# Microsoft’s WHAM: Progressing AI-Generated Gameplay, Yet Still Developing
## Introduction
Microsoft Research has introduced its newest AI-focused gaming model, the **World and Human Action Model (WHAM)**, representing a major advancement in AI-enhanced game creation. WHAM’s purpose is to produce interactive gameplay sequences by interpreting video content and player actions. Although the model has achieved remarkable progress, it is chiefly valuable for low-resolution prototyping rather than fully functional AI-generated games.
## The Progression of AI-Generated Gameplay
For years, AI researchers have been investigating **”world models”**, enabling AI to deduce how in-game entities and characters ought to act based on video content. The objective is to develop entirely interactive, AI-generated media that mimics new playable environments.
WHAM builds upon earlier AI systems like **Google’s Genie**, which sought to craft interactive 2D environments from singular images. Microsoft’s strategy, conversely, emphasizes training AI with real gameplay footage to enhance reliability and continuity in AI-generated sequences.
## How WHAM Operates
WHAM’s training involved gameplay footage from *Bleeding Edge*, a 4v4 online brawler created by Ninja Theory, a Microsoft subsidiary. By scrutinizing **seven years of player gameplay videos**, WHAM gained insight into how in-game objects and characters engage with one another.
At first, the AI faced challenges, producing inconsistent and warped clips. However, after **1 million training updates**, WHAM exhibited a foundational grasp of gameplay principles, such as item explosions and character dynamics.
### Evaluating WHAM’s Capabilities
Microsoft assessed WHAM’s performance by supplying it with **one second of actual gameplay footage** and requesting it to produce subsequent frames based on simulated actions. The AI-generated videos were then evaluated against real gameplay using the **Frechet Video Distance metric**, a method for measuring video realism.
Key outcomes include:
– WHAM-generated videos remained **consistent for up to two minutes**, outshining Google’s Genie 2 model, which held consistency for merely one minute.
– The AI reacted fittingly to **randomized inputs**, though it still did not achieve human-level precision.
– WHAM showcased **object persistence**, meaning newly added objects stayed in the scene with suitable interactions.
## WHAM’s Shortcomings
In spite of its progress, WHAM is far from generating entirely playable AI-produced games. Some of its primary shortcomings include:
– **Low Resolution**: WHAM currently functions at **300×180 resolution** (akin to a Nintendo DS screen) at **10 frames per second**—significantly below current gaming benchmarks.
– **Visual Artifacts**: AI-generated characters frequently appear **morphing and stretching**, lacking the solid structure found in traditional game models.
– **Slow Processing**: WHAM’s existing prototype produces video **at a considerably slower speed** than necessary for real-time gameplay.
## The Future of AI-Generated Gaming
Although WHAM is not yet equipped to create fully-fledged AI-powered games, Microsoft views it as a vital step toward **real-time AI-driven game creation**. Developers can explore WHAM through a **prototype tool on Azure AI Foundry**, allowing them to craft interactive gameplay sequences from a handful of video frames.
Microsoft imagines a future where AI can generate **high-quality, immersive experiences on demand**. Nevertheless, as indicated by WHAM’s current limitations, there remains considerable work to be done before AI-generated games become a commonplace reality.
## Conclusion
Microsoft’s WHAM symbolizes a significant advancement in AI-triggered game development, providing enhanced reliability and object persistence compared to earlier models. However, its low resolution, sluggish processing, and visual discrepancies emphasize the hurdles that still exist. While WHAM is not yet suited for real-time gaming, it offers a preview of the potential future of AI-generated interactive experiences.
As AI research evolves, models like WHAM could eventually transform game development, empowering creators to construct dynamic, AI-generated worlds with minimal effort. For the moment, WHAM is an intriguing yet **early-stage prototype** in the quest for AI-empowered gaming.
