An innovative AI-generated version of the classic video game Doom has been developed, allowing players to engage with the game despite the absence of traditional computer code or graphics. Researchers assert that this groundbreaking AI model could pave the way for the creation of entirely new games, leveraging similar technology to what is currently used for text and image generation.
This model, named GameNGen, has been crafted through extensive research that enables it to deliver a playable version of Doom for up to 20 seconds, maintaining key elements from the original game, including scores, ammunition levels, and map designs. Players can seamlessly interact with in-game features, such as battling enemies and opening doors, within this brief timeframe.
After approximately 20 seconds, however, the memory limitations of the model result in a breakdown of the gaming experience.
Originally launched in 1993, Doom has been frequently utilized in various computer science projects, often challenging developers to run the game on unconventional devices, from toasters to treadmills. Unlike previous attempts that utilized the original game code, GameNGen represents a notable departure from traditional programming methods; it employs a neural network to learn and recreate the game solely through observation.
The development involved an AI model that learned to engage with Doom as a human player would. This model repeatedly played the game while a second AI, inspired by Stable Diffusion, analyzed how a multitude of inputs altered the game’s state.
In this way, the second model effectively became a replica of Doom, capturing the essence, rules, and mechanics of the original without direct access to its code. In preliminary tests, human players struggled to differentiate between real gameplay and AI-generated simulations, performing only slightly better than random guessing.
The creators of GameNGen suggest that their findings could represent a proof-of-concept, indicating a future where neural networks generate games from simple text descriptions or conceptual artwork, potentially reducing production costs associated with human programming efforts.
Experts highlight the significance of this advancement while also noting the ongoing need for human creativity in game design. One commentator emphasizes that while AI can simulate environments, the imagination and narrative skills of human designers remain crucial for creating engaging gaming experiences.
