Video Game AI Lesson Plan (Grades 6-8): How It Works

A student raises their hand during a discussion on how AI works. “But the game already knows me,” they say, with genuine certainty. “It gets harder when I’m too good and easier when I’m losing. It learned my style.” Half the class nods. The other half is already typing something else. This moment happens in grade 6, 7, and 8 classrooms every week — not because students are wrong to be curious, but because nobody handed them the vocabulary to tell the difference between a rule and a mind. That confusion is a literacy gap. And video game AI is the perfect no-fear entry point to close it.

This is not a discipline problem or a tech problem. Students who play games daily come in with genuine intuitions about intelligence and adaptation. The lesson this post outlines gives those intuitions a name, a correction, and a framework they can apply to every AI system they meet next.

TL;DR — how does AI in video games actually work? Most game AI is not a learning system. It is a set of hand-written rules that react to player inputs in real time. Ghosts in Pac-Man follow fixed targeting formulas. Enemies in stealth games run behavior trees a developer typed out. Even “adaptive difficulty” systems — the ones that feel like they know you — are tuning dials following thresholds, not neural networks training on your playstyle. Real machine-learning AI trains on data and updates its internal model. Game AI almost never does that. That difference is one of the most important concepts in AI literacy for grades 6-8.

Is the AI in video games real AI?

Mostly no — and that answer is the whole lesson. Most game AI is hand-authored rules: finite-state machines, behavior trees, and scripted decision sequences a developer wrote in code. It is not a learning system like ChatGPT or a recommendation algorithm.

The clearest example in gaming history is the original 1980 Pac-Man. The four ghosts are famous for feeling unpredictable, but each one runs a single fixed targeting rule. Blinky always chases Pac-Man’s current position directly. Pinky targets four tiles ahead of wherever Pac-Man is facing. Inky calculates a point based on both Blinky’s position and Pac-Man’s direction. Clyde chases directly — but only until he gets close, then flees to the bottom-left corner. Four rules. Zero machine learning. Zero training data. The ghosts do not remember your last game, your preferred routes, or your reaction speed. They run their formula, every frame, all game long.

This distinction maps directly onto AI4K12 Big Idea #2 (Representation and Reasoning — intelligent agents use symbolic representations and rules to make decisions) and onto ISTE 1.5.d (students understand the basic concepts of algorithmic thinking). For a full lesson that pulls apart narrow AI, general AI, and what “intelligence” actually means in each, the Types of AI Deep-Dive unit breaks it down with classroom-ready slides and discussion questions for grades 6-8.

How game AI actually works: four techniques

Four techniques cover the vast majority of game AI students will ever encounter. The table below gives you a real game, a plain-English description of what it does, and the answer to the question students always ask.

Technique	Real game example	What it does	Does it actually “learn”?
Pathfinding	Pac-Man ghosts / A* algorithm in any RTS	Finds the shortest route to a target on a grid or map	No — follows a routing algorithm, every step recalculated fresh
Decision-making (planning)	F.E.A.R. (2005) Goal-Oriented Action Planning (GOAP)	Picks short action sequences at runtime — flank, take cover, reload — based on the current game state	No — searches through a pre-set list of actions and conditions a developer defined (source)
Difficulty adjustment	Left 4 Dead “AI Director” (2008) / Mario Kart rubber-banding	Changes spawn rates, item strength, or pacing to match a player’s current performance	No — follows tuning rules with thresholds; if player health drops below X, ease back; if too high, increase pressure (source)
Procedural generation	No Man’s Sky (2016), ~18 quintillion planets from one numeric seed	Builds worlds, levels, and creatures from mathematical formulas applied to a starting value	No — generates from deterministic formulas, same seed always produces the same output (source)

The phrase students keep using — “it adapts to me” — deserves a precise answer. Adaptation in game AI means a rule fires when a threshold is crossed. The Left 4 Dead AI Director monitors your health, item pickups, and pace in real time and adjusts spawn rates accordingly. That is not learning. Learning, in the technical sense, means a system updates its internal parameters based on new data. The Director’s parameters never update. They were set by designers before the game shipped. What changes is which rule fires, not the rule itself.

This is the same distinction that separates a thermostat from a recommendation algorithm — and it is worth naming out loud for students. For the deeper machine-learning mechanics behind systems that do actually learn, the How AI Works MEGA unit walks through 16 lessons on machine learning concepts, including training data, parameters, and feedback loops — everything game AI skips.

A 45-minute unplugged video game AI lesson (grades 6-8)

No devices required. The pacing below works for a standard single period and scales to 50 or 55 minutes by extending the card-sort discussion.

Time	What students do
0–5 min	Hook: Call out a game title. Ask: “Does the AI in that game know you? Does it remember your moves from last session?” Students share instincts — no right answer yet.
5–15 min	”Be the Ghost” role-play: Each student gets an index card with one targeting rule (Blinky = chase directly; Pinky = aim 4 tiles ahead; Clyde = chase until close, then flee). Students move on a floor or paper grid toward a “Pac-Man” marker using only their card’s rule. They feel exactly how a finite-state machine works — no communication, no memory, no learning.
15–30 min	Paper-grid pathfinding trace: Students receive a hand-drawn 8x8 grid with walls and a start/end point. Task: trace the shortest path using only cardinal moves, then compare routes with a partner. Debrief: “Could you do this without thinking? Could a rule do it without thinking?“
30–40 min	Sort the games: Four game-card strips (Pac-Man, F.E.A.R., Left 4 Dead, No Man’s Sky) — students sort them to the four technique labels from the table above. Small-group discussion: what evidence from the game description supports their sort?
40–45 min	Exit ticket: “Name one way game AI is NOT like a human brain.” Collect as the warm-up anchor for the next lesson.

A grade-7 teacher could run the ghost role-play with desks pushed back or on a hallway floor grid drawn in painter’s tape. A typical class finishes the sort in 8-10 minutes, leaving time for a full-class debrief on the exit ticket question.

Standards the lesson hits, by anchor code: the role-play and pathfinding trace address ISTE 1.5.d (algorithmic thinking — students understand how sequences of instructions perform tasks); the sort-the-games discussion addresses CCSS.ELA-LITERACY.SL.7.1 (collaborative discussions where students build on others’ ideas) and ISTE 1.3.d (build knowledge by actively exploring real-world issues through investigation). For a full set of activities in this format, the Unplugged AI Activities pack includes 10 no-tech lessons with ready-to-print card sorts and role-play instruction cards.

What students get wrong about game AI (and how to correct it)

Three misconceptions show up reliably across grades 6-8. Each one has a direct correction that takes about 90 seconds to address.

• “The game is sentient / thinking.” Correction: the game is running rules. A finite-state machine checks its current state and fires the matching action. There is no awareness, no intention, no experience. A traffic light does the same thing on a larger clock.
• “The NPCs learn my style.” Correction: most game AI reacts to your current inputs — your position, your health, your last action. It does not remember your playstyle across sessions or train on your moves. The AI Director in Left 4 Dead cannot recall that you always rush the finale; it only knows what your health is right now.
• “Harder difficulty = smarter AI.” Correction: difficulty is a tuning dial. On Hard mode, an enemy may have more health, faster reaction timers, or more accurate aim — all of which are numbers a developer typed. The AI is not more intelligent; it has more favorable parameters.

The “does it learn?” question ties directly to AI4K12 Big Idea #3 (Learning — machine learning systems train on data, updating their internal model to improve performance). Real learning systems do that. Most game AI does not. Drawing that line clearly is what separates students who use AI tools thoughtfully from students who credit chatbots with feelings or blame game NPCs for cheating.

For a lesson that goes deeper on what genuine machine-learning systems do — and how they differ from rules-based systems — the companion post how generative AI works (grade 7 lesson plan) walks through training data, parameters, and prediction in student-accessible language. And if students start asking why game NPCs don’t just act more autonomously — making their own plans and pursuing their own goals — the post on AI agents explained for middle school covers the Goal → Plan → Act loop that distinguishes a true agent from a scripted NPC.

Standards crosswalk for this video game AI lesson

Lesson step	Standard(s)
“Be the Ghost” FSM role-play	ISTE 1.5.d · AI4K12 Big Idea #2
Paper-grid pathfinding trace	ISTE 1.5.d
Sort-the-games small-group discussion	CCSS.ELA-LITERACY.SL.7.1 · ISTE 1.3.d
”Does it learn?” exit ticket + debrief	AI4K12 Big Idea #3

Standards source: ISTE Student Standards. ISTE is a registered trademark of the International Society for Technology in Education. These resources are not affiliated with or endorsed by ISTE.

The literacy gap here is real and specific. Pew Research reports that 85% of US teens play video games, with 41% playing daily — which means almost every student in a grades 6-8 class walks in with strong intuitions about how game AI behaves and almost no vocabulary for what is actually happening under the hood. That is not a failure of attention or curiosity. It is a gap nobody handed teachers the tools to fill. Game AI is a low-stakes, high-engagement on-ramp: the systems are concrete, the examples are familiar, and the “does it learn?” question has a clean, demonstrable answer. Get the misconceptions named, get the four techniques in front of students, run the ghost role-play once — and the conceptual foundation for every AI conversation that follows is suddenly sturdier.

When you are ready to go deeper, the shop carries full units on machine learning, AI types, unplugged activities, and more — all PDF, standards-documented, and ready to print for tomorrow.

This post was drafted with AI assistance and human-finalized.