How generative AI is changing game development (and UGC)

Artificial intelligence has been a fundamental part of games for decades. It’s neither a new nor revolutionary part of game development. However, generative AI is a new tool in developer’s arsenals. Proponents say GenAI has the capacity to revolutionize game development — but the technology also has fierce critics.

Typically AI is most commonly used in games to generate in-game behaviors in non-player characters (NPCs) or new environments in a procedurally-generated games (e.g., roguelikes). Generative AI, on the other hand, adds several new use cases for the technology from the earliest phases of a game’s development.

>>Follow VentureBeat’s ongoing generative AI coverage<<

Generative AI is rising in popularity, thanks in large part to tools like ChatGPT, Midjourney and others. The games industry is no exception, as generative AI tools have begun to integrate into game development and the creation of user-generated content.

Generative AI’s creep into game development

One way developers are already incorporating AI into their development process is to generate art and assets. Generative AI trained with a handful of images can produce a significant number of similar pieces faster than a human could. Additionally, tools such as DALL-E 2 and ChatGPT can be used to by game writers to craft original stories, expand upon core ideas or generate in-game text.

Recently, Scenario opened up its generative AI platform to developers. It purports to offer individually trainable generators, which developers can use to create art pieces tailored to their specific style. Scenario CEO Emmanuel de Maistre demonstrated to GamesBeat how this would work by generating dozens of small art pieces based on a series of prompts. de Maistre also showed how the process worked on Twitter, generating several images of potion bottles using Scenario.

Some developers are also utilizing AI in character creation. Ready Player Me, a metaverse avatar creation platform, recently launched a trial of its generative AI creator. Players can use AI to create their own in-game clothing.

According to RPM CEO Timmu Tõke, the point of using the AI is to give players more involvement in their avatar’s look. He told GamesBeat, “Generative AI will definitely change how 3D content is created and avatars are created. We are aiming to be at the forefront of that and understand how things are changing and then change ourselves with that.”

Other developers have used generative AI to improve on more “traditional” in-game artificial intelligence. For example, Inworld AI plans to use the tech to give NPCs in games more dialogue in response to what the player says to them. They’ve used this already to create a demo of a virtual Santa Claus who responds to children’s interactions with him.

User (and AI) generated content

However, some of the biggest waves AI is making in game development aren’t just in the world of mainstream game development. As Dean Takahashi wrote recently, AI tools are becoming a staple in the creation of user-generated content (UGC). This type of content abounds in games such as Roblox and Minecraft, and the creators who used the technology have accrued acclaim and profits. Roblox is even rolling out its own genAI tools so users can build content for the game more easily.

In the last few months alone, AI tools have helped users create a horror visual novel, multiplayer games on LG TVs and their own NFTs. Content creators are using tools like ChatGPT to mod existing games to have more dialogue (similar to the Santa demo above), while others are using it to make new text-based adventure games.

In the discussion of skilled human labor versus AI output, user-generated content would fall somewhere in the middle. While some users might have the requisite skills required to create in-game items to their liking, many do not. Generative AI would, in theory, bridge the gap between what users can envision and what they can create themselves.

Even this use case is not without its issues. Generative AI requires training sets — input data to power its AI model. There could be legal implications if users train AI with copyrighted images without permission.

Say that we can use genAI in games — should we?

Not everyone in the game development world is so sanguine about the use of AI in development. Recently, Justin Roiland, formerly of Squanch Games, revealed that the developers of first-person shooter High on Life used images generated by Midjourney as in-game decorations. Lead designer Erich Meyr also revealed the team used AI to prototype voice lines, and one minor character in the game kept its AI voice.

Around the same time, former Riot Games producer Jon Lai (now a general partner at a16z) tweeted that AI could reduce the time and costs of creating a large triple-A game. Using a variation on the expression, “Fast, Cheap and Good . . . pick two,” Lai said, “GenAI promises to break this triangle: enabling good + fast + affordable.”

The response to Lai’s tweet was decidedly mixed. While some applauded the idea of enabling smaller developers to create games with bigger scope, others pointed out that using AI to generate in-game assets or art could mean fewer opportunities for artists and developers to contribute to the game. High on Life provides an example of this, as AI was used to generate art that otherwise would have been made by one of Squanch Games’ artists.

Roiland told Sky News the AI art was used, in part, because it made the alien world feel more unlike our own. He also added, speaking about AI in game development, “I don’t know what the future holds, but AI is going to be a tool that has the potential to make content creation incredibly accessible. I don’t know how many years away we are, but all you will need to be is somebody with some big ideas.”

As generative AI becomes more sophisticated and omnipresent, it will likely be used more in game development. What form that will take — and whether the fears that it will replaced human creators will be realized — remains to be seen.