AI or Not AI?
Decide in a few seconds whether a request is work for AI or for a human (or a script), and what a fork can customise.
Colleagues line up at the player's desk, each with a request and the urge to "put AI on it". The player decides before the clock runs out: AI or NOT AI (a human or a script handles it). An arcade game drawn as a ligne claire comic, played solo in 4 to 8 minutes.
Play it now, for free and without an account: the AI or Not AI? game page.
Rules
Four red flags rule AI out. When a request carries several, the highest-priority one explains it:
| Red flag | Why not AI |
|---|---|
| Binding decision | Hiring, legal, medical, money: a human stays accountable. |
| Named personal data | Real names, not anonymised: they never go to a model. |
| Published unreviewed | Sent straight to the public: one hallucination and it is out there. |
| Mechanical work | Sums, sorting, formats: a script does it for free, without mistakes. |
Everything else (language work, nothing sensitive): AI is the right call.
- On the keyboard: left arrow (or 1) = NOT AI, right arrow (or 2) = AI; Enter skips the explanation. With a mouse or a finger: the two stamps.
- A right stamp scores points, with a round bonus, a speed bonus (full in the first half of the clock) and a streak bonus. A wrong call or a timeout costs a life; "wrong" never takes points away. After each answer, the rule that decided is shown.
- Token budget: every AI call costs tokens, a right NOT AI refunds a few, each round refills the budget. An empty budget ends the game.
- Flying budgets: now and then a winged pouch crosses the scene. Typing its word (TOKEN, PROMPT…) or tapping it on a touch screen banks it.
- A game is a shift of 8 rounds of 6 requests: the clock tightens every round and tricky requests (the surface misleads) arrive from round 4. The game ends on the last life, an empty budget or the end of the shift.
- At the end, a profile sums up the way the player played (the Guardian, the Budget Burner, the Data Leaker…) with the most frequent mistakes.
The score shown is recomputed by the server from the answers: lives, budget, streak and the end of the game are replayed. The maximum is a perfect shift (every request right, fast and in a streak): a good player reaches part of it.
Skills measured
| Skill | Requests |
|---|---|
ai.usage.good-fit: knowing when AI fits |
requests without a red flag |
ai.usage.personal-data: protecting personal data |
named personal data |
ai.usage.accountability: keeping a human accountable |
binding decisions |
ai.usage.human-review: reviewing before publishing |
unreviewed publishing |
ai.usage.right-tool: choosing a script over a model |
mechanical work |
What a fork can customise
Rules (side panel of the editor):
| Rule | Default | Effect |
|---|---|---|
lives |
3 | Mistakes allowed. |
roundCount, requestsPerRound |
8 × 6 | Length of a shift. |
answerSeconds, speedUpSeconds, minAnswerSeconds |
7 s, 0.45 s, 2.4 s | Clock in round 1, time taken off each round, floor. |
warmUpSeconds, warmUpRounds |
3 s, 4 | Extra reading time at the start, fading out. |
pointsPerRequest, streakCap |
100, 8 | Points for a right call, streak bonus cap. |
trickyFromRound, trickySharePercent |
4, 60% | When tricky requests arrive and their share. |
budget |
on, 500 / 800 | Token budget: start, maximum, cost of a call, refund, refill, alert. |
budgetDrops |
on, 150 | Flying budgets: amount, requests before the first one. |
sessionDurationSeconds |
no timer | Maximum game length (the clock of each request stays). |
passingScorePercent |
50 | Pass mark sent to the LMS. |
Content, editable in place in the editor, in the fork's source language (other languages are translated with a file):
- each request: department, text, colleague, red flags (none = AI fits), tricky;
- the red flags and the "AI fits" case: name, rule, related mistake, emblem;
- the eight end profiles (name, description, emblem), the flying budget words and every interface label.
Look: brand colour (buttons, gauge), highlight colour, the colleagues (one to four pictures each), the three-plane office scenery and its changes over the rounds, the crest, the coin and the flying budget. The fork's brand name presents the game on the welcome screen ("Your organisation presents").
Review the content
The 96 default requests are listed with their red flag in games/ai-not-ai/content-review.md
(generated by bun run content:review in the game package), in English and French. About a
third are work for AI; tricky requests are spread across the five cases.