Firms that create machine learning datasets aim to level up AI writing skills, with human help
Calling all aspiring novelists and poets: Your writing skills are needed to feed the machines. Firms that create training data for AI models are now seeking to hire creative writers, as Rest of World first reported. It’s an opportunity that could one day spell the end of writers’ marketable skills. But in an age where reams of books are being fed to AI without permission, it’s at least a more honest way to get one’s hands on literary expertise.
The job listings have all the trappings of a gig economy side-hustle. Companies are offering hourly pay to candidates with graduate degrees or professional experience who can also work remotely.
Scale AI, a San Francisco–based AI training startup, posted the job “AI training for creative writers” in May for those fluent in English, Japanese, and Hindi. Offering an hourly wage of $25 to $50 based on educational attainment and experience, it specifically seeks candidates with master’s degrees and PhDs. Australian firm Appen also posted a job for “creative writing expert” and cast a net for writers with advanced degrees.
Remotasks, a micro-task job site, has posted a list of AI creative writing training jobs in languages including Xhosa, Slovenian, and Malayalam. Hourly wages vary based on language (Turkish, for example, offers $3.24 to $4.51 per hour, while Catalan offers $15.50).
Across the job listings, the work appears to be project-based and aimed at gaining human feedback on AI writing. Scale AI and Remotasks even offer the opportunity to write short stories on assigned topics.
AI models can write, but they aren’t at the level of, say, Hemingway—yet. It’s not without trying. Generative AI companies have been using a pirated collection of ebooks, called “Books3,” to improve the creativity and nuance of their large language models (LLMs). The collection, whose contents are now searchable through an article published in The Atlantic yesterday (Sept. 25), contains more than 190,000 titles.
The library of pilfered works has become central to ongoing legal disputes between publishers and writers on the one side, and AI and fellow tech firms on the other. Earlier this month, a group of 17 writers including George R. R. Martin, author of the fantasy series that inspired Game of Thrones, sued OpenAI for training ChatGPT on their works without permission.
Another group of authors, among them Pulitzer Prize winner Michael Chabon, sued Meta $META this month for alleged copyright infringement when training its Llama AI, following an earlier suit against the tech giant filed in July on similar grounds. Copyright law, meanwhile, is still playing catch-up.
Even if authors end up winning their cases against Big Tech, AI companies can still amass writing data through less dubious means. The hourly gigs targeting poets, novelists, and other skilled creatives are just one example. Whether or not they’re actually tempted to apply, there is a certain irony that writers are being summoned by the writing on the wall.
🖼️ If an AI creates a work of art, who owns the rights to it?
Join 500,000+ readers who start their day with Quartz.
By subscribing, you agree to our Terms of Service and Privacy Policy.
