Procedural content generation (PCG) is the creation of game content - levels, maps, terrain, dungeons, items, even quests - by algorithm rather than by hand. A small program plus a random seed expands into a large amount of content, often deterministically, so the same seed always produces the same result. The technique is old: the 1980 game Rogue generated a fresh dungeon each run, giving the “roguelike” genre its name, and fractal and noise functions have long been used to synthesize natural-looking terrain.
PCG spans a range of methods. Constructive and grammar-based approaches assemble content from building blocks under rules. Search-based and evolutionary methods, often using genetic algorithms, evolve content toward a fitness target. Noise functions like Perlin noise produce coherent randomness for landscapes. More recently, machine-learning approaches train models on existing human-made content and generate more in the same style - the subject of the 2017 survey “Procedural Content Generation via Machine Learning,” which defines that subfield as “the generation of game content using machine learning models trained on existing content.”
The appeal of PCG is leverage and variety: a handful of developers can ship a world far larger than they could build by hand, and players get content that differs every session. No Man’s Sky famously generates over 18 quintillion planets from a single seed. The cost is control - procedurally generated content can feel samey or produce broken or unfair results, which is why much PCG research focuses on constraints, repair, and guaranteeing playability.
Why business readers should care: PCG is generative AI’s older cousin in entertainment. It shows both the promise (massive content from small teams) and the pitfall (uneven, hard-to-control output) that now define generative systems across every industry.