Human creativity and AI already coexist and compete at the same time, and the evidence shows each effect in a different place. Artists using text-to-image tools produced about 25 percent more work in a study of more than 4 million artworks, while demand for automation-prone freelance writing and coding jobs on one platform fell about 21 percent within eight months of ChatGPT's launch. The Writers Guild of America (WGA) contract is a third kind of evidence: a written rule that AI output cannot count as a writer's work.
Key takeaways
- A 2024 PNAS Nexus study of more than 50,000 users found that text-to-image AI raised artists' output by 25 percent and the chance of a favorite per view by 50 percent, while average novelty fell.
- A Management Science study of a global freelancing platform found a 21 percent drop in job posts for automation-prone writing and coding work, and a 17 percent drop for image creation.
- The WGA agreement for May 2, 2026 through May 1, 2030 keeps the 2023 rule that AI is not a writer and adds a notice right when companies license writers' work for AI training.
- Each source measures one slice, such as output, demand, or contract terms, so none settles whether AI and human creators will mainly compete or mainly coexist.
Output: more art per artist, with less variety on average
Eric Zhou and Dokyun Lee analyzed more than 4 million artworks from more than 50,000 users of an online art platform to see what happened after people adopted text-to-image tools such as Midjourney, Stable Diffusion, and DALL-E. In their PNAS Nexus paper, published in 2024, adopters' productivity rose by 25 percent over time and the likelihood of receiving a favorite per view rose by 50 percent. Peak content novelty also rose, meaning the most original works became more original, but average content novelty fell, as did visual novelty at the level of style. The authors read this as a wider but less efficient search for ideas and conclude that choosing among ideas, which they call filtering, matters more than before.
The sample is people who chose to adopt the tools on one platform. The study shows what happened to those artists, not what would happen to an artist who never adopts, and favorites are a measure of audience response, not of artistic worth.
Demand: fewer job posts for work AI can imitate
Ozge Demirci, Jonas Hannane, and Xinrong Zhu used data from a large global freelancing platform. Their study, in Management Science in October 2025, compared job posts for tasks that automation can do with posts for jobs that need manual skills. Eight months after ChatGPT's introduction, writing and coding posts prone to automation had fallen about 21 percent relative to the manual-skill group. Image-generating AI was associated with a drop of roughly 17 percent in image-creation posts. Competition among freelancers rose as postings declined, while the remaining automation-prone jobs tended to be more complex and better paid.
This is a measure of one marketplace in the months after a launch. A fall in posts is not a count of lost livelihoods, and the paper's comparison group is manual-skill work, not all creative work.
Rules: what the WGA contract says about AI and writers
The WGA's page on AI rights summarizes the 2023 rules: neither traditional nor generative AI counts as a writer, so its output is not literary material. A company cannot give a writer an AI-generated screenplay and pay a rewrite fee, because the writer is then treated as the first writer. Companies must tell a writer if material they hand over was generated by AI, and cannot require a writer to use AI software. A writer may choose to use AI if the company consents and the writer follows its policies. The Guild also reserves the right to argue that using writers' material to train AI is prohibited by the agreement or other law.
The 2026 agreement, which runs from May 2, 2026 through May 1, 2030, preserves all 2023 protections. Its new text says companies must give the Guild written notice if they license writers' work to train a commercial generative AI system, and the Guild may request a discussion that includes any remuneration for writers. The WGA's summary does not state that payment is required, or that writer consent is needed. The Guild announced on April 24, 2026 that members had ratified the agreement, with 90.38 percent in favor (4,282 votes) and 9.62 percent against (456 votes).
These rules cover writers who work under the WGA agreement. Independent authors, illustrators, and most freelancers are outside it.
Three kinds of evidence compared
| Source | What it measures | Main result | Main limit |
|---|---|---|---|
| Zhou and Lee, PNAS Nexus, 2024 | Artists' output and audience response | 25 percent more output; average novelty down | Adopters on one platform |
| Demirci, Hannane, and Zhu, Management Science, 2025 | Freelance job posts | 21 percent fewer automation-prone writing and coding posts; 17 percent fewer image posts | One marketplace, months after launch |
| WGA agreement, 2026 | Contract rules | AI is not a writer; notice right for training licenses | Covers WGA-covered work only |
Questions to ask when a headline says "AI will replace artists"
- Is the claim about output, pay, job posts, or credit?
- Who was measured: adopters, all creators, or a platform's customers?
- What is the comparison group, and how long after the tool launched?
- Does a contract, law, or court ruling govern the work, or is the market unregulated?
- Do the results describe the average creator or the best-known ones?
FAQ
Does AI make human artists more creative?
The PNAS Nexus study found more output and more favorites for adopters, and higher peak novelty, but lower average novelty. The data do not show whether an individual artist becomes more creative, only how the group's output changed.
Has AI already cost creative workers jobs?
One platform study found fewer job posts for automation-prone writing, coding, and image tasks. It does not measure total employment, and the authors' comparison is with manual-skill work on the same platform.
Can a studio use an AI-written script instead of hiring a writer?
Under the WGA agreement, AI output is not literary material, and a writer handed an AI-generated screenplay is treated as its first writer, not paid a rewrite fee. Work outside the agreement follows other contracts and laws.
Do the two studies describe current tools?
They describe tools available when the data were collected, which was before their publication dates. Newer tools may give different results, so each figure is a dated measurement.
Related reading
- Where AI and art meet
- The debate over AI replacing human intelligence
- Why human-centered design matters for AI
- A parallel question for accountants
More in AI Basics.
Last verified: October 2026
Sources
- Generative artificial intelligence, human creativity, and art, Zhou E, Lee D, PNAS Nexus 3(3), 2024
- Who Is AI Replacing? The Impact of Generative AI on Online Freelancing Platforms, Demirci O, Hannane J, Zhu X, Management Science 71(10), 2025
- Summary of the 2026 WGA MBA, Writers Guild of America West, 2026
- Artificial Intelligence (know your rights), Writers Guild of America West, 2026
- 2026 MBA ratified, WGA Contract 2026 site, April 24, 2026