Is D&D a Generative Adversarial Network?

Dungeons and Dragons (D&D) is a popular game that has influenced the world of video games and fantasy beyond measure. Its impact on popular culture and video game development are impossible to ignore, being the root influence on many of today’s most successful gaming franchises like The Elder Scrolls and Baldur’s Gate. 

However, when we look deeper into the mechanics of the game, we start to see a more interesting angle of the game that reflects artificial intelligence (AI) and, more specifically, generative adversarial networks (GANs). While D&D is much older than the development of GANs (and there is no claim that D&D influenced their creation), it is worth taking the time to analyze the similarities between the two systems to better understand how AI mimics traditional generation models. 

Basics of Dungeons and Dragons

Dungeons and Dragons is a tabletop game that utilizes pen and paper to construct characters and quests for players to embark on using a set of rule books and their own creativity. There are two major roles in a game of D&D - the players and the Dungeon Master (DM)

Each player is tasked with creating a character from a select set of races and classes, establishing a unique backstory and dynamic for storytelling and role-playing. The DM on the other hand is tasked with world-building, developing an environment for the characters to explore and venture through, adding complexity and intrigue to the game with recurring non-player characters (NPCs) and plot lines. 

The game is popular for its dynamic storytelling and open-ended scenarios which players and the DM resolve through the use of various dice rolls. Over time, through multiple sessions, characters begin to develop certain personalities as the world around them becomes more lifelike and unpredictable. 

Understanding Generative Adversarial Networks

Generative adversarial networks are a form of AI that poses two AI models against each other to produce generative content. They are a generator and a discriminator, using each other to improve their own functions and outputs. 

The Generator begins the process by taking random noises as input and creating data in the form of images, videos, audio, etc. This content is then passed to the discriminator which determines whether the content was artificially produced or not. If the discriminator identifies the content as artificial, then the generator must improve its algorithms to create higher-quality content. If the discriminator fails to identify the content as artificial, then it must improve itself to identify artificial content more accurately. This process then repeats until both models have reached an optimal rate of success. 

GANs can be used for a wide variety of digital applications, especially image and video generation, creating content like faces and people that have never existed. They can also be used to transfer styles from an artist and recreate similar images with different subjects. GANs are also especially adept at creating procedurally generated content in video games, developing radiant quests and rewards for players who want to continue dungeon crawling in their favorite games. 

Parallels between D&D and GANs

There are many comparisons that can be made between D&D and GANs. To begin, DMs often exhibit many similar traits to a generator model. Their role is to create an environment for players to interact with, adapting it to their responses and adding creativity so that content doesn’t get redundant while players follow a more reactionary course of action, trying to overcome the DM’s challenges and finding new solutions to increasingly difficult challenges. 

In both cases, the DM/generator and players/discriminators must adapt themselves to overcome the other. While the DM is not considered the enemy by the players, they must still pose challenges and scenarios that present a degree of difficulty that scales to the player’s level and abilities. 

Then, like an AI model, both parties must self-evaluate themselves and learn how to overcome the other. As players become more successful at dungeon crawling they begin to encounter more complex puzzles and stronger enemies. However, they are rewarded for overcoming their foes with improved weapons and armor that make their strategies more reliable. This adversarial, yet mutually beneficial relationship between players and DM highlights the core concept of GANs. 

Learning and Evolution in D&D and GANs

Both D&D and GANs exhibit similar mechanisms where each contributor to the system must learn to adapt and evolve to the actions of the others. This dynamic development creates a system of self-learning and improvement that is a core component of AI systems and could help us understand better ways of applying game theory and strategy building to machine learning algorithms. 

While the two models - D&D and GANs - are not totally similar, with one system [D&D] using subjective experiences and dice rolls to generate content and the other [GANs] more focused on algorithmic precision, both models showcase how new content can be produced almost infinitely when multiple contributors are placed in a mutual environment and given a set of rules that can be followed or bent to the party’s desire.