An experiment conducted and shared by writers and technology specialists has sparked debate over artificial intelligence’s ability to uncover authors’ identities through writing style, even in unpublished or anonymous texts.
The experiment, carried out by a technology columnist at The Washington Post, showed that an advanced AI model successfully identified the author of literary and personal texts based solely on short excerpts, without any external search, relying instead on linguistic style and narrative structure analysis.
Details
- In the first test, around 1,000 words from an unpublished novel were input, and the model correctly identified the author based on stylistic fingerprints.
- In other cases, the model required progressively shorter passages, down to around 124 words in one instance, to make an accurate identification.
- The analysis relied on linguistic features such as rhythm, vocabulary, narrative structure, and descriptive style rather than database lookup.
- The experiment included diverse texts: novels, science fiction, and eulogies, all producing similarly strong identification results.
- The model indicated that each writer’s patterns form a linguistic signature that can be traced even in small text samples.
What’s Next?
The findings raise broader questions about the future of digital privacy as AI-driven text analysis becomes more powerful. Experts warn that anonymous writing online may become increasingly difficult to protect as stylistic tracing tools evolve and improve.
