A recent study reveals that advanced AI models can replicate the writing style of famous authors after being fine-tuned on as little as two books of their work. The finding raises major questions around authorship, copyright and what “style” really means in writing.
What the research found
- Researchers at Stony Brook University and Columbia Law School conducted a study where they asked readers (both writing experts and non-experts) to judge passages written in the styles of 50 well-known authors.
- They compared three methods: in-context prompting with existing LLMs, fine-tuning models on specific authors’ works, and human professional imitators. The fine-tuned models were overwhelmingly preferred for stylistic fidelity and writing quality—even by experts.
- Surprisingly, the number of books used for fine-tuning did not strongly correlate with the quality of style imitation. Even authors with just two published books (used for training) could be mimicked just as well as prolific authors
- Costs were low: the median cost to fine-tune for a given author was estimated at US $81, compared to approximately US $25,000 for hiring a professional human imitator.
- The fine-tuned outputs were also very hard to detect as AI-generated by the best-available detection tools (~3% detection) versus ~97% detection for more generic AI outputs.
Why this matters
- Copyright / fair use concerns: If AI can convincingly imitate an author’s style from minimal training data, questions arise about the “market effect” on original authors’ works and whether this crosses into infringing derivative content. THE DECODER
- Style vs content distinction: The study highlights that “style” (vocabulary, sentence rhythm, tone) may be easier for AI to replicate than the deeper originality, voice or thematic substance of an author.
- Impacts on creative industries: With such low costs, AI-driven imitation becomes commercially feasible—potentially disrupting how publishing, ghost-writing, and style-based writing operate.
- Detection & authenticity: As AI-generated texts become increasingly indistinguishable in style, verifying authenticity or authorship becomes harder, raising uncertainty for readers, publishers and authors.
- Ethical and legal boundaries: The capability puts pressure on lawmakers, publishers and AI firms to define where imitation ends and copying begins.
What we still don’t know
- Depth of imitation: While style is replicated, how well these models replicate themes, unique voice, emotional depth, plot-craft remains less clear.
- Scope of generalisation: The work was focused on famous authors with published works. How this applies to lesser-known authors or very short corpora is less certain.
- Long-form output: The study evaluated passages (~450 words) rather than full-length novels—how models scale while maintaining style fidelity over thousands of words is open.
- Legal outcomes: How courts will adjudicate whether style-mimicry constitutes derivative work or infringement is still evolving.
Implications for India / Authors & publishers
- Indian authors and publishers should monitor this trend: style-imitation at scale might impact local markets, ghost-writing demand and licensing models.
- Authors should consider how their work may be used (or mis-used) for training AI models, especially when only a few works are enough for style replication.
- Publishers might need to update contracts, licensing agreements and rights frameworks for AI training and derivative uses of author style.
- For readers and creators: understanding that “in the style of” may increasingly be produced by AI, raises questions of transparency, authenticity and value of “human authorship”.
Final Thoughts
The research shows that AI models can truly mimic famous authors’ writing styles—even when trained on just two books. This capability challenges our notions of authorship, originality and creative value in the age of AI. As the technology matures, authors, publishers, and policymakers will need to grapple with how style-imitation intersects with copyright, marketplace dynamics and the future of creative writing.
