How can AI learning games support mixed-confidence participants?

A group described as AI beginners rarely starts from one place. One person may use a general assistant every day but have weak checking habits. Another may understand the risks but hesitate to touch the tool. A third may bring valuable domain judgement and little interface experience.

A game does not erase those differences. Its rules can either distribute meaningful practice or allow the fastest and most confident participant to take over.

Mixed confidence is a design condition, not a learner defect

Confidence changes with the task, tool, group and perceived consequence. Someone comfortable drafting private notes may become cautious when asked to handle customer information. A confident speaker may still accept weak AI outputs. Self-assurance is therefore not a reliable substitute for competence.

Prior experience also has several forms. Tool fluency, domain expertise, critical evaluation and knowledge of organisational controls are distinct. A game designed around prompt speed may reward only one of them while appearing to assess practical AI capability.

Treat learner variability as predictable. Ask about prior experience and access needs, but avoid permanent beginner and expert labels. People need opportunities to strengthen different parts of the capability, and their support needs may change during the activity.

Games can redistribute participation or concentrate it

Games can create reasons for everyone to contribute. Different roles, limited information and joint decisions can make a range of expertise useful. They can also amplify existing inequalities.

A public leaderboard may make uncertainty costly. A timed interface can disadvantage people unfamiliar with the tool or those using accessibility support. In team play, the person who grabs the keyboard may make every operational decision while colleagues become spectators. The room looks active, but practice is concentrated in one participant.

Competition is not required. Teams can work against a scenario, a shared time limit or a changing constraint. Where competition is used, score the target behaviour and avoid exposing individual competence. A checking decision should matter more than the number of prompts generated.

Keep one objective while varying routes and support

Inclusive design does not mean removing meaningful difficulty. It means reducing barriers unrelated to the learning objective and providing appropriate challenge and support.

Brief the tool, rules and scenario before time pressure begins. Offer examples and non-examples. Provide optional prompt cards, a glossary or an extra planning minute without giving away the decision. Let participants respond aloud, in writing or through a role representative where the objective does not depend on one communication mode.

Rotate roles such as problem framer, AI operator, evidence checker and decision owner. Each role should include an observable decision rather than an administrative task. The experienced tool user then has to practise checking or explaining, while a cautious participant can operate with support.

Use formative feedback to adjust difficulty. If a team is overwhelmed, reduce incidental complexity or provide a cue. If it succeeds by following a memorised route, vary the evidence or introduce a trade-off. Support should help participants engage with the same core capability, not replace it.

Look for meaningful participation, not visible enthusiasm

Enjoyment and confidence ratings can help describe the experience, but they do not prove learning. Look at what each participant had to notice, decide, explain and revise.

Private reflection can reveal learning that is not visible in rapid group discussion. Ask participants to record one choice they would change and the evidence behind it. A second unfamiliar round can show whether they can apply a principle rather than repeat a winning sequence.

Invite feedback on barriers as well as content. Did the interface, pace, language or group structure prevent a relevant decision? Adjust the design rather than attributing every difficulty to motivation.

No single design will suit everyone. Accessibility requirements should be sought directly and addressed through local processes. The objective is not identical activity; it is a fair opportunity to practise the capability and contribute evidence.

Example

An insurance operations group includes frequent AI users, cautious first-time users and colleagues who prefer not to operate a shared screen. In a facilitated challenge, teams rotate among problem framer, AI operator, evidence checker and decision owner.

Optional prompt cards and extra planning time are available to any team. The score rewards evidence use and justified escalation rather than speed. After the first round, each participant records one decision and then takes a different role in a changed scenario.

Every person makes a relevant AI-related decision. The activity preserves the shared objective while varying routes and support.

FAQs

• Should beginners and experienced AI users play separately?

  Mixed groups can support peer learning when roles and airtime are deliberately distributed. Separate or temporarily differentiated groups may be fairer when prior tool experience would otherwise dominate the task. Choose from the objective and observe what actually happens.

• Does removing competition make an activity less game-like?

  No. Games can use goals, rules, constraints, feedback and replay without ranking individuals. Cooperative and team-against-scenario structures may better support learning where public comparison would suppress uncertainty.

• How can a facilitator tell whether quieter participants are learning?

  Give every participant an observable decision, rotate roles and use private written reflection or varied response methods. Do not equate speaking frequency, visible enthusiasm or self-confidence with capability.