Artificial intelligence — or something like it — has powered video games since players tried to outwit a computer-controlled paddle in Pong.
Now, some in the industry are saying new generative AI programs can kickstart huge leaps in the way digital worlds are created for gamers to explore.
But others are more cautious, even warning against how the technology could be used to threaten the livelihoods of the programmers, artists and performers who make the games themselves.
“We are extremely against the idea that anything creative could or should take [the] place of skilled specialists, to which we mean ourselves,” said Rebecca Ford, creative director of Warframe, at Digital Extremes in London, Ont.
“The last world we want to live in is the one where the robots get to make all the creative decisions and we don’t. It should be the other way around.”
Artists have raised the alarm over AI programs like Midjouney and Lensa using artists’ work without consent to fuel their image generation via text prompts.
Generative AI programs currently can’t produce the code that forms the building blocks of games. But text and image programs could theoretically provide a font of concepts for characters, locations and storylines just like a novel, film or TV series.
In Hollywood, the actors union SAG-AFTRA and the Writers Guild of America (WGA) are both striking, partially over the use of AI in future film and television projects.
And earlier this month, SAG-AFTRA voted unanimously to send a strike authorization vote to its members against major video game companies like Activision and Electronic Arts. Such a strike would concern voice work and performance capture, not the game developers themselves.
What even is AI in games?
Tanya X. Short, director and designer at Montreal-based Kitfox Games, said “we’re definitely in a new world” when it comes to recent discussions about AI.
“Historically, AI has meant the opposite of what other industries mean by AI,” she said. “AI [in games] is something scripted cleverly in games that give the players the illusion of an entity.”
That could include actions like an enemy soldier in a first-person shooter announcing they’re reloading their gun — giving the player a brief moment for a counter attack. A more nimble enemy could juke, jive and leap to avoid the player aiming their weapon at the incoming threat.
Those behaviours have historically been described as artificial intelligence, even though they’re much simpler in scope than new generative AI programs.
Some developers, meanwhile, say generative AI could promise far more sophisticated feedback — for instance, character dialogue or tactical movements in the heat of battle — that directly responds to the player’s actions, rather than generated from a set list of pre-written behaviours.
John Romero, legendary game designer and co-creator of Doom, told CBC Radio’s The Sunday Magazine that AI could be “really groundbreaking” in fleshing out the world around a player, particularly the way secondary characters react to the human player.
“You can talk to them, and they can talk back to you — and it’s not canned lines that are pre-recorded, but they’re generated in real time based off of all of what’s going on in the game,” he said.
“It’s like true emergent gameplay, which is what we’ve been hoping for, for so long.”
LISTEN | Doom creator John Romero on The Sunday Magazine:
The Sunday Magazine3:10John Romero on AI’s potential for video games
Ubisoft, the studio behind series like Assassin’s Creed and The Division, recently announced they’re working on an AI-powered tool called Ghostwriter.
“Ghostwriter isn’t replacing the video game writer, but instead, alleviating one of the video game writer’s most laborious tasks,” the company stressed high up in a press release.
Ghostwriter could help generate “first drafts of barks,” or the incidental chatter a player might hear from background characters the player hears as they walk through a crowded city, for example.
Several developers told CBC that for now, generative AI could help them workshop early ideas and weed out potentially bad ideas in favour of the ones with more potential.
But they said none of those generated ideas can stand on their own in a final product without further tweaking by the creators themselves.
“You can use AI to prototype a lot of early-bird stuff, but there needs to be that human element to sort of push that towards the finish line,” said Antonio Miceli, founder of the Toronto indie team Megapower games.
Short said developers early in their careers, particularly, will find better inspiration by looking at a mood board, or selection of images and works found on Pinterest than by something made through image AI programs.
“It will show your future collaborators that you respect humans, that you value their art form, that you weren’t willing to throw away their life’s work into a wood chipper in order to forward yourself,” she said.
The oatmeal problem
Game developers have used generative programs for years, says Benjamin Rivers, co-founder of the Toronto-based independent studio Bancy Co.
Take a game like Starfield, whose planetary surfaces — including rocks, plains and mining outposts — are procedurally generated each time you land on the surface. It’s sort of like taking randomized tiles from a board game and setting them on the table in a different formation each time.
“There’s always some element of stuff that gets put in that’s somewhat templated, and then an artist comes in and makes sure everything makes sense. That is to bring an actual human touch on things. But oftentimes it starts from something that is less bespoke than you may think,” said Rivers.
Where generative AI takes it a step further would be to provide millions or more potential variables before that world is set: think thousands of separate alien species to choose from, instead of a dozen or more.
But Short argues that those variables won’t likely produce more meaningful moments that original work can summon.
Instead it would replicate the so-called “procedural oatmeal,” a term coined by game developer and researcher Kate Compton. Even if we can explore a million AI-generated worlds, they’d all feel the same, just like the individual grains in a bowl of oatmeal.
She pointed to Dwarf Fortress, a cult classic PC game published by Kitfox. It’s weird, it’s complex, and every time you play is different thanks to its myriad interlocking systems.
“I think adding AI to Dwarf Fortress would just make it more same-y; more oatmeal. It would not help it become more meaningful,” she said.
“We as designers want the player to encounter a wild variety of experiences that are unique to them, but each different from each other, in a way that right now feels very unrealistic to expect any AI to ever provide.
“But maybe I’ll be proven wrong.”