
AI or Intuition?
Make sensitive decisions by balancing evidence, algorithms, and human judgement.
In short. AI or Intuition? is a serious game (management) for 6–20 participants, lasting 60–90 min. Created by serious-games.com. Use it for: prise de décision, communication, résolution de problèmes.
Card by serious-games.com, updated 22 November 2026. Original game by serious-games.com.
Participants play decision-making committees faced with dilemmas where an AI offers an imperfect recommendation. In each round, they must choose their level of trust in the AI, argue their position, and then face the ethical, operational, and human consequences. The game highlights biases, blind spots, and the need for explicit human oversight. It is not about demonising or idealising AI, but about developing responsible decision-making reflexes.
Walkthrough
- 1
Framework and Game Contract
10 minThe facilitator sets up teams of 3 to 5 people and distributes 12 Trust tokens, 3 Right to Question cards, and a Score sheet to each team. They say: "Today, you will decide with a useful but never infallible AI; your challenge is to justify the right level of trust." Participants name their team and choose a rotating role: reader, ethical guardian, decision-maker. The aim is to create a safe, concrete, and learning-oriented environment.
Tip : Clarify that the scenarios are fictional and deliberately simplified: we assess the quality of reasoning, not legal, medical, or technical expertise.
- 2
Calibration: the 4 Levels of Trust
10 minThe facilitator presents the 4 possible decisions: 0 tokens = reject the AI's advice, 1 token = take inspiration from it, 2 tokens = follow under human control, 3 tokens = follow automatically. They say: "The more you delegate, the faster you can win, but the more you expose the team if the AI hides a bias or error." Teams quickly place a personal or professional example under one of the 4 levels. This step provides a common language to avoid vague debates.
Tip : Display the 4 levels on the wall or in a shared board and keep them visible throughout the game.
- 3
Round 1: Decision Under Limited Information
15 minThe facilitator gives each team a Scenario card, an AI Recommendation card, and two minutes for silent reading. They say: "You have 7 minutes to decide how many Trust tokens to wager and write down two reasons: one reason for effectiveness, one ethical reason." Teams discuss and can spend a Right to Question card to obtain an additional Hint card. The purpose is to simulate a real decision, constrained by time and incomplete information.
Tip : Examples of scenarios: candidate pre-selection, prioritisation of client files, detection of internal fraud, allocation of limited aid.
- 4
Revealing Blind Spots and Scoring
10 minThe facilitator reveals a Blind Spot card for each scenario: incomplete data, indirectly discriminatory variable, poorly formulated objective, lack of recourse, or unrecognised human context. They say: "Compare your level of trust with the revealed blind spot; a good decision is not only quick, it is defensible." Teams score 2 points if their justification anticipated the risk, 1 point if their level of trust was cautious, and lose 1 point if they automated without safeguards. The score makes the value of vigilance visible.
Tip : Do not try to trap the teams: highlight the good hypotheses made, even if the final decision was imperfect.
- 5
Round 2: Pressure and Stakeholders
15 minThe facilitator distributes a new scenario and adds a Pressure card: urgency, financial objective, reputation, team overload, or hierarchical injunction. They say: "This time, a stakeholder is pushing you; note who stands to gain, who risks losing, and what recourse you foresee." Teams have 8 minutes, then wager their tokens and write a human oversight rule in one sentence. The aim is to introduce social complexity: ethics also plays out in constraints and power.
Tip : If a team gets stuck, ask: "What decision would you be willing to explain publicly to the person concerned?"
- 6
Challenging Audience
15 minEach team presents their decision, level of trust, and main safeguard in 2 minutes. The facilitator assigns another team the role of challenger and says: "Your mission is not to attack, but to test the robustness of the decision." Challengers ask one question about the data, one about the affected people, and one about possible recourse. This phase develops the quality of argumentation and shows that trust in AI must be debatable.
Tip : Use a strict timer: 2 minutes for presentation, 2 minutes for questions, 1 minute for responses per team.
- 7
Creating a Golden Rule
10 minThe facilitator asks each team to formulate a golden rule applicable in their context: "We can delegate to the AI only if..." They say: "Your rule must contain a data quality criterion, a threshold for human oversight, and a mechanism for contestation." Participants write their rule on a poster, then vote with 3 stickers for the most useful and realistic rules. This step transforms the game into reusable operational principles.
Tip : Reject overly general rules like 'be aware of biases'; always ask: 'How will you see this concretely?'
- 8
Debrief and Transfer
10 minThe facilitator has the group sit in a circle and reviews the scores without designating an absolute winner. They say: "The score indicates your reflexes, not your worth; what matters is what you change in your real decisions." Participants share a learning point, a risk to monitor, and an action applicable from next week. The purpose is to consolidate awareness and prevent AI ethics from remaining an abstract discussion.
Tip : Conclude with an individually completed sentence: "The next time an AI recommends a sensitive decision, I..."
Variants
- Short version 60 minutes: play a single complete round, reduce the challenging audience to one question per team, and keep the final golden rule.
- Advanced version: add an external auditor role who can award 1 bonus point if the team clearly explains the data used, the limitations, and the planned recourse.
- Large group version: form up to 5 teams and give the same scenario to two teams to compare different levels of trust.
- Remote version: use a collaborative board with columns for Scenario, Bet, Reasons, Blind Spots, and Golden Rule; tokens become digital dots or numbers recorded on the team sheet.
Debrief guide
- When did you tend to trust the AI too much, and why?
- What clues helped you distinguish a useful recommendation from a dangerously incomplete one?
- How did the pressure of time, performance, or authority change your ethical reasoning?
- Which individuals or groups risked being invisible in your decisions?
- What human safeguards seem essential before delegating a decision to an AI?
- How can a decision assisted by AI be made contestable, explainable, and improvable?
- What concrete rule could you test in your team or organisation?
Facilitator reviews
No review yet. Have you run this game? Your feedback will help the next facilitators.
Similar games
Same objectives or same category, to vary or follow up.

The Beer Game
A supply chain simulation to experience the bullwhip effect and understand the impact of delays.

Data Detective
Investigate a 30% drop and make decisions without being fooled by the data.

Common Ground
Mediate a fictional conflict, lower the tension and turn positions into needs.

The Socratic AI Challenge
Question better, use AI judiciously, decide more clearly.

Ugli Oranges
A quick negotiation to reveal the difference between defending a position and exploring true interests.

Walt Disney Method (Dreamer, Realist, Critic)
Maturing an idea by separating dreams, action plans, and constructive criticism.