Unique every time,
Procedural content gen,
Enhances design

Procedural content generation is a method of creating content - usually with limited user input - via an algorithm. [3] This can cover a broad variety of techniques and methods of content generation, modification and enhancement, from random map-generation to procedurally generated enemies or sounds.

There are usually two axes which PCG roles can be placed: players and designers. In the figure below we can identify the key roles of PCG in games across the dimensions of designer initiative and player experience.

The roles shown in the picture act either autonomously or as a collaborator in the design process. Those are autonomous, mixed-initiative, experience-driven or experience-agnostic. All of the four will be briefly discussed later on this page.

In this form of PCG, there is almost no human input to the generation process, apart from the parameters used in the algorithm. Autonomous and mixed-initiative PCG systems often overlap in design and functionality, with map creation being a prime example. PCG can be used fully autonomously for map generation, such as in Minecraft or Dwarf Fortress, where new maps are created whenever the player starts a new game. Here the developers have only specified parameters and algorithms and are not present when the map is actually created. In contrast - as discussed below - procedural dungeons and maps can be used as tools to help give designers a base to work off of or to fill in areas of the game around a framework of important scenes.

In mixed-initiative PCG, algorithms are used in concord with human designers to create content for games. In this scenario, PCG is usually considered a tool to aid the human designer, saving them time by automating tasks. Randomly-generated dungeons and maps are one of the most common examples of this.

Experience-driven PCG systems are designed to enhance or further the player experience in some way, often in an aesthetic sense. One of the most well-known examples of this is the AI-Director of Valve's Left 4 Dead 2, which alters the difficulty and pacing of the game based on player performance, changing the types of enemies encountered if the player is doing better or worse. Metal Gear Solid V also displays a similar, if simpler version of this, where enemies adapt over time to the player's tactics. For example, if the player performs most of their missions at night, the enemies will start to equip torches in order to counter the player's preferred play-style. This encourages the player to experiment with different approaches and also increases difficulty as the player becomes more adept at the game.

This is simply PCG that is not focused on directly enhancing player experience. This covers many forms of PCG, especially tools for developers, such as a foliage brush in a game engine, that randomly places vegetation within a radius.

[1] G. N. Yannakakis and J. Togelius, "Artificial Intelligence and Games". Springer, 2018.
[2] Liapis, Antonios, Gillian Smith, and Noor Shaker. "Mixed-initiative content creation." Procedural content generation in games. Springer, Cham, 2016. 195-214.
[3] Togelius, J., Kastbjerg, E., Schedl, D., Yannakakis, G.N., "What is procedural content generation?: Mario on the borderline." In Proceedings of the 2nd Workshop on Procedural Content Generation in Games (2011)
[4] V. Vashistha and S. Malik, "Procedural Content Generation in Games towards Semantic Web", International Journal of Computer Sciences and Engineering, vol. 6, no. 9, pp. 803-812, 2018.