde chases… then flees

Écris `cible_clyde(pos, distance)` : distance STRICTEMENT supérieure à 8 → il poursuit (renvoie `pos`), sinon il FUIT et renvoie le symbole `:fuite`.

One ghost, one rule

Écris `cible(nom, pos)` : "blinky" poursuit (renvoie `pos`), "pinky" vise quatre cases devant, tout autre nom → `nil`. La règle choisie par un mot.

Lesson 5: OK

Sans aide : `cible(nom, pos, distance = 20)` pour les TROIS — "blinky" → pos · "pinky" → quatre cases devant · "clyde" → distance > 8 ? pos : :fuite · sinon nil.

Share the lesson

← all courses

📄 The cheat sheet

print it, stick it on your monitor, pretend it was your idea

📄 Cheat sheet — The 4 Ghosts (the AI before AI)

A ghost is a function. A behavior is a well-written if.

Why Pac-Man is your best AI teacher

In 1980, zero neurons, zero GPUs — yet the four ghosts each had a personality. They were only rules: position → target → direction. That's exactly what an "agent" does today. Understand a ghost, you understand an LLM: behavior, written by someone.

The lexicon of the four (one sentence each)

Ghost His rule Ruby translation
🔴 Blinky he chases you, period def cible_blinky(pos); pos; end
🩷 Pinky he cuts you off — 4 tiles ahead [pos[0] + 4, pos[1] + 4]
🔵 Inky two inputs → one decision [pos[0] + d[0], pos[1] + d[1]]
🟠 Clyde distance > 8 ? chase : flee distance > 8 ? pos : :fuite

The structure that shows up everywhere: case

def cible(nom, pos, distance = 20)
  case nom
  when "blinky" then pos
  when "pinky"  then [pos[0] + 4, pos[1] + 4]
  when "clyde"  then (distance > 8 ? pos : :fuite)
  end            # other name → nil: no rule, no behavior
end

One word → one rule. It's a dispatch, a state machine, a "router" — the same word everywhere.

The 3 player traps

  1. > vs >=: Clyde flees from 8 on — test the edges (8 and 9) like in the video.
  2. Mutating the input (pos[0] += 4) instead of computing a new one: your Pac-Man moved, not your ghost.
  3. Forgetting the default case: without else, an unknown name → nil → the ghost stops. (An agent without a rule is a dead agent.)

The sentence that unlocks everything

AI is well-written ifs. Today's difference? The ifs are written by other machines. The reader is still you.

Never say again

  • "it's magic / it's AI" → it's a rule someone wrote. Read it.
  • "a script isn't AI" → Blinky is a script. And he scared you for 40 years. 😉

SlopCourse Microduck — from slop to clean code.