Planning summary recorded 15 October 2026. This brings together the latest discussion; it is not a new story revision or a validated run sheet.
Purpose to keep in view
Give students the understanding to read the logic, predict its behavior, and make a machine carry out their intentions, even if AI helped write the code. Understanding mechanisms, access, and the ability to change or carry work away matters alongside successful expression. A classroom activity can expose a mechanism without solving the larger problem of dependence on complex systems and their owners.
The opening should establish why this matters before the hands-on exercises. Image-led slides and rich speaker notes remain the preferred format. Keep the distinction between repeatability and reasoning about behavior under stated conditions.
Current story
Chloe and Bridget, revision 3 is the latest five-minute draft. It adds the museum scene: Chloe recognizes pencil strokes in an inspiring work and returns home to try a branch herself. No real artist, specific artwork, or museum has been selected. The earlier storyboard and revision 2 are retained as history, not competing current scripts.
The story should show people, actions, growing artistic power, intimidation, and changing access. It should not resolve the systemic problem with a simple lesson or assert complete control. Chloe and Bridget are fictional.
Paper-first workshop
The latest direction emphasizes a paper experience that reveals mechanisms without requiring heavy tools. The earlier BeagleY-AI/candy-bar transfer-learning investigation remains background material, not a required workshop component. The older one-hour schedule therefore needs revision and rehearsal; it is not the current timed plan.
Next-word hat
- Start with a shared prompt. Everyone proposes one next word or STOP based on the current text.
- Collect slips, randomly draw one, append the selected word, and clear the hat. If STOP is drawn, end the run; also agree a word cap in advance.
- Ask everyone again after each selected word. Suggestions must respond to the new context; do not prewrite a whole sentence and cut it up.
- Repeat the same process from the same initial prompt for two full runs. A third is optional if timing allows. Different paths can arise without changing the rules or deliberately choosing a different first word.
- This illustrates sampling during generation, not model training. People are not a neural network, and model tokens are not always words. Temperature is not required for this exercise.
A phone/URL version remains a teacher-toolkit wish, not an implementation commitment.
Procedural card activity: selection still open
The goal is to start with shuffled cards and follow explicit simple operations toward an explainable, predictable result. Avoid relying on an undisclosed preset card or implying that a trick demonstrates Turing completeness.
Candidates researched for rehearsal:
- NRICH Spot the Card: a nine-card adaptation, dealt into three columns. Collect the indicated column in the middle and repeat once; the chosen card reaches the fifth position. The nine-card mechanics and collection order need clear facilitator instructions and rehearsal.
- NRICH Card Trick 1: begin with a known count of face-up cards, mix, remove that many cards and flip the removed packet. The two packets then contain equal numbers of face-up cards, an invariant to explain.
- Colorful Piles, Bridges 2018: another invariant-based alternative using balanced colors and matching dealing operations.
No final card activity is selected and no classroom timing is verified. A sorting algorithm was suggested as another possible way to turn a shuffled input into a known arrangement, but a full-deck bubble sort may consume too much workshop time.
Magic reveal: exploratory
An idea under discussion is to turn the paper group's generated text into an image that appears unexpectedly in a later slide, then reveal the mechanism and connect it with the Mechanical Turk story. Visible typing could weaken the surprise.
Audio capture is a candidate, not an established solution. Triggering a capture, reading back the finished text, recognizing keywords, or capturing one word at a time remain undecided. Future tests should check microphone placement, room noise, transcription fidelity, missed or false triggers, generation latency, and a fallback if the system fails. No recording system, keyword listener, image-generation pipeline, or live-slide integration is commissioned by this note.
Next preparation decisions
Rehearse the procedural activity and the hat, select the card mechanism, revise the one-hour allocation with room for discussion and training economics, and decide whether the optional reveal is worth its complexity. Preserve the original notes as dated records rather than silently rewriting earlier decisions.