Change One Thing

What you will make

A comparison of two runs of a small, invented model. You will predict what happens, ask AI to inspect the rules, run them yourself, and change one parameter—a setting you can change, such as how far the reference line moves or how many tokens are lost each turn. The result explains the model's behavior and its limits. You can work aloud, on paper, or with counters; saving and sharing are optional.

Before you start

Required: access to a chatbot for the live AI activity. Operate it yourself, direct a partner, or guide a facilitator-operated exchange. No learner account, recording, or saved work is required. A calculator or table is optional; no coding or simulation app is needed.

Open the required 📎 Resource · Model cards. For self-study, inspect one 📎 Resource · Gallery example, then choose an unworked variation for your own run. Zoom and in-person sessions begin with a facilitator example followed by individual, pair, or group practice. Watching alone is preparation. If live AI is unavailable, prepare your question and return for the live step.

Lab materials

• 📎 Resource · Model cards — Two small rule sets, reset conditions, and hand checks for running a comparison.
• 📓 Notebook · Lab Notebook — The matching six-step activity with optional notes and local export.
• 📎 Resource · Gallery — Two constructed walkthroughs with predictions, explicit rules, and numeric comparisons.
• 📎 Resource · Facilitator guide

Activity

Follow the complete activity here or open the optional 📓 Notebook · Lab Notebook. Speak, write, or work with a partner; note fields may stay blank.

1. Choose a question and model

   Open the required 📎 Resource · Model cards. Choose the mountain reference-line model or the ice-token store. Ask a question the rules can answer: “What changes when the shared reference line moves faster?” or “How does a larger fixed loss change the tokens left?”

   For self-study, inspect one 📎 Resource · Gallery example, then use an unworked variation, such as comparing reference changes of 1 and 3, or token losses of 1 and 2. In Zoom or in person, inspect the facilitator's example, then choose your own individual, pair, or group comparison. Repeating the displayed answer is not your own run.

2. State the rules and predict

   Read your chosen 📎 Resource · Model card aloud or copy its small rule block. Name the starting state, update rule, number of turns, and output you will compare. Choose one parameter—a setting you can change, such as reference-line movement or token loss per turn. Keep all other starting values and rules fixed.

   The baseline is your first comparison run. Before calculating, predict what it will do and how the changed run will differ. Say why. These are invented models: mountain coordinates stay fixed while a reference line moves; ice tokens lose a fixed amount each turn. Neither predicts a real mountain, climate, refrigerator, or food-storage outcome.

3. Ask AI to inspect the rules

   Required: access to a chatbot for the live AI activity. Operate it, direct a partner, or choose the request in a facilitator-operated exchange. No learner account, recording, or saving is required. Use the invented values from the 📎 Resource · Model cards; consult 🧩 Module · Safety: Share Less, Notice More before sending other material.

   Ask: “Here is my invented model: [paste the rule block]. I plan to compare [baseline parameter] with [changed parameter], keeping everything else fixed. Audit the rules for ambiguity, or propose one hand check. Do not change assumptions, add physical effects, or run the comparison for me.”

   Inspect the actual response against the card. Accept a useful check, correct a misunderstanding, or reject it. Reject invented additions such as melting rates that vary with weather or peaks that move with the reference line. The live AI response is not evidence that the model works; 🧩 Module · Reality Check helps distinguish a suggestion from a checked result. You must execute and hand-check the rules yourself.

4. Run and check the baseline

   Start from the initial state on your 📎 Resource · Model card, called turn 0. Apply the rule once for each turn. Keep the intermediate values available aloud with a partner, on paper, with counters, or in optional notes. A calculator may help with arithmetic.

   Hand-check at least one transition and the card's zero-change or zero-loss test. If the result does not follow the rule, correct the arithmetic or restate the rule before continuing. If you discover that you changed an assumption, restart the baseline under the stated rules. An AI-generated table alone does not complete this step.

5. Reset and change one parameter

   Return to the same turn-0 state. Change exactly the parameter you chose: the reference increment or the fixed token loss. Keep the turn count, starting values, and all other rules on the 📎 Resource · Model card unchanged.

   Run the same number of turns yourself and hand-check a transition. Compare corresponding turns, not the end of one run with the start of another. If you accidentally continued from the baseline's final state, reset and redo this run.

6. Compare the results and limits

   Compare the baseline and changed result. Give the values, explain the difference using the rule, and return to your prediction. Correct a prediction that did not fit; explaining the correction is part of the work. Use the 📎 Resource · Model card to name something that stayed fixed and something the model leaves out.

   Changing a shared reference cannot change the mountains' physical heights or reverse their ordering under either fixed measurement definition. Ice tokens and turns are arbitrary; their loss says nothing about real refrigeration rates or food safety. For a real-world claim, follow 🧩 Module · Reality Check and the original research links in the 📎 Resource · Gallery. A model result alone is not that evidence.

   Explain what you accepted or rejected from AI and what you checked yourself. Choose to stop, try a new one-parameter comparison, or investigate a real question with sources. A spoken explanation is enough; keeping or sharing a record is optional.

Check your result

• Both runs begin at the same initial state and use the same number of turns.
• Exactly one parameter differs; the rules remain visible and unchanged.
• Your hand check supports the values you compare.
• Your explanation distinguishes model behavior from a claim about the world.

You are done when

• You chose a question, stated the rules, and predicted before calculating.
• You participated in a live AI rule audit or check request and inspected its response.
• You ran and hand-checked a baseline, reset, and ran a comparison with exactly one changed parameter.
• You compared numeric results with your prediction and explained a model limit.
• You named your own checks and the AI contribution, then chose what happens next.

Try another version

• Try another value of the same parameter after making a new prediction.
• Use the other Model card and explain what its rules leave out.
• Follow an original Gallery research link and identify evidence a real-world model would need.