Shutting Down A.I.

Overview

In this science-fiction escape room, students use critical thinking strategies to review the depth study “AI – Recreating the Human Mind” and collaboratively solve six puzzles to stop an AI takeover. Students rotate through puzzles, using codes and deductions to reach the final “shutdown” instruction.

Learning intentions

Students will:

• WALT review key ideas about AI, including how it “learns”, limits, bias, and human-like cognition
• WALT use pattern recognition and deduction to solve puzzles and build reliable reasoning
• WALT communicate justification clearly and check answers against clues
• WALT assemble correct codes to complete the final shutdown step

Success criteria

Students can:

• I can answer knowledge questions accurately and use the answer set to form a code
• I can recognise relationships between AI progression levels and represent them correctly
• I can decode a cipher using AI-related terms as the key
• I can use clues from previous puzzles to make a logical final deduction
• I can combine all codes to produce the correct shutdown phrase

Curriculum links

• MA5-FNC-P-01 — Students use function notation to describe and graph functions of one variable and graphs inequalities in one and 2 variables (reviewed through data/logic reasoning tied to function-style representations of AI “progression levels” used in puzzles)
• Students practise interpreting representations and building explanations, aligned to the outcome’s “functions and other graphs” reasoning mindset (how variables change, what outputs mean)

Lesson structure (60 minutes)

1. 0–5 min · Narrative Hook. Teacher reads the scenario: the class has access to an offline “mind-matrix” but the AI will self-upgrade in 15 minutes. Students must solve six mission puzzles to shut it down safely. Students repeat the mission goal and listen to rules: collaborate, justify, and record reasoning.

2. 5–12 min · Direct teaching: Critical thinking moves. Teacher models two think-alouds: (a) “If my answer is X, why does it match the clue?” (b) “Check for bias/assumptions.” Students complete a 1-minute cold call: “What does ‘AI learning’ mean in plain language?” then quick discussion.

3. 12–44 min · Escape room rotation: 6 hybrid puzzles (teacher-guided). Teacher splits class into 5 teams (about 5 students each) and assigns a start puzzle; teams rotate when time or checkpoints hit. Students solve each puzzle, then record the code for that puzzle on their mission sheet.

Puzzle 1 (Knowledge quiz → Code 1) | 6–8 min

• Puzzle (display/print): Answer the following about AI. Write the letter of the correct option.

1. Which best describes machine learning? A Learns rules by explicit programming only B Learns patterns from data to make predictions C Thinks like a human without data D Stores memories like a brain
2. A key risk of AI systems is: A No risk of bias B Bias from training data can affect outputs C Perfect accuracy always D No need for evaluation
3. A “model” in AI is best thought of as: A A spreadsheet with no calculations B A system that maps inputs to outputs C Only a hardware component D A random number generator
4. Evaluation of an AI system primarily helps: A Prove it is always correct B Measure performance and identify errors/limits C Remove the need for data D Replace all human judgement

• Solution + explanation (for teacher/answer key): 1) B, 2) B, 3) B, 4) B. Map to letters: B B B B. Code 1 = BBBB (because all correct options are B).
• Hint card: “Look for the definitions: learning from data, bias from data, model as input→output, and evaluation as measuring performance.”
• Links to next: Teacher unlocks Puzzle 2 only when Code 1 is written on the team sheet.
• Required knowledge: machine learning definition; bias; model as input-output mapping; purpose of evaluation.

Puzzle 2 (Pattern recognition using AI progression levels → Code 2) | 6–8 min

• Puzzle: A lab document shows “progression levels” of AI capability. Match each level to its typical change in outputs. Levels: L1, L2, L3, L4 Statements: P) Outputs improve mainly by learning patterns from more/training data Q) System rules become more data-driven than hand-coded R) Performance becomes more reliable but still has error margins S) Behaviour appears flexible, yet may still reflect bias/limitations Task: Write the correct pairing in order L1→L4 (letters).
• Solution + explanation: L1→Q, L2→P, L3→R, L4→S. Code 2 = QPRS (just the chosen letters).
• Hint card: “Think: early = rule-like, middle = learns patterns, later = more reliable, advanced = flexible but biased/limited.”
• Links to next: Code 2 letters become the key for Puzzle 3’s cipher order.
• Required knowledge: progression concepts (data-driven learning, reliability vs errors, bias).

Puzzle 3 (Cipher using AI terms as key → Code 3) | 6–8 min

• Puzzle: Cipher text: “K-NO-RA-ML” Key (use AI terms in the order of Code 2): Bias, Data, Model, Evaluation. Instructions: Replace each hyphen-separated token with the first letter of the matching key word. Then write the resulting 4-letter code.
• How it works (for students): Code 2 = QPRS. Use a mapping given on the sheet: Q→Data, P→Model, R→Evaluation, S→Bias. So K-NO-RA-ML is a decoy; you must use the mapping + key order.
• Solution + explanation: QPRS → Data, Model, Evaluation, Bias → first letters D M E B. Code 3 = DMEB
• Hint card: “Don’t decode the dashes—use Code 2 to order the AI terms, then take first letters.”
• Links to next: Puzzle 4 uses Code 3 letters as placeholders for logic answers.
• Required knowledge: AI terms; mapping from previous code.

Puzzle 4 (Logic deduction using clues from Puzzles 2 and 3 → Code 4) | 6–8 min

• Puzzle: Three security agents predict shutdown settings: Agent A: “If Model improves, Evaluation must check errors.” Agent B: “If Bias exists, outputs can be unfair even when performance is high.” Agent C: “If Data quality rises, capability tends to increase, not instantly become perfect.” Deduction grid:

1. A is true and involves letters from Code 3 that start with M or E.
2. B is true and involves letters from Code 3 that start with B.
3. C is true and involves letters from Code 2 that include R or P. Question: Which statement must be used to form Code 4 as a 3-letter abbreviation? Choose the correct 3-letter abbreviation: A = MEL, B = BIA, C = DPA (Data/Pattern/Able)

• Solution + explanation: From Code 3 starts: DMEB → M and E are present (A fits MEL), B present (B fits BIA), Code 2 includes R and P (C fits DPA). Choose the one that matches all given conditions: Agent B fits item 2 exactly, but condition 1 and 3 also need satisfaction; only the combined abbreviation that best aligns with “bias affects unfairness even with performance” is BIA. Code 4 = BIA
• Hint card: “One clue is a hard match: Code 3 includes a B-start letter, so look for the abbreviation that begins with B.”
• Links to next: Code 4 determines the pattern in Puzzle 5.
• Required knowledge: reasoning about bias, evaluation, and data quality.

Puzzle 5 (Knowledge quiz → Code 5) | 6–8 min

• Puzzle: Choose A/B/C/D.

1. In an AI system, “training” is mainly: A Testing once only B Learning from data to adjust parameters C Manually writing every rule D Storing photos
2. “Bias” means: A Equal outcomes for everyone always B A systematic tendency caused by data or design C No relationship to data D Random answers only
3. A human mind is: A Not modelled by any computation at all B Complex and not fully replicated by AI today C Identical to today’s algorithms D Only a database

• Solution + explanation: 1) B, 2) B, 3) B → Code 5 = BBB
• Hint card: “Training = learn patterns from data; bias = systematic tendency; current AI doesn’t fully replicate human mind.”
• Links to next: Code 5 sets the final assembly length.
• Required knowledge: training purpose; bias definition; limits of AI vs human cognition.

Puzzle 6 (Final assembly: all codes → Shutdown) | 6–8 min

• Puzzle: You must assemble the shutdown phrase using codes in order 1–5. Rule card: “Take the first letter of Code 1, then first of Code 2, then first of Code 3, then first of Code 4, then repeat Code 5’s first letter for remaining slots.” Codes: Code 1 = BBBB, Code 2 = QPRS, Code 3 = DMEB, Code 4 = BIA, Code 5 = BBB
• Solution + explanation: First letters: Code1 B, Code2 Q, Code3 D, Code4 B. Code5 first letter B repeated 3 times to fill 7-letter phrase length: BQDBBBB. Teacher announces: “Shutdown command accepted: BQDBBBB”.
• Hint card: “Use the rule card exactly; don’t overthink the middle letters.”
• Links to wrap-up: Students submit their final code + one sentence of justification.
• Required knowledge: synthesising across AI concepts.

4. 44–55 min · Debrief + cold calls. Teacher cold-calls 3–5 students to explain one puzzle strategy (e.g., “How did you use Code 2 to decode Puzzle 3?”). Students share what clue mattered most and why.

5. 55–60 min · Exit ticket. Students complete: “One limitation of AI is ____ because ____.” Teacher collects for quick checks.

Resources

• Printed puzzle sheets (6 sections) and team mission sheet
• Hint cards for each puzzle (one per team; replenish as needed)
• Code recording template (blank boxes for Code 1–6)
• Timer/rotation plan card
• Answer key for teacher only
• Slide(s) with narrative hook and critical thinking moves
• Coloured pens/highlighters for reasoning evidence
• Optional: “unlock cards” for teacher to hand out per checkpoint

Assessment

• Formative: teacher circulates, checks recorded codes and reasoning sentence quality
• Formative: cold-call explanations during debrief (focus on justification, not just answers)
• Exit ticket: one AI limitation with a justification sentence; check for bias/evaluation/training language

Differentiation

• For teams needing support: provide sentence starters (“I think the answer is __ because __”), and allow one hint card per puzzle; don’t remove choice—scaffold with definitions.
• For struggling readers/EAL: colour-code letters, mapping steps, and use larger font for puzzle instructions; read aloud the Puzzle 4 statements.
• For high achievers: require an extra “reasoning proof” line for each code (“My choice follows because…”), and encourage checking links between puzzles.
• Extension (advanced learners): after shutdown, ask teams to modify Puzzle 2 mapping by proposing an alternative “progression level” description, then predict how Codes 3–6 would change logically (no redoing all work—brief written plan only).