AI Still Struggles to Create Compelling Video Game Worlds, and It May Never Succeed
This is The Stepback, a weekly newsletter diving into a key story from the tech landscape. For insights on the video game industry’s response to generative AI, keep an eye on the latest updates. The Stepback lands in subscribers’ inboxes at 8 AM ET.
How it started
Long before the surge of generative AI, game developers were crafting titles that could create their own worlds. Games like Minecraft and the original 1980 Rogue, which inspired the term “roguelike,” generate environments dynamically based on specific rules and parameters. Developers invest significant effort to ensure these worlds are engaging and filled with activities, making them replayable for years as each playthrough feels fresh and unique. However, as other creative sectors resist the encroachment of AI, the gaming world is also facing the challenge, even if generative AI may never match the creativity of human developers.
How it’s going
Generative AI has sparked heated debates among gamers, with many expressing frustration over subpar in-game content, while a significant portion of developers view it as detrimental to the industry. Despite the backlash, major gaming companies are diving into AI. For instance, Krafton is rebranding itself as an “AI First” company, EA is collaborating with Stability AI to develop innovative game-making tools, and Ubisoft is committing to “accelerated investments” in player-facing generative AI as part of a major restructuring. The CEO of Nexon, which owns the studio behind last year’s hit Arc Raiders, ominously stated, “I think it’s important to assume that every game company is now using AI.” (Some indie developers hold a different perspective.)
Larger companies often frame their AI initiatives as a means to streamline game development, which is becoming increasingly costly. However, the integration of generative AI tools poses a potential threat to jobs in an industry already known for frequent layoffs.
Recently, Google introduced Project Genie, an “early research prototype” that allows users to create sandbox worlds using text or image prompts, which they can explore for 60 seconds. Currently, this tool is only accessible in the US to subscribers of Google’s $249.99-per-month AI Ultra plan. Powered by Google’s Genie 3 AI world model, which the company touts as a significant step toward AGI, it aims to enable “AI agents capable of reasoning, problem-solving, and real-world actions.” Google claims the model’s applications extend beyond gaming, but it has garnered attention as a first glimpse into how generative AI could influence game development, similar to how tools like DALL-E and OpenAI’s Sora have showcased the potential of AI-generated images and videos.
In my experience with Project Genie, the generated experiences were far from captivating. The “worlds” offered little more than a chance to wander using arrow keys. Once the 60 seconds elapsed, the only option was to download a recording of the session, with no ability to integrate the generated content into a traditional game engine. While I did manage to create some poorly executed unauthorized Nintendo knockoffs (likely influenced by the online content Genie 3 was trained on), they were nowhere near the quality of a crafted Nintendo game. The environments were silent, physics were clunky, and the overall design felt basic.
Following the announcement of Project Genie, stock prices for major gaming companies like Take-Two, Roblox, and Unity dipped. This prompted some damage control, with Take-Two president Karl Slatoff firmly stating in an earnings call that Genie does not pose a threat to traditional games. He emphasized, “Genie is not a game engine,” and argued that it does not replace the creative process, likening it more to “procedurally generated interactive video at this point.” (Stock prices rebounded in the days that followed.)
What happens next
Google is likely to continue refining its Genie world models and tools for generating interactive experiences. It remains uncertain whether the focus will be on enhancing these experiences as games or on leveraging Genie to support its broader ambitions toward AGI.
Meanwhile, other AI leaders are advocating for interactive AI experiences. Elon Musk of xAI recently claimed that “real-time” and “high-quality” video games tailored to individual players will be available “next year,” and he mentioned that establishing an “AI gaming studio” is a significant project for xAI. Meta’s Mark Zuckerberg, now promoting AI as the next frontier in social media after job cuts in its metaverse division, envisions a future where users can create games from prompts and share them in their feeds. Roblox is also exploring how creators can utilize AI world models and prompts to dynamically generate and modify in-game environments, a concept they refer to as “real-time dreaming.”
However, even in the most optimistic scenarios where AI technology can generate worlds as engaging and responsive as those in traditional games, creating a video game involves much more than just world-building. The best games feature compelling gameplay, diverse activities, and original art, sound, writing, and characters. Human developers often spend years ensuring that all these elements harmonize perfectly.
Currently, AI technology is not equipped to create fully realized games, and anyone who believes otherwise is mistaken. While AI-generated video has shown some improvement, it still struggles with quality, as evidenced by the lackluster ads produced for the Super Bowl. Consequently, tech companies are likely to continue investing in generative AI for gaming. In an already volatile industry, the mere notion that AI tools could rival human creativity could have significant implications in the future.
The complexity of game development is distinct from AI-generated video, which has advanced rapidly but involves fewer variables. While AI game-making tools will likely improve, they may never fully bridge the gap between AI-generated content and what human developers can create.
By the way
* Unity CEO Matthew Bromberg argues in a lengthy post that world models are not a threat but a “powerful accelerator.”
* Although the gaming industry may not need to feel threatened by AI world models just yet, generative AI tools will remain a contentious topic in game development. Even Larian Studios, known for titles like Baldur’s Gate 3, faces backlash.
* Steam mandates that developers disclose when their games utilize generative AI for content creation, but a recent change allows developers to omit disclosure if they used “AI-powered tools” in their development processes.
* Some games, such as the text-based Hidden Door and Amazon’s Snoop Dogg game on its Luna cloud gaming service, are integrating generative AI as a fundamental aspect of gameplay.
