The Winning Number Game
Activity: Have students test numbers and try to discover the rule that makes a number win.
AI relevance: AI can become better at predicting outcomes without developing a more accurate understanding of why they happen.
Before the reveal: Students see only a number input, winner/loser feedback, their test history, and a place to propose a rule; ask for both an explanation and a prediction strategy before revealing the rule.
Reveal: the principle and how the activity works
Principle: Better prediction is not the same as a more accurate understanding of the process producing the result.
What to compare: Compare students’ ability to predict a winner with the accuracy of their proposed explanation: did their predictions improve even while their account of the rule remained wrong, and which new examples broke the shortcut?
How it works:
The hidden rule compares the current number’s English name with the previous number’s English name. The current number wins if its name comes later in alphabetical order. It is not a rule about numerical size. The app uses an English text comparison.
At the start, and after Reset, the previous number is 0 ("zero"). Test 9: "nine" comes before "zero", so it loses. Then test 2: "two" comes after "nine", so it wins. Then test 3: "three" comes before "two", so it loses. The order of guesses matters.
Repeating the immediately previous number keeps its last result. A first test of 0 loses. The app spells numbers from 0 to 999 in English, using forms such as "twenty-one" and "one hundred one" (without "and").
The AI lesson: predictions depend on the data and clues a learner can use. Someone looking only at numerical properties may find a shortcut that works for some guesses while missing the English-name relationship entirely. That shortcut can fail when tested with a different sequence. For AI, more examples or more computation can help, but success is not guaranteed if the information needed for the right explanation is still missing. Ask separately: "What is the actual rule?" and "What strategy helps you predict?"
Keep the analogy clear: the game gives people the numbers and their sequence, but people may not think to use their English names. The clue is available but overlooked. An AI may instead lack a relevant clue in its data, fail to represent it usefully, or fail to learn its importance. These are different ways a learner can miss what matters; the game illustrates the problem rather than showing that all AI has the same limitation.
Next-Word Stories
Activity: Have students build shared stories by adding a few words at a time while seeing different amounts of earlier text.
AI relevance: More context helps people write coherent continuations, much as transformer attention helps AI draw on relevant earlier text to predict what comes next.
Before the reveal: Run the three rounds in order, showing the last 3 words, then the last 15, then the whole story; let students compare the resulting stories before discussing attention.
Reveal: the principle and how the activity works
Principle: Coherence depends on access to relevant context and on using earlier information to connect the next words to the story.
What to compare: Compare the 3-word, 15-word, and full-story rounds for consistent characters, events, and meaning: which connections became possible when earlier text was visible?
How it works:
Round 1 shows the last 3 words; Round 2 shows the last 15; Round 3 shows the whole story so far. The rounds let learners experience how access to earlier text can make a story more coherent: a continuation can fit the last few words yet conflict with a character, event, or idea introduced earlier.
Ask students to point to examples of connections they could preserve with more context. Which earlier words mattered to their next choice? Did they need to treat every earlier word as equally important? Use those observations to introduce transformer attention: the model gives different parts of the available context different weight when forming the information used to predict the next token.
The parallel is the usefulness of relevant context, rather than identical human and machine processes. People write the continuations here; the activity does not compute transformer attention. Coherence can improve with context, but the classroom results are something to examine, not a guaranteed improvement in every story.
Attention Driving
Activity: Have students drive with different levels of automated help while completing a competing task.
AI relevance: Reliable AI can encourage people to shift their attention elsewhere, making rare mistakes harder to notice and correct.
Before the reveal: Students experience the assigned driving conditions and competing task before comparing driving outcomes and discussing when they could intervene.
Reveal: the principle and how the activity works
Principle: The safety of an AI-assisted activity depends on human attention and the ability to intervene as well as on the AI’s usual reliability.
What to compare: Compare manual, advisory, and increasingly automated driving on both driving performance and the competing task: did more automation free attention while making rare interventions harder?
How it works:
The activity demonstrates how trusting automation changes attention and the opportunity to catch mistakes. Compare the dashboard’s driving outcomes across cars with the attention taken by the competing task. Use the experience to discuss whether a person can notice and respond to a rare problem when reliable automation has encouraged them to look elsewhere.
Think Like the AI
Activity: Have students choose helpful answers to questions after seeing different clues about the person asking.
AI relevance: AI uses context to shape its answers, but a clue that influences an answer may not be a good reason for changing it.
Before the reveal: Students see a question, a context cue, and shuffled response options; compare choices across the randomized contexts before discussing which clues should matter.
Reveal: the principle and how the activity works
Principle: Context changes what seems relevant, but helpful adaptation and unjustified assumptions can arise from the same use of contextual clues.
What to compare: Compare answers to the same question under different context cues: what changed in the preferred answer when only the information about the person changed?
How it works:
The activity demonstrates how clues about a person can change what seems like a helpful answer. Context is assigned at random and answer order is shuffled. Use the dashboard to compare responses to the same question under different context, then ask whether that context should have influenced the answer.
Human + AI Work Allocation
Activity: Have students decide which parts of a marketing project to do themselves, give to AI, or complete together with AI.
AI relevance: AI can make work easier while making different people’s outputs more alike, so individual productivity and the variety of ideas are different things to evaluate.
Before the reveal: Students assign two tasks to each work mode before seeing the results; they use their own descriptions to prompt an external AI service and either keep its first substantive answer or actively revise it in the collaborative condition.
Reveal: the principle and how the activity works
Principle: Making individual work easier can also make a group’s outputs more alike; productivity and the variety of ideas are different outcomes.
What to compare: Compare human-only, AI-only, and collaborative work on effort or completion time and the variety of outputs: did AI make a task easier while producing more similar names, images, or ideas across students, and did human revision change that?
How it works:
Students lock in two human-only, two AI-only, and two collaborative tasks. Human-only work uses no AI. In AI-only work, students write their own product description and submit the first substantive AI response without refining it. Collaborative work allows students to question, redirect, and revise the response.
The dashboard groups allocations, completion times, and completed or partial outputs. Compare the effort needed with how many distinct approaches appear across students. AI may make producing a good-looking individual result easier while narrowing the range of ideas across the class. Show different student descriptions beside similar AI outputs where possible: identical starting descriptions are not required for that pattern to appear.
Ask whether collaboration added different ideas, better judgment, or mainly a more polished version of the same idea. Students choose both their task allocations and their AI services, so the classroom comparisons illustrate questions about work design rather than proving that one mode always causes a particular result.
Finding a Helpful Tone
Activity: Have students choose which AI responses to send as a company representative and which responses they prefer to receive as a user.
AI relevance: Similar preferences can shape both response design and user feedback, reinforcing the same patterns without necessarily improving accuracy.
Before the reveal: Keep the company and user roles distinct across the two rounds, then compare selections, offered options, and preferred responses on the dashboard.
Reveal: the principle and how the activity works
Principle: The same attractions and constraints can operate when people design an AI response and when they reward it.
What to compare: Compare the company role’s choices with the user role’s preferences: do the same styles attract selection at both stages, after accounting for which options users were offered?
How it works:
Round 1 puts students in the role of an AI company representative choosing what to send. Round 2 puts them in the user’s role choosing what they prefer to receive. The first round also shapes which options are offered in the second. The activity demonstrates how similar preferences and expectations can influence both response design and the feedback that reinforces responses.
Compare the response styles selected in both roles. Do confidence, agreement, or making someone feel understood make an answer attractive at both stages? Ask what happens when those features compete with accuracy or a useful correction. A tendency can be supported twice: first by the choice of what to offer, then by the reward for what users like.
Use the dashboard to compare first-round selections, second-round exposures, and second-round wins. Exposure matters because users cannot choose an answer they were never offered. The exercise models the choices behind design and reinforcement; it does not train an AI model or guarantee that the class will show the same pattern in both rounds.
Safeguard Game
Activity: Have students present truthful facts about a request and decide whether another person’s request should be allowed.
AI relevance: AI safeguards must judge human behavior from partial information, which can be difficult even when nobody can lie.
Before the reveal: Requesters choose and order game-supplied facts without changing them; responders see one fact first and can allow, refuse, or pay a points cost to ask for more.
Reveal: the principle and how the activity works
Principle: Truthful information can still be insufficient to identify whether human behavior is appropriate.
What to compare: Compare decisions about the same request with fewer versus more truthful facts and with different facts shown first: which missing fact changed whether the action looked appropriate?
How it works:
The requester chooses four game-supplied facts and their order; they cannot invent, rewrite, or falsify evidence. The responder initially sees one fact and may ask for more. A request has a stated policy and a hidden reason that determines whether it is appropriate. This lets the class experience the difficulty of judging behavior from partial information even with lying removed.
Use the dashboard to compare decisions, revealed facts, and hidden reasons. Which truthful facts were compatible with both an appropriate and an inappropriate action? Which additional fact settled the question? Choosing which facts to show can affect a judgment without any fact being false.
Asking reveals more evidence but reduces available points. Discuss the cost of deciding too soon versus the cost of asking. The AI connection is that a safeguard reasons from evidence about a person’s situation and purpose, not direct access to the whole situation. More reasoning cannot guarantee recovery of a missing fact; obtaining that fact can change the decision.
Delivery Robot
Activity: Have students choose instructions for a delivery robot and see whether it delivers the food as they intended.
AI relevance: AI can satisfy the words of an instruction while finding an unwanted way to achieve the result.
Before the reveal: Students choose one word in each of four slots, observe an outcome, and revise their instructions before seeing the successful combinations.
Reveal: the principle and how the activity works
Principle: Specifying a desired result does not necessarily specify an acceptable way to achieve it.
What to compare: Compare outcomes after changing one instruction word, and compare what the words literally require with what the person intended: which unwanted routes remain possible?
How it works:
There are exactly four successful combinations in the current app: safely / send / hot / waiting person; carefully / serve / fresh / table; quickly / deliver / intact / customer; efficiently / route / prepared / person who ordered.
There are 256 possible combinations. Each successful combination differs from every other in all four choices; changing just one word in a winner produces a failure. The app uses authored outcomes, not a live AI model.
The reveal is about what instructions fail to guarantee: each selected word is meant to remain true while the robot finds an unwanted way to satisfy it. Connect searching these possibilities to prompting and to searching for ways around safeguards.
The Market
Activity: Have students produce and sell goods while their classmates’ choices change the market.
AI relevance: When many people pursue the same AI opportunity, their combined actions can change the opportunity itself.
Before the reveal: Students see three unfamiliar goods, forecasts, and production/pricing choices; keep the historical name reveal until after they have experienced the market.
Reveal: the principle and how the activity works
Principle: Social effects emerge from people responding to one another, rather than simply adding up isolated individual benefits.
What to compare: Compare an attractive opportunity early in the market with what happens after classmates invest in it too: how do other people’s responses change supply, competition, and returns despite correct buyer forecasts?
How it works:
Pilut is TULIP reversed; Aduog is GOUDA reversed; Ragel is LAGER reversed. Save this reveal for the debrief.
The app does not schedule a crash. Players’ production and pricing decisions can create one. Production takes one year; buyers choose the lowest available price they will pay. Pilut’s rising prices make it increasingly attractive, but everyone acting on that opportunity changes supply and competition.
Ask: "The forecasts about buyers were right. What did they leave out?" The answer is how other producers would respond. Relate this to how widespread AI adoption changes the setting in which people use it.
The Game of AI-Displaced Life
Activity: Have students explore how AI changes the tasks, skills, and choices along a career.
AI relevance: AI taking over some tasks can reshape a job and its career path without automatically eliminating the whole job.
Before the reveal: Students choose a starting path and respond to changing work and retraining opportunities; compare what happened to tasks and careers before discussing job displacement.
Reveal: the principle and how the activity works
Principle: Task automation and job displacement are different changes, with outcomes shaped by adaptation and opportunities to retrain.
What to compare: Compare how individual tasks change with how the whole job changes, and compare career paths with different starting skills and retraining choices: what lets someone adapt rather than lose their role?
How it works:
The activity demonstrates how changes to individual tasks can reshape a career without automatically eliminating the whole job. The simulation changes tasks and the value of skills over time. Compare paths and opportunities to retrain; do not read a simulated outcome as a forecast for an actual person or occupation.
Grow the Firm
Activity: Have students run a company as AI changes its work and experienced employees retire.
AI relevance: AI can improve current output while reducing the work through which future experts learn.
Before the reveal: Students make staffing and AI-use choices over successive years, then compare profits with skill development and the ability to replace experienced staff.
Reveal: the principle and how the activity works
Principle: Optimizing today’s measured performance can undermine the development of expertise needed for tomorrow’s work.
What to compare: Compare staffing and AI-use choices on current profit, junior workers’ skill growth, and the ability to replace retirees: can a choice improve today’s results while weakening the future workforce?
How it works:
The activity demonstrates how gaining efficiency with AI today can reduce the opportunities that develop tomorrow’s experts. Human effort on work builds expertise; reducing that effort with AI also reduces some learning opportunities. Compare current profit with who is gaining experience and who can replace retirees. The tradeoff is built into the simulation and is not an empirical estimate.
Advising the Leaders and Best
Activity: Help students choose their college courses and see how those choices affect their life trajectory.
AI relevance: As AI changes work, learning across fields can help people recognize new opportunities and develop the skills to pursue them.
Before the reveal: Students advise several people by choosing a major and electives, then explore careers and alternative futures; compare curricula within the same major before discussing what education should develop.
Reveal: the principle and how the activity works
Principle: Education can prepare people to notice connections and adapt to changing work, while visible performance and earnings capture only part of what learning is for.
What to compare: Compare concentrated and broader curricula within the same major across alternative futures: how do starting skills affect noticing opportunities, adapting to AI changes, and developing expertise through later work?
How it works:
The activity demonstrates how initial learning can help people notice new opportunities and adapt as AI changes their work. Education supplies starting skills; practice builds expertise; AI changes the mix of work. Compare different curricula within a major and across simulated futures. The model illustrates the possible value of noticing connections across fields; it does not prove that a particular curriculum causes better real careers.