As Video Game Industry Goes All-In On AI, Workers Race To Secure Their Rights

We’re not even two weeks into 2024, and already Square Enix has declared that it plans to be “aggressive in applying AI” to its game development processes. Nvidia has announced that partners – including Ubisoft, Tencent, and MiHoYo – are now using its “Ace” production services, which draw on generative AI to create fully-voiced NPCs. Wizards of the Coast and Electronic Arts have been caught haphazardly cranking out AI-touched images, the former after late-2023 layoffs ravaged its art department. But game developers are pushing back.

With a well-documented overreliance on crunch and layoffs even in the best of times, the video game industry has long demonstrated a tendency to treat the human beings that bring its worlds to life like so much trimmable fat. AI represents a new frontier in executives’ endless quest for “efficiency” (which curiously doesn’t involve not paying themselves tens of millions of dollars per year): a desire to combine multiple jobs into one, or automate parts of game development entirely. Nvidia aims to do this with NPCs, as does Ubisoft through the aforementioned Nvidia partnership and also with its own “Ghostwriter” tool that generates “barks,” or basic lines of character dialogue. Microsoft also hopes to be a major player in the AI space via a partnership with Inworld, a self-described “character engine” for NPCs that can enable “entirely new narratives with dynamically generated stories, quests, and dialogue for players to experience.” 

Game developers and performers are doing their best to get out ahead of the impending AI-fication of their profession and protect their rights. 

"We're trying to make sure that all of AI's uses become very human centric,” Conor O’Donnell, a tester and union member at Zenimax, a Microsoft subsidiary, told Aftermath. “We want it to be centered around people and the workers and making sure that the work is the work and the workers are being respected.”

The Communications Workers of America (CWA), of which the 376-member Zenimax union is a part, announced in December that Zenimax workers had reached an agreement with Microsoft to curb potentially destructive uses of AI. The legally binding agreement commits Zenimax – responsible for series like The Elder Scrolls, Fallout, Starfield, and Doom – to uses of AI that "augment human ingenuity and capacities ... without causing workers harm" and requires that Zenimax provides notice to the union in cases where "AI implementation may impact the work of union members and to bargain those impacts upon request." It also states that workplace AI implementations must adhere to six guiding principles: fair, reliable and safe, private and secure, inclusive, transparent, and accountable.

"We have agreed that AI will not be used to replace a person's job and that it would only be used to assist a job that exists – and not to either reduce the number of positions or to replace an open position or anything along those lines," Dylan Burton, senior QA tester and Zenimax union member, told Aftermath. “The other thing that's important is that there is a person who is accountable if a decision or process or whatever AI is being used for causes harm. And it's not like, 'Oh, AI did it. So who knows why this happened?'" 

Claude Cummings Jr, president of the CWA, views AI running rampant as something only organized workers can prevent – not just in the video game industry, but across the board.

"AI is a threat,” he said. “One of the CEOs, [Bobby Kotick, now-former CEO of Activision], met with me, and he's like, 'I got you a gift.' It was a poem that AI had written in less than a minute about me. ... My hope and desire is that whether it's the gaming industry or the telecommunications industry, we can negotiate standards that will not allow a lot of work to be taken away from us because of AI."

Game workers see other issues with AI beyond it potentially replacing jobs. AI needs source material to create its output, which raises questions about what data sets tools are being trained on to, say, generate voices. 

“AI sources content from places,” Burton said. “It doesn't make it from nothing. So especially in cases of commercial projects, you have to be very aware that if your AI is sourcing content to present to you from copyrighted sources, that alone is a pretty big potential liability – let alone a more subjective conversation about artistic value.”

The developers of The Finals, a new multiplayer first-person shooter that’s blown up on Steam, say they source their AI-powered voices from “a combination of recorded voice actors and AI-based [text to speech] that is based on contracted voice actors.” But even that level of transparency – which allows the studio to publish the game on Steam despite Valve’s policy of removing AI-powered games when it’s unclear if they have “sufficient rights to the training data” – is rare at this stage.

Other developers have said that where video games are concerned, generative AI is a solution to a problem that doesn’t exist: Video game designers are overflowing with creativity. They don’t need help coming up with characters, dialogue, or quest lines. That’s the fun part of the job, and while the idea of AI generating those things in real time, in a way that reacts to the player, presents some interesting possibilities, the end result would still need to be curated – effectively crafted by human beings so as not to devolve into nonsense unrelated to a game’s world or story. It could take as much (or more) work than current game development methods. And if those elements of a game were generated from a limited – to avoid copyright issues – data set, they’d probably become repetitive pretty quickly, in a way players would doubtless notice. 

“I think there are so many considerations that you need to take into account whenever you're using [AI] that I personally wouldn't ever,” said Burton. “But I know that people will, so we just have to be careful about how they do.”

Zenimax workers, at least, are optimistic. In their eyes, they wrestled Microsoft to the table before a worst-case scenario could unfold. 

"We haven't gotten to that point yet with AI where it is a nightmare scenario and things are falling apart,” said O’Donnell. “So the fact that Microsoft is coming out right at the start and saying they're willing to talk [means] we'll be able to push back on a lot of it. The fact that the video game industry itself is so early in its unionization stages, this is setting a really great precedent for all companies and union efforts moving forward. It's saying that these conversations are very possible."

But so far, Microsoft is the only video game giant to adopt these sorts of contractually binding standards, and even then, evidently not across all of its divisions. When Microsoft first trumpeted its shiny new Inworld AI partnership last November, another group of game workers who stand to be impacted by AI, voice actors, were none too pleased. Video game performers, in conjunction with the Screen Actors Guild-American Federation of Television and Radio Artists (SAG-AFTRA), authorized a strike over AI, pay raises, and other concerns last October following a year of contract negotiations with companies like Activision Blizzard, Electronic Arts, and Epic Games. (Authorizing a strike, notably, is different from actually engaging in one, which video game performers still hope to avoid.) Against this backdrop, Microsoft nonetheless announced the Inworld partnership.

"The thing about Microsoft that really frustrates me is, that announcement was the first I heard about it,” Sarah Elmaleh, who has voiced characters in big-name games like Halo Infinite, Gears 5, and Spider-Man: Miles Morales, told Aftermath. “It is a mandatory subject of labor law that AI rights are something you have to bargain. So for all these projects and subsidiaries under Microsoft, they can't just go ahead and use AI without coming to us and talking about it. I was surprised that that was the first we were hearing about it, like, 'Are you encouraging your developers to commit an unfair labor practice?’" 

For this reason and many others, Elmaleh and other video game performers ended up demonstrating outside The Game Awards in Los Angeles in December, a tactic which – in conjunction with others – has borne fruit: Today SAG-AFTRA announced an agreement with voice tech company Replica that allows game developers to "access top SAG-AFTRA talent.” The agreement, according to The Hollywood Reporter, codifies actors’ negotiation and consent for their digital voice and requires that they be allowed to opt out of their voice double’s continued use in new works. SAG-AFTRA continues to negotiate “in other areas where AI is concerned.”  

Shortly after the demonstration at The Game Awards, while scarfing victory pancakes at a local Denny’s, Elmaleh and fellow performers Linsay Rousseau, David Errigo Jr, and Zeke Alton discussed the process of recording background chatter for Insomniac’s Spider-Man 2– the exact kind of dialogue video game executives might think they can automate. 

"They brought in a bunch of actors for a day of messing around and reacting to each other in real time,” said Elmaleh. “Improvising and the idiosyncrasies and sense of humor and all of that wonderful, beautiful humanness is what fans have been responding to. I've never seen so much love for NPCs in my life." 

Sure enough, people have made numerous viral TikToks just standing still and observing Spider-Man 2 NPCs chatting at length about, for example, how to properly moisturize a baby. These conversations happened almost entirely on the fly.   

"They literally just ran different scenes for us and had us play the scene as that character,” said Rousseau. “So it was like, 'OK, there's a guy talking to a woman. And... go!'"

"In a game like Spider-Man 2 where you have a large open world like New York City, it can be bland and boring as you swing by, but Insomniac really wanted to make it an interesting place where you would stop and see the scenery,” said Alton. “So you could stop and watch a basketball game and listen to the trash talk, or you could stop and watch people getting arrested. And what brought color to that was putting a bunch of actors in a room, giving them a prompt, and then allowing them to improv through it. They did that with 30 different actors in differing loop sessions. So they built out the entire world of New York."

"It's wonderful that it is human,” said Errigo Jr. “There's an organic flow to the conversations. And yes, we are actors. Yes, we're often aiming for the joke. But there's also just bits and bobs of real interaction. I mean, there were a couple of rounds where we were different permutations of couples, and we had conversations about, 'Oh, it's been so long since we've seen you. We've got to have you over, blah, blah, blah.'"

Elmaleh believes that even though you could get the contours of such a conversation with AI, it’d be missing the heart that has made these Spider-Man 2 exchanges so popular.

“They're bringing their own relationship and their own sense of humor,” she said. “They know what makes that person tick and what'll make them light up. You don't get that with computers. You only get that with real, lived experiences." 

Like other unionized game workers, Elmaleh and her colleagues do not seek to prevent companies from using AI. They – like their Hollywood counterparts before them, with whom they are sharing resources and a knowledge base – simply want a fair and consistent set of standards not just in video games, but across all industries.

"I love collaborating. I love the work, so I might be a little bit more on the conservative side of having a replica [of my voice] out there,” Elmaleh said. “But I know folks who are happy to have a replica do work for them so long as there's consistency across the entire spectrum of use: transparency, consent, and compensation. So I'm informed; I get to decide with you and we're collaborating at least on the terms. I'm being fairly paid so that the value of my work, the material that you're going to use, is carried through to the end result."