Exploring the challenges of AI-generated art in game development

Speaking at Devcom today, Red Meat Games creative content strategist Judy Ehrentraut discussed the importance of ethically training generative AI models, and how certain tools can be utilised during game development.

"AI is the biggest buzzword and it's either hyped as the new way to solve every productivity problem, or it's received with a groan," Ehrentraut acknowledged.

"I think both of these takes are really valuable, because disruptive technologies are not necessarily good or bad. It depends on how they are used, and it really depends on the approach that we take and whether it's ethical or not."

But as Ehrentraut highlighted, AI can only learn what it's being trained.

"AI is not intelligent in the way that we keep thinking that it is, or that people are hyping it up to be. It's just a tool that can learn from instructions that we give it, whether it's a large language model or algorithms collecting a bunch of data, sweeping it from the internet and creating something called art when really it's just a mashup of many artists' work – there's no real human intent behind it."

She noted that, because of this, gen AI tools are giving people a certain perception of the art that it creates.

"A lot of artists are debating the effects of this disruptive technology, and many will say they enjoy the process of creation," Ehrentraut said.

"They enjoy the time it takes, it's all part of learning and production. They don't tend to enjoy that something can be produced in one second by a machine, because that takes away the intent behind it. It takes away the work, the practice, the collective intelligence that people have shared with others."

She continued: "We know at this point that removing artists from the creative process is not the best idea. It's not fair to use the art one time then produce, produce, produce, until you've reached something that's almost nothing – and that's what we're seeing now.

"The internet is becoming littered with AI images at a very rapid rate, and AI is using the content that it's produced to further train itself. It's not being trained on human content – it's training itself on its own content, and then it doesn't understand that it's getting things wrong and keeps going."

There is currently a lot of contention around gen AI, especially with tools such as Midjourney and Open AI's Dall-E which are accused of scraping artwork from the internet without an artist's permission.

"AI is not intelligent in the way that we keep thinking that it is, or that people are hyping it up to be"

Red Meat Games have been very open about using gen AI during development of its latest game, Moriarty.

But instead of using something like Midjourney, it opted for Scenario – a tool that enables artists to input their work and train an AI model instead of it taking art from the internet without consent.

"Our artists have drawn all of these things, and we're training Scenario to create more versions of our characters," Ehrentraut explained.

"We're not using Scenario to replicate subjects, objects and style. We're using it to generate new art in real time, because we're trying to give our game long term replayability.

"We have a small team of artists, and this [sort of tool] can be very helpful for studios who have a small team and want a greater output, because you're essentially just expanding on what you've already created. You're not just creating something from nothing. And when I say nothing, I mean art that the AI you're using has been stolen from Google."

Ehrentraut clarified that by using Scenario, the team is more in control of the process as the tool is specifically doing what is being asked of it.

"I believe this is more ethical, as we're not so much training the AI from the ground up – the system already has a basic understanding of concepts and objects. What it's learning now is what you actually give it."

For Red Meat Games, it's about utilising gen AI as a tool to "encapsulate a more symbiotic relationship between the human and AI system."

"We need to ascribe a lot more ethical responsibility and transparency to AI," Ehrentraut concluded. "We need to think of AI as one [aspect] in a larger production so that it's not the artist or creator – it's just a tool that the artists and creators are using, and together everybody creates this thing we're then calling art."