The SAG Deal Sends a Clear Message About AI and Workers

On Monday, the leadership of the Screen Actors Guild–American Federation of Television and Radio Artists held a members-only webinar to discuss the contract the union tentatively agreed upon last week with the Alliance of Motion Picture and Television Producers. If ratified, the contract will officially end the longest labor strike in the guild’s history.

For many in the industry, artificial intelligence was one of the strike’s most contentious, fear-inducing components. Over the weekend, SAG released details of its agreed AI terms, an expansive set of protections that require consent and compensation for all actors, regardless of status. With this agreement, SAG has gone substantially further than the Directors Guild of America or the Writers Guild of America, who preceded the group in coming to terms with the AMPTP. This isn’t to say that SAG succeeded where the other unions failed but that actors face more of an immediate, existential threat from machine-learning advances and other computer-generated technologies.

The SAG deal is similar to the DGA and WGA deals in that it demands protections for any instance where machine-learning tools are used to manipulate or exploit their work. All three unions have claimed their AI agreements are “historic” and “protective,” but whether one agrees with that or not, these deals function as important guideposts. AI doesn’t just posit a threat to writers and actors—it has ramifications for workers in all fields, creative or otherwise.

For those looking to Hollywood’s labor struggles as a blueprint for how to deal with AI in their own disputes, it’s important that these deals have the right protections, so I understand those who have questioned them or pushed them to be more stringent. I’m among them. But there is a point at which we are pushing for things that cannot be accomplished in this round of negotiations and may not need to be pushed for at all.

To better understand what the public generally calls AI and its perceived threat, I spent months during the strike meeting with many of the leading engineers and tech experts in machine-learning and legal scholars in both Big Tech and copyright law.

The essence of what I learned confirmed three key points: The first is that the gravest threats are not what we hear most spoken about in the news—most of the people whom machine-learning tools will negatively impact aren’t the privileged but low- and working-class laborers and marginalized and minority groups, due to the inherent biases within the technology. The second point is that the studios are as threatened by the rise and unregulated power of Big Tech as the creative workforce, something I wrote about in detail earlier in the strike here and that WIRED’s Angela Watercutter astutely expanded upon here.

Both lead to the third point, which speaks most directly to the AI deals: No ironclad legal language exists to fully protect artists (or anyone) from exploitation involving machine-learning tools.

When we hear artists talk about fighting AI on legal grounds, they’re either suing for copyright infringement or requiring tech companies to cease inputting creative works into their AI models. Neither of these approaches are effective in the current climate. Copyright law is designed to protect intellectual property holders, not creative individuals, and the majority of these infringement lawsuits are unlikely to succeed or, if they do, are unlikely to lead to enforceable new laws. This became evident when the Authors Guild failed in its copyright lawsuit against Google in 2015; and it faces similar challenges with its new suit, as outlined here.

The demand to control the ability of AI to train on artists’ work betrays a fundamental lack of understanding of how these models and the companies behind them function, as we can’t possibly prevent who scrapes what in an age where everything is already ingested online. It also relies on trusting tech companies to police themselves and not ingest works they have been told not to, knowing it’s nearly impossible to prove otherwise.

Tech entities like OpenAI are black boxes that offer little to no disclosure about how their datasets work, as are all the major Big Tech players. That doesn’t mean we shouldn’t fight for greater transparency and reform copyright protections. However, that’s a long and uncertain game and requires government entities like the US Federal Trade Commission to be willing to battle the deep-pocketed lobbyists preventing meaningful legislation against their Big Tech bosses. There will be progress eventually, but certainly not in time for this labor crisis that has hurt so many.

The absence of enforceable laws that would shackle Big Tech doesn’t make these deals a toothless compromise—far from it. There is great value in a labor force firmly demanding its terms be codified in a contract. The studios can find loopholes around some of that language if they choose, as they have in the past, but they will then be in breach of their agreed contract and will face publicly shaming lawsuits by influential and beloved artists and the potential of another lengthy and costly strike.