create video Games require hard, repetitive work. How could it not? Developers are in the business of building worlds, so it’s easy to see why the games industry would be excited about generative AI. With computers doing all this nonsense, a small team can make a map the size of San Andreas. Crunch is a thing of the past; games are released in a finished state. A new era is calling.
There are at least two interrelated problems with this narrative. First, the logic of the hype itself—reminiscent of the crypto/Web3/metaverse mad gold rush—seems to see the automation of artists’ jobs as a form of progress, whether intentionally or not.
Second, there is a gap between these claims and reality. Back in November, when DALL-E seemed to be everywhere, venture capital firm Andreessen Horowitz published a long-form analysis on their website touting a “generative AI revolution in games,” from shortening development time to Everything from changing the type of game being made.The following month, Andreessen partner Jonathan Lai posted a twitter thread elaborate a “cyberpunk Most of the world/text is generated here, enabling developers to move from asset production to higher order tasks such as storytelling and innovation” and theoretically AI can achieve “good + fast + affordable” The game’s production. Eventually, Lai’s mention was filled with so many annoyed replies that he posted a second thread Acknowledging that “there are definitely a lot of challenges that need to be addressed.”
Patrick Mills, CD Projekt Red’s acting head of franchise content strategy, said: “Frankly, I’ve seen some ridiculous claims about what’s coming.” Cyberpunk 2077“For example, I’ve seen people suggesting that AI could build Nightside. I think we’re a long way from that.”
Even those advocating the use of generative AI in video games argue that many of the exciting discussions about machine learning in the industry have gotten out of hand. It’s “ridiculous,” says Julian Togelius, co-director of NYU’s Game Innovation Lab, who has written dozens of papers on the topic. “Sometimes it feels like the worst crypto bros left the crypto ship when it was sinking, and then they come here and it’s like, ‘Generating AI: Start the hype machine.’”
That’s not to say generative AI can’t or shouldn’t be used in game development, Togelius explained. People are not realistic about what it can do. Sure, an AI could design some generic weapons or write some dialogue, but compared to text or image generation, level design is diabolical. You can forgive a generator that produces a face with crooked ears or a few lines of gibberish. But a crappy game level, no matter how amazing it looks, is useless. “It’s bullshit,” he said, “and you need to throw it away or fix it manually.”
Basically – and Togelius has had this conversation with multiple developers – nobody wants a level generator that works less than 100% of the time. They make the game unplayable and ruin the whole game. “That’s why it’s so hard to put generative AI that’s hard to control,” he said.
