Microsoft Creates An AI-Rendered 'Quake II': An Achievement With Many Limitations

Microsoft creates Quake II using AI

​AI is the hype of technology, as it can revolutionize industries, enhance human capabilities, and transform the way people live, work, and interact with the world.

And this time, based on the technology that was originally popularized by OpenAI ChatGPT, Microsoft, through its Copilot AI platform, has successfully rendered an AI version of Quake II.

But unlike the original ’90s first-person shooter classic, the artifacts, locations and enemies are not hardcoded into the software. Instead, they're generated on the fly by a Microsoft AI model, but based on the content and style of the original Quake II.

According to Microsoft in a post on its website, the game is created by what it calls the 'WHAMM':

"Muse is our family of world models for video games at Microsoft. Following on from our announcement of Muse in February, and the World and Human Action Model (WHAM) that was recently published in Nature, we introduce a real-time playable extension of our model. Our approach: WHAMM which stands for World and Human Action MaskGIT Model allows generating visuals much faster than WHAM. This means that you can interact with the model through keyboard/controller actions and see the effects of your actions immediately, essentially allowing you to play inside the model."
Microsoft creates Quake II using AI
WHAM is Microsoft can tokenize gameplay data of image, action, image, action, and others' sequences into a longer sequence of tokens

Since the release of WHAM-1.6B, which was the first WHAM trained on Bleeding Edge, the researchers at Microsoft have changed and improved on a number of aspects that affect the overall experience.

For example, the team has improved the speed of generation, making WHAMM capable of generating images at 10+ frames a second, enabling real-time video generation. In contrast WHAM-1.6B can generate about 1 image a second.

Which this achievement, the researchers put WHAMM into a test, by making it create an AI version of Quake II.

However, when compared to Bleeding Edge, Quake II is a faster-paced first person shooter game which plays very differently. To be able to render the game, the team substantially reduced the quantity of data required for training WHAMM, which was achieved through more intentional data collection and curation.

By only focusing on a single level with intentional gameplay, the team managed to obtain enough high quality and diverse data.

And as a result of this, only one week of data being used for WHAMM training. This is a substantial decrease from the 7 years of gameplay that Microsoft used to train WHAM-1.6B.

Lastly, the team doubled the resolution of WHAMM’s output, increasing it to 640×360 from WHAM-1.6B's 300×180.

And in order to make the game capable of providing real-time experience like the original Quake II, the team ditched the autoregressive LLM-like setup, where WHAM-1.6B would generate 1 token at a time, to a MaskGIT, which allows the team to generate all of the tokens for an image in as many generations as they want.

Microsoft creates Quake II using AI
WHAMM tokenizes each image using a ViT-VQGAN, exactly like WHAM (left). The Backbone transformer takes in the context to predict the tokens for the next image (center). The Refinement transformer iteratively refines the image token predictions by repeatedly masking and predicting them (right)
"The fun part is then being able to play a simulated version of the game inside the model. After the release of WHAM-1.6B only 6 short weeks ago, we immediately launched into this project on training WHAMM for a new game. This entailed both gathering the data from scratch, refining our earlier WHAMM prototypes, and then the actual training of both the image encoder/decoder and the WHAMM models."

The result is a Quake II-like game, where players can wander around, move the camera, jump, crouch, shoot, and even blow-up barrels similar to the original game.

Microsoft researchers said that the game, which is created by Microsoft's Muse family of AI models for video games, allows users to "interact with the model through keyboard/controller actions and see the effects of your actions immediately, essentially allowing you to play inside the model."

While the demo is indeed an achievement on its own, which also boasts the gaming capabilities of Microsoft’s Copilot AI platform, the company said that the experience isn’t quite the same as playing a well-made game.

More specifically, the "limitations and shortcomings" include enemies that are fuzzy, the damage and health counters that can be inaccurate, and object permanence that is difficult to achieve.

In other words, the game is far from polished, come with a bad resolution, and laggy by modern standards.

Then, there is the fact that AI consumes a lot of resources to train and operate, meaning that behind the scene, Microsoft must've spent a lot of computing power to make this AI work.

And let's not forget how the gaming industry would be impacted by this kind of innovation.

[block:block=87]

Despite the potential for generative AI to assist developers, the world has seen way too many creatives in fields like music and art being replaced by AI, and that the video game industry was hit by some serious layoffs over the past year, including at Microsoft's own gaming division.

"The most important of which is this is a generative model," said Microsoft.

"Thus, we are learning an approximation to the real environment whose data it was trained on. We do not intend for this to fully replicate the actual experience of playing the original Quake II game. This is intended to be a research exploration of what we are able to build using current ML approaches. Think of this as playing the model as opposed to playing the game."

Quake II, first released in December 9, 1997, was developed by id Software and published by Activision.

The game that was released one year after the original Quake, is considered amongst true legends in the world of first-person shooters, like Wolfenstein 3D in 1992, DOOM in 1993, and Half-Life in 1998.

Published