Why the SAG-AFTRA contract might let AI kill voice acting
When SAG-AFTRA members cheered their acting union's contract win, Jesse Inocalla's voice didn't raise in celebration with the rest.
Despite the historic agreement, wrestled out of Hollywood studios after a 118-day strike, Inocalla was simply more worried than ever because to his eyes there was the equivalent of a con‐ tractual time bomb buried in that achievement.
"Not to be too fearmon‐ gering about it, but there are definitely huge swaths of the industry that will be af‐ fected by this," he said.
Inocalla was referencing one of the most hotly con‐ tested aspects of the con‐ tract: Generative artificial in‐ telligence [AI], the technology behind everything from Chat‐ GPT to the deepfake song Heart on My Sleeve with computer-gener‐ ated voices of Drake and the Weeknd. It uses machine learning on vast amounts of data to produce high-quality original text, graphics, sounds, images and videos.
And while it has disrupted multiple art industries like music, visual arts and liter‐ ature in the past two years, Inocalla says generative AI poses a special threat to per‐ formers like him.
"For a lot of voice actors, [generative] AI has been the canary in the coal mine that we've been shouting about for years," he said.
WATCH | Deepfake Drake and The Weeknd song shocks music experts:
Because of how voice ac‐ tors are positioned in the in‐ dustry, he says they're at most risk of being taken ad‐ vantage of by movie studios using the contract's new lan‐ guage.
"A lot of companies are looking to make the bottom line, and if they can spend $100 on licensing Male Voice No. 3 off of [text-to-speech platform] ElevenLabs instead of paying $500 to a living, breathing voice actor … then they're going to do that," said Inocalla.
Voice clones and con‐ tracts
Those fears come from two places: Both in practises already seen in the entertain‐ ment industry before the agreement was reached, and by language in the contract it‐ self.
Inocalla says it has already affected him.
Though rumblings of liveaction actors being fully recre‐ ated to appear in productions — like Magic City Films' neverrealized plan to have James Dean digitally resurrected to play a role back in 2019 — deepfakes in general have popped up in increasing frequency, and it is still easier to recreate a voice than a per‐ son's entire physical likeness.
For that reason voice ac‐ tors, whose jobs consist of creating thousands of publicly available examples of their performance, are often tar‐ geted by generative AI. Inocal‐ la says he found a cloned ver‐ sion of his voice from an episode of My Little Pony on a fan website, while the cloned voices of actors from Spongebob have proliferated enough to create their own AI-based rap battle subgenre.
While those are not done by studios, Montreal voice ac‐ tor Tod Fennell says it proves the ease of creating artificial voice actors — and the lack of protest from audiences with the result.
"As a voice actor, ... you used to feel competition from other voice actors and you all kind of want to get better. Now I'm literally feeling the push from AI," he said.
Combined with the fact that voice actors are much
less likely to gain widespread recognition than actors whose faces are regularly seen, Fennell said they are easier for studios to re‐ place while avoiding any back‐ lash.
"We're trying to get to the next level and get really, really good so that the audience will hear the difference," he said.
Fennell says that leads to worries some voice actors have with the contract itself. SAG-AFTRA's tentative agreement with the Alliance of Motion Picture and Televi‐ sion Producers (AMPTP) breaks down the way studios can use generative AI into dif‐ ferent categories.
Contract outlines differ‐ ent AI categories
The three that could most directly affect them are em‐ ployment-based digital repli‐ cas, independently created digital replicas and synthetic performers.
The first is a digital likeness created from an actor already working on a project — where a studio would hire an actor who acted in most of their scenes, and has a sort of com‐ puterized stunt-double act in additional scenes for them.
The second is a digital ac‐ tor based on a performer, but who wasn't actually hired for the project. That could be used for studios to, for exam‐ ple, create an AI version of Bugs Bunny voice actor Mel Blanc to continue voic‐ ing Looney Tunes long after his death. While studios would need to bargain with his estate or the union to do that, Inocalla noted Canadian actor Eric Bauza — who took over many of Blancs roles in recent years — may never have had that opportunity if it had been possible.
WATCH | Eric Bauza is the Canadian voice behind Bugs Bunny:
The third category is much what it sounds like: Fully syn‐ thetic actors, not made to re‐ semble any specific person.
Michael Duboff, an enter‐ tainment lawyer at Edwards Creative Law, said he and his colleagues have been prepar‐ ing for a predicted onslaught of legal fallout caused by these rules. While the lan‐ guage puts up guardrails against generative AI-abuse that weren't previously there, much of the contract is propped up by language in‐ structing the two parties to "acknowledge the importance of human performance in mo‐ tion pictures" and to act in "good faith" toward one an‐ other — requiring studios to, essentially, promise to be‐ have nicely.
"How do you actually im‐ plement that, and protect that? Acknowledging some‐ thing doesn't really mean much at the end of the day if there isn't a resulting action that comes from that," Duboff said. "So I have no doubt that there will be more battles to be fought from this."
And those battles could very well start around voice acting because while the con‐ tract states studios need to get consent from performers before creating digital like‐ nesses of them — a require‐ ment that Fennell and Inocal‐ la say will have little effect in an industry with thousands of people auditioning for a single job — there are other loop‐ holes.
A summary version of the contract states that actor's consent isn't needed for using generative AI to change "the voice of the performer to a foreign language." Even out‐ side of the risk of synthetic performers taking roles in ani‐ mated movies, Fennell and In‐ ocalla say that poses a danger for a huge avenue of work for voice actors called dubbing.
Renée Desjardins, an asso‐ ciate professor at Université de Saint-Boniface and transla‐ tion researcher, says that's an even bigger possibility now as translated media is growing in the wake of things like Squid Game and Parasite, creating a boom in multilingual content in media.
She said the possibility of studios relying on some form of artificial intelligence instead of human voice actors would follow a trend translators have been observing for decades.
WATCH | South Korean entertainment is huge in North America. Here's how it happened:
"There's something a little paternalistic or infantilizing to suppose that AI will always be better and is what the audi‐ ence wants," she said, adding it's a general trend profes‐ sional translators have con‐ tended with since computerbased machine translation re‐ search began shortly after the Second World War.
"Big tech seems to think that translation is always a problem to solve, but never consults — or rarely consults — the language industry and the end users."
Inocalla and Fennell say they are both afraid of that possibility. And while this con‐ tract does not specifically deal with animated TV produc‐ tions and video games — as those agreements will come separately at a later time — similar wording there could represent further hurdles for them.
And the more that acting roles are taken by AI, they both noted, the less likely any future strikes will be able to achieve what this one did.