Ethics
Hollywood Creatives Grapple With Training AI Systems That May Eliminate Their Jobs
Creative workers across film and television are being paid to help develop artificial intelligence tools, while increasingly concerned that the same systems could automate significant portions of their future work. The tension between immediate training income and long-term labor displacement is reshaping how the entertainment industry negotiates with technology companies.
By Michael C ·
Screenwriters, animators, visual effects artists and other creative professionals in Hollywood are confronting a paradox: they are being hired to train the artificial intelligence systems that may one day replace them. As The Guardian report documented, this arrangement has created a complex ethical landscape where workers balance immediate financial need against existential occupational risk, raising difficult questions about informed consent, fair compensation and the future of creative labor in the entertainment industry.
The work itself is straightforward. Studios and technology companies contract with creative professionals to annotate scripts, label scenes, describe visual compositions and tag dialogue in ways that help train machine learning models. Compensation ranges from project-based fees to hourly rates, and for some workers facing industry downturns or between-project gaps, the income provides necessary financial stability. One screenwriter described the work as accessible and relatively well-paid compared to entry-level industry positions, making it an attractive option during uncertain economic periods. The Guardian report documents the reporting behind this account.
Yet the longer-term calculus troubles many participants. They are teaching machines to recognize and replicate the same creative decisions they themselves make for a living. An animator working on character motion capture labeling acknowledged the tension directly: she understands that each project she tags makes the underlying system more sophisticated, potentially reducing demand for human animators. This awareness sits uneasily alongside the practical reality that she needs the work now. The bargaining position is asymmetrical. Workers lack meaningful leverage to negotiate terms that might protect their future interests, and most lack formal channels to raise concerns about the cumulative impact of their participation.
The entertainment industry has recent memory of labor disputes tied directly to technological change. The Writers Guild of America and SAG-AFTRA have both negotiated agreements addressing AI use in scriptwriting and actor likeness in the past eighteen months, establishing precedent for contractual protections. Yet these agreements apply primarily to unionized workers and to deployment decisions by studios themselves. The training phase, where individual creatives are hired as contractors or through gig arrangements, exists largely outside those frameworks. Workers engaged in AI training often lack union representation, clear long-term employment status or access to the same contractual protections that their unionized counterparts have secured.
The Gap Between Training Income and Labor Market Risk
The financial incentive structure creates pressure that complicates ethical deliberation. A visual effects supervisor estimated that AI training work might constitute 15 to 20 percent of her annual income during slower periods, meaningful money that affects her ability to meet mortgage and healthcare obligations. At the same time, she articulated awareness that widespread AI adoption in visual effects could reshape job requirements and reduce overall headcount in her field. She described a cognitive dissonance: she benefits individually from training work, but the collective participation of her profession in that training may harm the profession itself over time.
Informed consent remains incomplete in many cases. Most training contracts include non-disclosure agreements that prevent workers from discussing the scope, duration or actual capabilities of the systems they are helping to build. This confidentiality, while standard in technology contracting, obscures the scale of AI development underway and prevents workers from making collective assessments about cumulative risk. A screenwriter noted that she does not know how many other writers are being hired for similar annotation work, what the total training dataset size is, or how aggressively the companies plan to deploy completed models. She is asked to make individual economic decisions without access to information that would allow her to evaluate the systemic implications. The operational tradeoff is also reflected in SAG-AFTRA.
The question of attribution and residuals adds another layer of complexity. Unlike unionized creative work, where residuals are negotiated and paid when content is re-used or distributed in new formats, AI training work typically includes no such provisions. A creative worker who helps train a model that generates thousands of scripts or visual assets receives a flat fee for the initial training work, nothing more. There is no mechanism to recognize their contribution as the trained model produces commercial value over months or years. This represents a departure from traditional entertainment labor practices, where ongoing revenue streams acknowledge the enduring value of creative work. U.S. For broader context, U.S. Copyright Office AI initiative outlines the relevant standard or institution.
Accountability and Enforcement Questions
Regulatory frameworks remain underdeveloped. The U.S. Copyright Office AI initiative and other government bodies are examining AI training practices, but enforcement authority remains unclear and rules remain nascent. Companies have broad latitude to set their own terms for training work. Without clear standards for disclosure, compensation equity or protections against technological unemployment, individual workers negotiate from positions of relative weakness. A documentary filmmaker who participated in some training work described frustration with the lack of transparency around what she was being asked to help build and how the resulting models would be deployed.
Industry bodies and academic researchers have proposed frameworks to address these gaps. The NIST AI Risk Management Framework and OECD AI principles both emphasize transparency, accountability and stakeholder engagement in AI development. Applied to entertainment labor, such frameworks would require companies to disclose the purpose and scope of training work, provide clear information about expected model deployment, and establish mechanisms for workers to raise concerns about labor market effects. Whether these principles will translate into binding industry standards or regulation remains uncertain.
Several creative workers acknowledged that they have discussed forming informal coalitions to share information about training work opportunities and terms, attempting to build collective knowledge in the absence of formal channels. One animator noted that she and colleagues exchange information through group chats about which companies are hiring for annotation work, what they pay, and whether contracts include problematic clauses. This peer-to-peer information sharing provides some counterweight to information asymmetry, but it is fragile and relies on sustained personal connection.
The deeper question is whether the current arrangement serves the long-term interests of either creative workers or sustainable entertainment industry labor markets. Companies benefit from low-cost training data generated by the people most expert in the creative domains being automated. Workers benefit from immediate income. But the structure creates incentives for workers to participate in their own technological displacement, without robust mechanisms to ensure that transition happens equitably or that those most harmed by automation receive support in job redesign or retraining. As AI capabilities advance and deployment accelerates, the cumulative effect of these individual decisions may reshape the entertainment industry's labor structure in ways that were not explicitly negotiated or collectively endorsed. The final point can be checked against OECD AI principles.
Topics: AI labor, entertainment industry, worker rights, technology ethics, automation