SkinDeepRESEARCHSteve Seguin

Use case · Idea to test

Game levels and difficulty

Suggest levels a player wants to play.

  1. Play a short level
  2. Rate the experience
  3. Suggest another layout
Illustration of the proposed workflow.

A game could learn whether a player enjoys exploration, tight timing or short puzzles, then choose or generate a suitable level. Difficulty, visual style and rewards can be considered separately.

Where the small model fits

A level generator proposes layouts. A small scorer predicts preference from level features, while a separate playability check rejects impossible levels.

What would need to work

Measure enjoyment and completion on new levels, not just the score assigned to a generated layout. The current maze policy chooses moves; it does not learn player taste or generate preferred levels.

Related experiments and evidence