brightlab.Made for curious mindsBrightlab
Foundation · Instructions
Help Pip reach camp. Put the arrows in order, then try one at a time.
Go to the investigation ↓
Investigate with Pip the possumAn arrow program with a drawn prediction, marked failure and revised route.
Help Pip reach camp. Put the arrows in order, then try one at a time.
Our rover follows a list of instructions. Forward means one square in the direction its arrow points. Left and right turn the arrow without moving the rover. It does not know that we want to reach camp. When a route fails, we inspect the instructions and change them. This is an ordinary programmed machine, not a trained AI model. AI systems also depend on people specifying how their outputs become actions.
Before this lesson: Point forward, then turn your body without moving your feet. No prior AI lesson is required.
Our rover follows a list of instructions. Forward means one square in the direction its arrow points. Left and right turn the arrow without moving the rover. It does not know that we want to reach camp. When a route fails, we inspect the instructions and change them. This is an ordinary programmed machine, not a trained AI model. AI systems also depend on people specifying how their outputs become actions.
Point forward, then turn your body without moving your feet. No prior AI lesson is required.
Open the route lab and reset it. Tape a 5 by 5 grid on the floor, mark a rock at row 3, column 3 and place a camp at the upper right. Prepare forward, left and right paper arrows. Check children can turn in place safely; use a toy if space is limited.
ACARA V9 AI curriculum connection ↗ · Technologies ↗
A complete teaching sequence · 35 minutes
Ask a partner to get the toy to camp using only three instruction words. Execute literally, including turns in place.
Ask: “If I turn right, have I moved a square?”
Listen for: “No, but I am looking a different way.”
Read the default program without running it. Children draw an endpoint or point to a square; an adult records their words.
Ask: “Where will the rover stop, and which arrow makes you think that?”
Listen for: “It goes forward twice, then turns.”
Step the browser program one instruction at a time. Follow the highlighted instruction and mark visited squares on paper. Swap navigator and operator after four steps.
Ask: “Which instruction happened just before the rover hit the rock?”
Listen for: “That forward instruction; the turn was too late.”
Load the turned-start edge case with the same instructions. Compare endpoints.
Ask: “The words stayed the same. Why did the journey change?”
Listen for: “It started looking a different way.”
Edit the route to reach camp without crossing the rock. Save the original trace, reset the position and replay the revised program. Children dictate a reason for one edit.
Ask: “What will you change first, and what will you keep the same?”
Listen for: “I will move the turn earlier and keep the starting square.”
Each pair shows one predicted step, the failed step and its repair. Collect an annotated route.
Ask: “Did the rover guess what you meant?”
Listen for: “No; we had to change the instructions.”
The machine knows what I meant.
Keep the program fixed but turn the starting arrow. The same list reaches a different endpoint because initial direction is part of the state.
Write a route that reaches camp without the rock; then swap with a pair who must run it literally. Repair the first failing instruction.
Watch one child predict a turn without moving their counter. If they confuse a turn with a sideways step, replay only that instruction physically. Credit an accurate explanation of a failed route, even if the whole route is unfinished.
| Criterion | Beginning | Secure | Extending |
|---|---|---|---|
| Sequence reasoning | Names actions without an order | Predicts the next move correctly | Predicts the route including direction changes |
| Debugging | Changes several arrows without a reason | Identifies and repairs one failing step | Uses the trace to justify a minimal repair and retests |
Use three instructions and a 3 by 3 paper grid; allow pointing and adult scribing. Use tactile arrows for children who cannot access the screen.
Find two safe routes and compare instruction counts. Ask whether the shorter program always makes the shorter physical journey.
No names or locations are needed. Keep the floor route clear and offer a seated equivalent. Ask a trusted adult if another application requests a photo or login.
Shared device? Turn remembering off. A project file lets you continue on another device.
Your browser is the laboratory
Help Pip reach camp. Put the arrows in order, then try one at a time.
A 5×5 map starts at column 1, row 5, facing north. A rock is at (3,3); camp is at (5,1). F moves one square; R and L turn in place. First predict the four-step program FFRF.
The experiment opens after your prediction.
There is no penalty for being surprised.
Apple Silicon · PyTorch MPS
Execute two route programs over every grid position and starting direction on MPS/CUDA; group failures by first blocked step. Repeated state sets increase workload, not independent evidence.
Teacher-operated extension: students predict and interpret the visual report together.
The bundle contains lesson-specific working code, a configuration file, a reactive notebook, a deterministic CPU check and hardware setup instructions. Acceleration is reported only after a tensor operation and result read-back succeed.
Download Mac Studio investigation ↓unzip f-trail-mac-pathway.zip -d f-trail-mac cd f-trail-mac bash setup-mac.sh source .venv/bin/activate python experiment.py --device mps --output results marimo edit notebook.py
CPU and available-device execution status is recorded in the downloaded README and validation report. DGX Spark execution requires that hardware; static validation alone does not establish GPU compatibility or performance. The browser lesson remains fully available without this extension.
NVIDIA DGX Spark · PyTorch CUDA
Execute two route programs over every grid position and starting direction on MPS/CUDA; group failures by first blocked step. Repeated state sets increase workload, not independent evidence.
Teacher-operated extension: students predict and interpret the visual report together.
The bundle contains lesson-specific working code, a configuration file, a reactive notebook, a deterministic CPU check and hardware setup instructions. Acceleration is reported only after a tensor operation and result read-back succeed.
Download DGX Spark investigation ↓unzip f-trail-dgx-pathway.zip -d f-trail-dgx cd f-trail-dgx bash run-dgx.sh # Open the localhost notebook URL printed by the container.
CPU and available-device execution status is recorded in the downloaded README and validation report. DGX Spark execution requires that hardware; static validation alone does not establish GPU compatibility or performance. The browser lesson remains fully available without this extension.
An arrow program with a drawn prediction, marked failure and revised route.
Download the editable handout →