Microsoft recently announced the beta launch of Copilot for Gaming, planned for April on the Xbox platform. This AI assistant is envisioned as a constant companion for players, offering tips and hints on demand. Examples of its intended use included providing advice on selecting a character in Overwatch 2 or helping a new player get started in Minecraft.
According to a Microsoft representative, Copilot for Gaming is designed to access public information sources, primarily leveraging the Bing search index and results. It aims to provide tailored responses based on its understanding of the individual player’s activity and the games they are currently playing on the Xbox platform. Microsoft also plans for Copilot’s advice to source the “most accurate game knowledge,” which includes working directly with game studios.
Table of Contents
Scepticism and the “Solution in Search of a Problem” Argument
Some sites like GeekWire question the fundamental need for this feature, posing the question: Why would someone want to use this at all? What is the actual target audience?. Machine learning, as a whole, is currently in a phase where many new projects are “solutions in search of a problem,” with advocates focusing on what technology can do rather than what it should do. They place Copilot for Gaming squarely in this category, viewing it as an attempt to improve the gaming hobby by removing certain human elements. The stated goal of not having to stop playing a game to look something up, with Copilot offering online hints or personalised coaching, is presented as part of this.
Concerns About Accuracy and Content Integrity
A significant concern raised is the accuracy of Copilot for Gaming. Unless every Xbox game is planned to have direct integration with internally written Microsoft material, Copilot is expected to still have issues with overall accuracy, similar to the standard Copilot. A recent study by the Columbia Journalism Review found that Copilot is wrong more often than it is right.
Microsoft's demo of how its AI Copilot will interact with Xbox games on Windows in the future was one of the most interesting demos today. It's just a canned demo, but it shows the potential for an AI assistant that can see what you're doing and guide you pic.twitter.com/6MCtTWmckt
— Tom Warren (@tomwarren) May 21, 2024
A Microsoft representative acknowledged this, noting that it’s important to check AI results against other trusted sources. The review questioned the utility of Copilot for Gaming if players still need to verify the information, especially given that it might take away “much-needed web traffic away from actual humans who want to provide me with the same guidance”. The Columbia Journalism Review author, who writes strategy guides, admits a bias but argues that players value accuracy over convenience. They predict an “unrecoverable PR hit” the first time Copilot for Gaming provides incorrect gaming advice.
A Promising Feature: In-Game Memory Assistance
Among the planned features, one stood out as potentially useful: the ability to remind a player where they were and what they were doing the last time they played a particular game. This feature is compared favorably to a menu option in the 1993 game Phantasy Star IV. It can be noted that many games are long or complex enough that players might forget their progress if they step away for a few days, making a built-in refresher appealing. Aside from this reminder feature, the rest of Copilot for Gaming is viewed as another example of an AI project whose “reach that exceeds its grasp”.
Microsoft is asking players to trust an AI for accurate information at a time when “AI-driven search is still distinctly flawed”. This is seen as a “dumb move from a company that’s smarter than people think”. Microsoft’s successes in accessibility efforts, such as the Xbox Adaptive Joystick and the Adaptive Controller, as well as Xbox Game Pass and outreach to indie developers, suggest the company is generally more intelligent about the video game business than it’s given credit for.
Frustration with AI in Gaming and Unrealistic Expectations
The frustration stems from seeing AI researchers and developers attempting to bring products to market that are perceived as not working as advertised. The best case scenario for Copilot for Gaming, in this view, is being sort of helpful sometimes, at the cost of removing more of the human element from gaming. Complementing the broader Copilot for Gaming initiative is a recent technical demonstration: Microsoft’s release of a browser-based, AI-generated version of the classic game Quake II. This demo is powered by Microsoft’s Muse AI model, which is part of the Copilot AI platform. Quake II, released in 1997 by id Software, is considered one of the most iconic shooters of all time.
The AI-generated demo is intended as a technical showcase of Microsoft’s AI capabilities in gaming. Unlike traditional games with preset levels, the content in the AI version of Quake II is generated “on the fly by an AI world model,” based on the style and content of the original game. Microsoft researchers explained that their Muse family of AI models for video games allows users to “interact with the model through keyboard or controller actions and observe the effects of their actions instantly, essentially enabling them to play within the model”.
Microsoft trained their model on Quake II, a game they own through the acquisition of ZeniMax. Muse was reportedly trained on data, including over a billion images and controller actions, from human players of the 2020 Xbox game Bleeding Edge. Microsoft has also been vocal about the potential of Muse AI to help preserve classic games by learning how a game plays without needing the original engine or hardware.
However, many have stated that the Quake II AI demo is a limited and technically flawed experience. While it is playable in a web browser, users should temper their expectations. The demo runs at a resolution of 640 x 360. The frame rate is estimated at around 15fps and fluctuates significantly, with some sessions being very laggy and others almost passable. Enemies appear incredibly blurry, which diminishes the horror element intended in the original game. The demo also lacks audio. Furthermore, Microsoft has restricted the amount of time users can spend playing in a single session.
Limitations and Shortcomings
Microsoft researchers also acknowledged the limitations and shortcomings of the demo. Beyond the blurry enemies and inconsistent performance, issues include inaccurate damage and health counters. Most notably, the AI model struggles with object permanence, often “forgetting about objects that are out of view for 0.9 seconds or longer”. Other limitations noted include inconsistent texture generation, lack of dynamic animation, and difficulties with contextual understanding compared to a human artist. Compared to playing the original Quake II or its remasters, the AI demo provides less sharp visuals, inconsistent performance, and diluted immersion.
The release of the AI-generated Quake II demo prompted varied and strong reactions, particularly from the gaming community. When Geoff Keighley shared a demo video on X, it received overwhelmingly negative responses. Critics on platforms like Shacknews called it “regurgitated AI slop” and “absolute garbage”, noting its worse performance than the 1997 original, despite requiring more computing power. Some commenters linked the perceived poor quality to Microsoft’s acquisition strategy and recent layoffs. The demo was seen by critics as a disappointing use of the Quake IP, especially given the state of titles like Quake Champions.
However, the demo also found a prominent defender in John Carmack, one of the co-creators of the original Quake. Carmack argued that AI tools will allow the best developers to reach even greater heights. He also stated that AI can enable smaller teams to accomplish more and “bring in some completely new creator demographics”.
While acknowledging that AI might lead to fewer jobs, he dismissed anti-AI sentiment by saying, “‘don’t use power tools because they take people’s jobs’ is not a winning strategy”. He believes we will reach a world where interactive games (or novels, or movies) can be generated from a prompt, but predicts that “far better exemplars of the medium” will still be created by passionate development teams.
Muse AI Potential: Prototyping, Remastering, and Accessibility
Microsoft continues to promote Muse as an AI model that can aid game developers in prototyping their games. They frame the Quake II demo as a research exploration, describing the experience as “playing the model as opposed to playing the game”. The project serves as a tech showcase for what AI might achieve in game development. Experts view these limitations as a natural part of technological evolution and see the demo as a valuable proof-of-concept, demonstrating the feasibility of automating certain aspects of game design. The current imperfections are not seen as entirely discouraging but rather as steps toward future breakthroughs, suggesting a future involving hybrid methods where AI-generated assets are refined by human artists.
The broader implications for AI in game development, particularly for Windows gaming, include the potential to dramatically reduce the time and cost associated with remastering classic games, potentially leading to more affordable, high-quality remasters. AI could also assist developers in producing rapid prototypes, shortening development cycles. Generative AI could enhance accessibility and inclusivity by automating localisation processes. For developers, AI could serve as a “co-creator,” helping generate assets or design concepts quickly, allowing them to focus on refining gameplay and storytelling. However, this also presents new challenges in integrating AI outputs and may lead to specialised roles focused on optimising AI contributions.
Industry perspectives, like that of Nintendo of America President Doug Bowser, still emphasise the crucial role of human developers, their artistic capabilities, and insight into how people play. The general consensus seems to be that AI is best used as a tool that amplifies human creativity rather than replacing it entirely. The AI-generated Quake II demo highlights the ongoing journey towards a “seamless marriage of technology and art,” acknowledging that while the future is digital, it is also intertwined with the human legacies that inspire it.
Conclusion
In conclusion, Microsoft’s Copilot for Gaming initiative and the AI-generated Quake II demo represent the company’s active exploration of artificial intelligence in the gaming space. Copilot for Gaming is designed as a player assistant offering tips and hints, although concerns about its utility and accuracy exist. The Quake II demo, powered by Muse AI, is a browser-based technical showcase demonstrating the potential of AI to generate game content on the fly and assist in areas like prototyping and game preservation.
Despite these potentials, the demo itself is described as rudimentary, with significant limitations in performance, resolution, graphical clarity, and AI understanding. Reactions are polarised, with some viewing it as disappointing “slop” and others, like John Carmack, defending AI as a powerful tool for empowering creators. While challenges remain and human artistry is still seen as critical, these experiments suggest a future where AI will play an increasingly significant role in game development, potentially transforming areas like remastering and prototyping, though likely in a hybrid approach alongside human input. This iterative innovation, acknowledging both successes and shortcomings, points towards a future where technology and creativity continue to redefine interactive entertainment.
