Roguelike Radio Episode 149: Chaos Theory
Dec 2, 2018
Alexei Peppers, a professional game designer and long-time NetHack player, explores chaos theory and procedural generation. He explains Lorenz’s butterfly effect, visualizing generators with expressive range, and how small changes shape outcomes. Conversations cover metrics for tuning generators, emergence from nonlinear rules, and how players learn and feel ownership of procedurally created worlds.
AI Snips
Chapters
Books
Transcript
Episode notes
Run Wide Sweeps Instead Of Eyeballing Samples
- Run many permutations and visualize metrics instead of eyeballing a few samples to avoid false confidence about generator behavior.
- Use clusters on plots to identify parameter sets that produce desired difficulty or linearity.
Use Humans For Pattern Picking
- Let humans inspect visual plots because our brains detect patterns better than many automated optimizers.
- Use a hybrid: computer to generate many samples, humans to spot meaningful clusters and iterate.
Metric Choice Determines What You See
- Picking the right metrics is crucial: bad metrics produce misleading visualizations and bad design choices.
- Alexei recommends using human visual pattern recognition to spot interesting clusters then work backwards to parameters.
