Garbage In, Garbage Out

Overview

Computer scientists have an old saying: garbage in, garbage out. If the instructions are unclear, the result will be too. In this game, a Director secretly draws a picture from a shape challenge and writes instructions, and a classmate Robot follows those words exactly, with no peeking and no questions. Round 1 results are often wildly off. Round 2, after the class builds a "prompt recipe," gets much closer. Students then upgrade vague AI prompts and learn the other half of the truth: even a great prompt can get a wrong answer, so people still check.

Objectives

• Students will write step-by-step instructions that another person can follow exactly, and revise them after seeing the result.
• Students will explain that a generative AI tool fills in missing details with likely guesses from patterns, so vague prompts get unpredictable results.
• Students will identify missing ingredients in a vague prompt (who it's for, what, how many or how long, details, style) and rewrite it.
• Students will explain why a clear prompt still doesn't guarantee a correct answer.

Step by step

1. 10–5 min

   Hook

   Program the teacher

   Stand at the board and say: "I am a robot. I do exactly what you say, nothing more." Ask the class to tell you how to draw a cat. When a student says "Draw a cat," draw a tiny circle in the corner and stop. "Draw ears" gets two ears floating far away from the head. Keep following the words literally. After a minute, ask: "Why doesn't my cat look like the cat in your head?" Take answers, then write Garbage in, garbage out on the board.

   Facilitator noteStay deadpan. The laughter is the hook, but the point is that the robot can't read minds; it only has the words.

2. 25–10 min

   Model

   How AI fills in the gaps

   Explain simply: "A generative AI chatbot learned from huge amounts of writing. When you type a prompt, it predicts the words that are most likely to come next, one after another. It doesn't know what's in your head. When your prompt leaves something out, it fills the gap with a likely guess." Then read the Robot Rules aloud: the Robot does only what the words say; no peeking; no questions; the Director can't point or gesture. Show a Shape Challenge Card and model writing two precise instructions: "Draw a big square in the middle of the page, about as wide as your hand."

   Facilitator noteStudents often ask whether AI "understands." A fair answer for this age: "It's very good at predicting what words usually come next. That can look like understanding, but it can also be confidently wrong."

3. 310–22 min

   Practice

   Round 1 and Round 2

   Round 1 (5 min): Directors take a Shape Challenge Card, hide their paper behind a folder, draw their picture, and write instructions in the Round 1 box of Handout B. Robots read and draw in the Round 1 drawing box, following exactly. Reveal and compare. Quick huddle (2 min): Ask pairs what went wrong and build the first version of the Prompt Recipe on chart paper. Round 2 (5 min): Switch roles with a new card. Directors write using the recipe. Robots draw. Compare again.

   Facilitator noteLook for Round 2 instructions that name size, position ("top left corner"), count, and order. Those are the details AI prompts need, too: specifics instead of assumptions.

4. 422–28 min

   Debrief

   Build the Prompt Recipe

   Put two pairs' Round 1 and Round 2 drawings under the document camera. Ask: "What changed between rounds? Which words made the biggest difference?" Finalize five ingredients on the chart: Who is it for? What exactly? How many or how long? Which details matter? What style or format? Then ask: "When your Robot still drew something wrong in Round 2, whose job was it to notice?" (The Director's, by checking the result.)

   Facilitator noteConnect to S3.2: clear instructions are clear communication. The same recipe helps when you ask a classmate for help or explain your idea.

5. 528–36 min

   Apply

   Prompt Upgrade Lab

   Pairs complete Handout C: for each vague prompt, name what's missing using the recipe, then write an upgraded prompt. Share a few aloud. If you have the projected demo, type one vague prompt and the class's upgraded version into the chatbot and compare the answers. Then point out any mistakes or made-up details in either answer and ask: "Did a better prompt make it automatically correct?"

   Facilitator noteRow 5 on Handout C adds wrong information to the prompt ("garbage in"). Students should notice that a tool may repeat the mistake instead of fixing it.

6. 636–40 min

   Reflect

   Exit ticket

   On the back of Handout B, students answer: "One thing I changed in Round 2 was ___. It helped because ___." and "Even with a clear prompt, I still need to ___." Collect as evidence.

   Facilitator noteStrong answers name a specific revision ("I said where on the page") and a checking move ("check if the answer is right").

Adaptations

Grade 3

Use Shape Challenge Cards 1–4 (three shapes each) and limit instructions to five sentences. Do Handout C Rows 1–3 as a class.

Grade 5 / advanced

Add a Round 3 with a "picky Robot" who misreads any word that could mean two things. Directors hunt for ambiguous words like "next to" or "big."

Emergent bilinguals

Post a word bank of position and size words with sketches (top, bottom, left, right, center, above, below, inside, big, small), and let pairs write in either language.

Students with fine-motor needs

Robots build with cut paper shapes instead of drawing.

Pairs well with

• Gen AI Literacy Breakouts (K–8)A teacher-led digital breakout follow-up on how generative AI works.
• AI Essentials Prompt LibraryFor teachers: model prompts that show the same ingredients students list in the Prompt Recipe.