Skip to content

Unplugged Backups

The Wi-Fi will go down. A site will get blocked. A Chromebook cart will be double-booked.

This isn't a disaster plan — unplugged activities are genuinely excellent teaching, and this curriculum schedules some of them on purpose. But when something breaks at 9:03 on a Friday, you need a real lesson in your hand within sixty seconds.

The go-bag

Keep one box in the classroom closet all year:

  • The printed binary card set (16 · 8 · 4 · 2 · 1)
  • 6–10 plastic cups
  • A deck of playing cards
  • Scratch paper and pencils
  • Printed copies of this page

Total cost: about four dollars. It has saved more class periods than any other four dollars you will spend.

Pick by concept

If the lesson was about… Run this Time
How computers store things Binary cards 20–40 min
Sequence, precision, instructions My Robotic Friends (cups) 25–40 min
Algorithms, efficiency Sorting networks 30–45 min
Searching Binary search with cards 20–30 min
Loops Human loop / dance notation 25–35 min
Conditionals If-this-then-that card game 25–35 min
Variables The cup and the sticky note 20–30 min
Functions Recipe abstraction 30–40 min
Debugging Broken instructions 30–40 min
Networks / the internet Paper packet routing 35–45 min
AI and training data Human classifier 30–45 min
Bias in AI The lopsided training set 30–40 min
Language models Next-word prediction 25–35 min
Data representation Pixel grid drawing 30–40 min

The activities

Binary cards

Concept: everything in a computer is ones and zeros

Five cards showing 16, 8, 4, 2, 1 dots. Card facing out = 1, hidden = 0.

Show 1, 2, 3, 4, 5, 7. Then ask for the biggest possible (31) and smallest (0). Then write the pattern on the board.

Extend: A=1 … Z=26, students write their initials in binary and trade to decode. 🔴 Extender question: With 8 cards, how many values? (256 — one byte, and the reason so many things in computing stop at 255.)

📄 Full printable set: Binary cards handout

My Robotic Friends

Concept: computers do exactly what you say, not what you mean

One student is the robot. The class gives instructions to stack cups into a pyramid using only four commands:

FORWARD    TURN LEFT    PICK UP    PUT DOWN

The robot must obey literally. "Move forward" with no distance means the robot walks into a wall. This gets very funny, very fast, and the debrief writes itself:

"The robot did exactly what you said. That's not a robot problem. That's a specification problem — and it's most of programming."

Round 2: groups write their algorithm on paper, trade with another group, and execute someone else's code. Watching your own instructions fail in someone else's hands is the entire lesson.

Sorting networks

Concept: algorithms, parallelism, efficiency

Tape a sorting network on the floor (search "CS Unplugged sorting network" for the layout). Students hold number cards, walk the network, and compare at each node — smaller goes left. They come out sorted, and nobody was in charge.

Debrief: "Nobody could see the whole list. Everyone made one tiny comparison. How did it end up sorted?" That's a distributed algorithm.

Needs floor space — a hallway works.

Binary search with cards

Concept: why sorted data is worth the trouble

Lay 16 numbered cards face down, in order. Student A picks a target. Student B finds it, guessing one card at a time.

Then shuffle and repeat unsorted. Count the guesses both times.

Debrief: sorted takes at most 4 guesses; unsorted takes up to 16. Ask 🔴 Extenders: "How many guesses for a thousand cards?" (10.) "A million?" (20.) That doubling is the whole point.

Human loop

Concept: loops replace repetition

Have a student write instructions for walking a square. They'll write eight lines. Then: "Write it in three lines." They discover repeat 4 times.

Extend: choreograph a short dance in loop notation and have another group perform it from the notation alone.

If-this-then-that card game

Concept: conditionals

Each student gets a rule card:

IF the card is red THEN clap · ELSE stand up

Deal cards from a deck one at a time; students execute their own rule. Then combine two rules with AND / OR and watch it get hard fast.

Debrief: nested conditions are hard for humans and easy for computers — which is exactly why we write them down.

The cup and the sticky note

Concept: variables

A cup is a variable. The sticky note on the front is its name. What's inside is its value.

Physically demonstrate: put a 5 in the cup labelled score. Add 3 — take out the 5, put in an 8. The old value is gone. That's assignment, and it's the thing beginners get wrong most often.

Extend: score = score + 1. Do it slowly with the cup. Read the right side first. This single demo prevents weeks of confusion.

Recipe abstraction

Concept: functions

Write out "make a sandwich" in full detail. It's long.

Now write three recipes that all include it. Ugly.

Now name it MAKE_SANDWICH() and just call it. That's a function — and the point isn't that it's shorter, it's that you can now change the sandwich in one place.

Broken instructions

Concept: debugging

Hand out a paper "program" with 2–3 deliberate bugs: an off-by-one, a missing step, a step in the wrong order. Students execute it literally, find the bugs, and — importantly — write down how they found each one.

Debrief: collect the strategies on the board. That list is a debugging method, and the students built it.

Paper packet routing

Concept: how the internet actually works

Students are routers, arranged around the room. A message is torn into 4 slips, each labelled with a destination and a sequence number. Slips are passed router to router.

Then make it real: crumple and discard one slip. What happens? The receiver notices #3 is missing and requests it again. That's TCP.

🔴 Extender: hand two students the same destination and let them figure out congestion.

Human classifier

Concept: training data and machine learning

  1. Teacher secretly picks a rule (e.g., "things with wheels").
  2. Show examples one at a time, labelling each ✅ or ❌.
  3. Students guess the rule after each example.
  4. Around example 6–8, someone gets it.

Debrief: "You just did what a machine learning model does. You never saw the rule. You inferred it from labelled examples. That's training."

Then the twist: show a genuinely ambiguous example (a wheelbarrow? a suitcase?). The class will disagree. "That's exactly where AI systems fail, and it's why they can be confidently wrong."

The lopsided training set

Concept: bias

Run the human classifier again — but every example of "dog" you show is a golden retriever.

Then show a chihuahua.

Students will hesitate or say no. Nothing was wrong with their reasoning; the data was lopsided.

Debrief — the key move: don't let this land as "AI is bad." Land it as:

"The model wasn't broken. It learned exactly what we showed it. The problem was upstream — in what we chose to show. So whose job is it to notice that?"

Next-word prediction

Concept: how language models work

Write on the board: The cat sat on the ___

Take votes. "Mat" wins. Write it. Continue, word by word, class voting each time.

After ten words, stop:

"There is no plan. Nobody in this room knows how this sentence ends. Each word is just the most likely next word given everything before it. That's what a language model does — at enormous scale, with far better statistics than us."

Then the crucial follow-up: "Notice it never checked whether any of this is true. It only checked what sounds likely." That is the cleanest explanation of hallucination a 5th grader will ever get.

Pixel grid drawing

Concept: images are numbers

Graph paper. One student draws a simple shape by filling squares, then writes it as run-length encoding: 3 white, 2 black, 4 white…

Partner reconstructs the image from the numbers alone.

Debrief: that's a black-and-white image file — and run-length encoding is real compression. 🔴 Extender: "How would you add colour?" (More numbers per pixel — three, for red/green/blue.)

Running an unplugged session well

  • Say why. "The internet's down, so we're doing the version of this that I actually like better." Never apologise — the moment you frame it as a downgrade, they treat it as one.
  • Get bodies moving. The advantage of unplugged is physical. Use the floor, the hallway, the whole room.
  • Older students run the activity. Hand a 9th grader the instructions and let them lead. Frees you to watch, and it's the near-peer model in action.
  • Debrief hard. Unplugged lessons live or die on the last five minutes. Always close the loop back to the computing concept explicitly.
  • Keep the artifacts. Paper algorithms and pixel grids go on the wall. They're good evidence of learning for anyone who asks.

Sources for more