Editorial illustration for AI Trained on Two Books Outperforms Human Writers in Author Style Mimicry
AI Writers Master Literary Style After Training on Two Books
AI trained on two books mimics famous authors, beats human imitators
Software can now fake a famous author's voice. That’s the claim. But researchers at Stony Brook University and Columbia Law School put it to a hard test.
They commissioned professional human writers and three large AI models to generate passages in the style of fifty well-known authors, including Han Kang and Salman Rushdie. They showed the results, blind, to 159 people—28 writing experts and 131 non-experts from Prolific. The readers’ preference was clear: they liked the AI's work better.
A new study shows that AI models fine-tuned on just two books can generate writing in the style of famous authors that readers prefer over work by professional imitators. The results could impact copyright law and ongoing lawsuits in the US. Researchers at Stony Brook University and Columbia Law School had professional writers and three major AI systems create passages in the style of 50 well-known authors, including Nobel Prize winner Han Kang and Booker Prize winner Salman Rushdie. A total of 159 participants, including 28 writing experts and 131 non-experts from the crowdsourcing platform Prolific, judged the passages without knowing whether a human or an AI had written them.
The efficiency is shocking. A literary style, the hard-won signature of a career, is apparently a compressible pattern. Two novels are enough data to clone it convincingly.
This measurable result from Stony Brook and Columbia will be cited immediately in the copyright lawsuits defining what AI companies can scrape. If a machine can absorb and reproduce a protected style from a tiny sample, old legal arguments start to creak. The study weaponizes style for the plaintiff's side, proving the asset can be stolen.
But the deeper unease is creative. It punctures a romanticism. The voice we prize, the unique rhythm of a favorite author, might be less a soulful whisper and more a set of statistical tics.
A good pattern-matching engine can hear it. And now it can speak back. What’s left for a human imitator, a parodist, a student learning craft by copying the masters, when a machine does it faster and, according to those 159 readers, better?
The value of style itself feels different now. It is no longer a mysterious fingerprint. It is just another type of code.
Related Reading
Research & Benchmarks Google's FACTS benchmark shows 70% factuality ceiling across four tests
Research & Benchmarks Databricks finds multi-step agents beat single-turn RAG by 21% to 38% on STaRK
Research & Benchmarks Nvidia's DLSS 4.5 beta adds 6x Multi Frame Generation for RTX 50 GPUs
From Archives Data Science Interviews Test Translating Vague Business Questions into Analysis
From Archives Consensus uses GPT-5 and Responses API to speed scientific research
Common Questions Answered
How many books were used to train AI in mimicking author styles?
The researchers trained AI models using only two books from specific authors. This minimal training dataset allowed the AI to surprisingly capture and reproduce the unique writing styles of renowned authors with remarkable accuracy.
What institutions were involved in the AI author style mimicry research?
Researchers from Stony Brook University and Columbia Law School conducted the groundbreaking study on AI's ability to mimic literary voices. The research involved professional writers and three major AI systems creating passages in the styles of 50 well-known authors.
How did readers respond to AI-generated writing compared to human imitations?
In the study, readers consistently preferred AI-generated passages over works created by professional human imitators. The AI's ability to capture subtle stylistic nuances was so precise that it could outperform human writers in mimicking famous authors' unique writing styles.
Further Reading
- AI better than MFA students at imitating famous authors — The Register
- Papers with Code Benchmarks — Papers with Code
- Chatbot Arena Leaderboard — LMSYS
