SAG-AFTRA’s ongoing strike has shone a light on how major Hollywood conglomerates are looking to use generative artificial intelligence as an alternative to paying for human labor—but questions about the technology’s use are also creeping up on actors in the video game industry.
Though no major studios have committed to using AI voices in lieu of human performers, a new phenomenon is putting stress on the voices behind characters like Evie Fry of Assassin’s Creed Syndicate and Long Caster in Ace Combat 7: modders using AI tools to recreate their voices in other games.
In the last year, the modding community has begun using the technology to fully recreate the voices of their favorite actors. Sometimes it’s to make original stories set in the world of their favorite game, other times it’s to create pornographic content that falls outside the scope of the original performance.
Speaking to Game Developer, actors Victoria Atkin and Tim Friedlander (who is also the head of the National Association of Voice Actors) laid out what they and their colleagues have been experiencing: in the last two years, modders have begun using tools like those from ElevenLabs to synthesize their voices, and it’s leaving them feeling vulnerable about the future of their work.
Modders are using AI synthesizers to put Evie Fry in other games
In movies, spies and sci-fi hackers will synthesize voices using just a few sentences from an unwilling subject. In real life, it takes much more audio than that. But voice actors build their profession by producing hours and hours high-quality audio that’s then packaged into commercially available games—making it easy to feed their voice into generative AI tools.
Atkin and Friedlander say their development partners haven’t done anything to directly contribute to this problem—but they’re left in a position where they aren’t able to take much action. Friedlander pointed out that when he, Atkin, and other actors signed their performances over to employers like Ubisoft, nothing in their contracts offered any protection or guarantees about how this technology might be used. “We didn’t even know to have safeguards in our contracts years ago because these things didn’t exist,” Friedlander explained.
As recently as June 2023, modders have been using voice technology to bring Atkin’s voice into games like The Elder Scrolls V: Skyrim. Atkin said she’s been sent videos of her voice appearing in other games—it prompted her to go to Twitter and say that AI had “stolen her voice” for the role of Evie Fry.
Here’s where things get complicated: Atkin and Friedlander own their own voices, but they don’t own the performances that are behind the characters they play (and they’re quick to acknowledge that). That means it’s up to their partners at studios to scrutinize if mods like the one above violate copyright law.
But corporate takedown of not-for-profit mods invites questions of fair use, and can raise antagonism with modding communities. Friedlander wanted to stress that modders have been “great advocates of voice actors, and great advocates of the games themselves.”
“I have friends in the modding community, and we all know the modding community respects other mod authors. There’s respect within the community and we would ask that respect to be sent to voice actors as well.”
Concern over AI voice mods reached a peak earlier in July when Skyrim fans began noticing mods uploaded that used AI synthesis technology to generate new pornographic dialogue for characters in the game. As Kotaku and other sites noted, NexusMods has decided not to ban AI-generated content from the platform.
“However, if we receive a credible complaint from a party who feels your work is damaging to them – this could be the voice actor or the body which owns the rights to the character/assets – we will remove the content if appropriate,” the platform operators stated.
What support do voice actors need?
Atkin and Friedlander both expressed frustration and a sense of helplessness about the situation. Atkin is especially worried because she also performed motion capture for Assassin’s Creed Syndicate, and though the raw motion capture data is likely not available to players, advances in generative AI animation could let them repurpose a performance that she literally threw her whole body into.
Friedlander said that voice actors need a mix of regulatory and contractual protection. “Is there a way to protect this audio and video from being used in an unauthorized way?” he asked rhetorically.
But he’s not blind to the fact that copyright law and fair use clauses makes it difficult to stop the modding community from what it doing what it wants to with these voices. In theory, the terms of service of companies like ElevenLabs would be sufficient protection—but if users aren’t worried about being banned from the service, they’re free to generate the audio they need and then stop using the platform.
Or more realistically, they could generate the audio they need, be banned from the service, and then create a new account to do it over again.
Voice actors’ fears over performances generated by modders speaks to an issue that resonates in the world of game development and on the SAG-AFTRA picket lines: that companies creating and employing generative AI tools are not always considering the human labor and effort that goes into a memorable performance.
It’s always easy to focus on the impact such tools can have on labor and access to work, but for performers the impact can be deeply personal. Actors pour a lot of emotional energy into a role, and to see that role reshaped without them—to hear their voice saying words they never said—has a psychological impact that you can’t necessarily capture in questions about who owns the copyright on a given piece of audio.