How can scenario-based games help people practise framing AI problems?

AI demonstrations often begin with a clean task: summarise this document, generate these options or rewrite this message. Real work frequently begins earlier, with an unclear request, competing goals and incomplete evidence.

If learning starts at the prompt box, people may practise giving instructions without learning what problem deserves attention. A scenario-based game can place problem definition inside the activity so that framing choices become visible.

Put the framing decision before the prompt

Problem framing determines what success means, whose interests matter, which constraints apply and what role AI should have. A fluent prompt cannot repair a task aimed at the wrong goal.

Generative AI also creates metacognitive demands. People need to plan an approach, monitor whether the output serves the purpose and adapt when it does not. A learning activity can support that work by requiring decisions before access to the tool.

Ask participants to state the problem, identify the decision owner, name essential evidence and define what AI may contribute. They may decide to use AI for option generation, comparison or drafting—or decide that the task should remain outside AI. These choices create a baseline against which later outputs can be judged.

Use purposeful ambiguity, not hidden rules

Workplace problems are often ill-structured: important information is missing, several frames are plausible and the path is not given. A scenario can represent that condition without becoming unfair.

Make missing information discoverable. Teams might receive a limited number of information tokens to spend on customer evidence, process data, policies or stakeholder perspectives. Their choices reveal what they believe matters. If an essential rule is impossible to find, failure teaches only that the designer withheld it.

Give participants a clear activity goal even when the work problem is ambiguous. Explain what resources are available and how consequences operate. Productive ambiguity lies in interpreting the situation, not guessing the facilitator's private answer.

Subject-matter review is important. Fictional consequences should reflect credible relationships in the work. A simplified scenario can focus attention, but it should not teach a false operational rule.

Make goals, evidence and AI boundaries observable

Game mechanics can turn invisible framing into choices. An information budget exposes priorities. Different role cards reveal stakeholder interests. A time constraint creates a trade-off between further inquiry and action. A required “AI role” card makes appropriate non-use a legitimate option.

Score or feedback should follow from the target capability. Rewarding the longest prompt or greatest output volume shifts attention back to tool operation. Instead, show how an overlooked stakeholder changes the outcome, how weak evidence undermines a recommendation or how an unsuitable AI use creates avoidable risk.

Allow revision. After seeing an initial consequence, teams can spend another token, reframe the task or reduce AI's role. The second attempt tests whether they can use feedback, not merely recognise the correct answer after it is revealed.

Compare frames before comparing outputs

In the debrief, display each team's problem statement and evidence choices before showing its AI output. Ask what each frame made visible and what it excluded. Two teams may reach different but defensible frames because they prioritised different stakeholders or uncertainties.

Then examine whether the AI output served the chosen frame. Did it answer the actual question? What evidence would be needed before action? Did using AI narrow the team's view prematurely? Where would professional judgement or direct inquiry be more useful?

Connect the lesson to work by choosing an upcoming low-risk task and recording its frame before AI is used. Later, compare the original purpose with the output and decision. The scenario does not prove workplace capability, but it gives people a repeatable way to practise defining the work before accelerating it.

Example

A product team receives a fictional request to use AI to reduce customer-support demand. Before accessing an assistant, teams have five information tokens to spend on customer interviews, operational data, policy constraints, cost information or stakeholder conversations.

Each team submits a problem statement, an evidence plan and a decision about AI's role. One discovers that the original request hides an accessibility problem requiring direct user research. Another uses AI to compare process options but keeps the service decision human-led.

The debrief compares the frames before the outputs. Participants see that better AI use began with different questions, not a more elaborate prompt.

FAQs

• Is ambiguity always useful in a learning game?

  No. Ambiguity is useful when it represents genuine uncertainty and prompts inquiry. Unclear instructions, inaccessible essential information or arbitrary consequences create guessing rather than problem-framing practice.

• Should learners use AI to help frame the problem?

  AI can propose questions, stakeholders or alternative frames, but people remain responsible for purpose, evidence and boundaries. It can support framing without owning the decision about what work should be done.

• How is this different from teaching prompting?

  Prompting gives instructions within a selected task. Framing determines the task, goal, context, evidence needs and appropriate AI role. A well-written prompt can still serve a badly framed problem.