Generative AI is not yet revolutionizing game development

Video creation the game requires hard, repetitive work. How could it not be? Developers are in the business of creating the world, so it’s easy to see why the gaming industry is excited about generative AI. While the computers were doing boring things, a small team could create a map the size of San Andreas. The crunch is a thing of the past; release of games in the finished state. The new age beckons.

There are at least two interrelated problems with this narrative. First, there is the logic of the crypto/Web3/Metaverse gold rush hype itself, which, consciously or not, seems to consider automating the work of artists as a form of progress.

Secondly, there is a gap between these statements and reality. Back in November, when DALL-E was seemingly everywhere, venture capital firm Andreessen Horowitz posted long analysis on their website touting a “generative AI revolution in games” that will do everything from reducing development time to changing the type of games they create. The following month, Andreessen partner Jonathan Lai posted Thread on Twitter exposition “Cyberpunk where most of the world/text was generated, allowing developers to move from asset production to higher-order tasks like storytelling and innovation” and theorize that AI can enable “good+fast+affordable” game creation. Eventually, Lai’s mentions filled with so many annoyed replies that he posted second thread recognizing that “there are definitely a lot of problems that need to be addressed”.

“Honestly, I’ve seen some ridiculous claims about things that are supposedly just around the corner,” says Patrick Mills, current head of franchised content strategy at CD Projekt Red, developer Cyberpunk 2077. “I saw people suggesting that AI could create Night city, For example. I think we are far from that.”

Even those who advocate generative AI in video games feel that a lot of the industry’s rave talk about machine learning is getting out of hand. It’s “ridiculous,” says Julian Togelius, co-director NYU Game Innovation Lab, who is the author of dozens of articles on the subject. “Sometimes it seems like the worst crypto-bros left the crypto-ship as it was sinking, and then they came in here and said, “Generative AI: start the hype machine.”

This is not to say that generative AI cannot or should not be used in game development, explains Togelius. The point is that people are not realistic about what he can do. Of course, the AI ​​can design standard weapons or write some dialogue, but compared to generating text or images, the level design is just devilish. You can forgive generators that create a face with crooked ears or a few lines of gibberish text. But a broken game level, no matter how magical it looks, is useless. “This is bullshit,” he says, “needs to be thrown out or repaired by hand.”

Basically – and Togelius has talked to several developers – no one needs level generators that work less than 100% of the time. They make games unplayable by destroying entire titles. “That’s why it’s so hard to take generative AI, which is so hard to manage, and just put it in there,” he says.