To put it mildly, longtime Dragon Age narrative lead David Gaider, who's long since left Bioware to become a co-founder of Malys and Stray Gods maker Summerfall Studios, is not impressed by generative AI. To put it less mildly, he reckons the technology is a "virulent plague" among games.
Speaking with GamesRadar+ as part of a larger report unpacking why so many game developers don't want to use generative AI, Gaider examines the proposed benefits of the tech alongside its observable effects.
"I think the fact that generative AI is frequently trained on data regardless of whether creators or owners have agreed to have their data pillaged in this manner opens up any use of it to all sorts of future legal issues - even if one chooses to ignore the moral implications, which one really shouldn't," Gaider begins.
It's often unclear exactly what materials AI models were trained on, but given the track records of many of the biggest companies in the space, you can safely assume that a significant chunk of that material was scraped without permission. Gaider says, "'If we're not allowed to steal whatever we need, then the AI won't work as well!' isn't a very compelling argument."
He examines one commonly supported use case for gen AI in games: the "concepting or iterating phases," where there may be no plans for AI-made assets to be included in the final game. But he sees risks even with sparing or early use. We've already seen plenty of games mistakenly ship with placeholder AI assets. "All you'd need is one lazy developer or one temp asset that's been forgotten or was placed by someone who's since left the team and you'd have an issue on your hands," Gaider says.
"Honestly, what does it help with?" he wonders. "Does it make the work more efficient? Does it improve the work? It wouldn't be so bad if generative AI was seen more as an assistant, doing the drudgery while leaving more important tasks for the worker, but we seem to be seeing more and more of the reverse: the AI is set to do the important work and the worker is around to 'clean up'.
"In all my time as a narrative designer I've never once encountered a situation where editing an inferior product took less time than simply throwing it out and redoing it would have or resulted in anything better than mediocre. And while there's potential for AI handling the drudgery, I also think we have to be very careful about not eliminating every task which is useful for training juniors. How are we going to train up the next generation of devs if we eliminate every entry-level task?"
Gaider struggles to find an angle on gen AI that doesn't render "the entire thing problematic." He sees "some potential" for it as a "useful development tool" in theory, but not in the technology we have today. "It's not ready for prime time. There's just a lot of executives who really, really want it to be," he says.
So, he doesn't want to use gen AI at all. While he hopes there's potential in it, he stresses that "generative AI is terrible at iteration" because you can't "tell it to adjust minor things and get a consistent result," which is kind of the whole point of iteration.
"It would be frustrating as hell," he continues. "I can't even imagine using it for bigger tasks like programming. How does one bug fix 'vibe coding'? What's the point of creating prototypes with AI when the result is that nobody on the team has actually learned anything about how to make the final product? Why use AI to create concepts which are inevitably going to be soulless and contain errors and which aren't going to be something your own artists can replicate? Why have systems that nobody on your team really knows how they work? I could go on and on."
And he does, indeed, go on. Gaider echoes many developers who don't wholly reject the idea of supporting development with new technology in this space, but do wholeheartedly reject gen AI as we know it today. Asked how he wants to see gen AI treated in the games industry, Gaider minces no words.
"Until some regulation is in place? Until we can be confident that it's only trained on legally sourced data? Until the people making decisions regarding its use finally realize that it's not the source of cheap labour replacement they want it to be, and don't cut off their teams at the kneecaps to force it on them while expecting unrealistic results? It should be treated like the virulent plague it is."
