1 free read left. Create a free account for unlimited reading + AI research.
Create free accountWhitepaperNew York University
Large language models serve as effective co-creative muses in game development, though they currently lack the capacity to replace human designers. By evaluating three distinct game genres—space shooters, platformers, and roguelikes—researchers compared baseline games against those developed with human-only input and those assisted by ChatGPT. The findings indicate that while AI-assisted iterations consistently outperformed baseline versions lacking additional features, they failed to match the quality of human-designed games. Player evaluations consistently favored human-led development, citing superior game feel, cohesive logic, and nuanced implementation as the primary differentiators.
The technical limitations of current AI models remain a significant barrier to autonomous game creation. ChatGPT frequently struggled with complex C# code generation within the Unity engine and often failed to grasp context-specific mechanics. These technical shortcomings highlight a fundamental gap between the model’s ability to generate structurally sound code and its inability to replicate the creative intent and iterative refinement necessary for high-quality player experiences. Consequently, the model functions best as a source of conceptual inspiration rather than a reliable engine for direct implementation.
Ultimately, the most effective application of generative AI in this field is as an ink splotch—a tool for providing diverse, non-obvious prompts that help designers overcome cognitive biases and expand their creative horizons. By acting as a collaborative assistant that offers a broader perspective during the early stages of development, the technology aids in the brainstorming process. However, the final synthesis of these ideas requires human curation to ensure that the resulting gameplay remains cohesive and engaging. The study confirms that while AI can significantly augment the creative process, the nuanced understanding of player experience remains a distinctly human domain.
Each row is a game or company this document names, each column a document, oldest first. A mark means that document names it. Select a row to see those documents.
University of California Santa Cruz · 2024