The controversy surrounding Canadian-Haitian writer Thelyson Orelien's best-selling debut novel has sparked questions about the effectiveness of AI detectors in identifying AI-generated text. Orelien was accused of using AI tools to write his novel, "C'était ça ou mourir", which has been sold to publishers in over 20 countries. He denies the allegations, stating that he wrote the book with his "guts and heart". The debate highlights the challenges faced by AI detectors in accurately identifying AI-generated text.
Various AI detection tools were used to analyze excerpts of Orelien's book, yielding different results. Some tools indicated that the text was "very likely" written by a human, while others claimed it was "100 percent" AI-generated. Dozens of AI detection tools are available online, many designed for use in education or publishing. These tools compare large troves of AI-generated text with human-written content to identify patterns.
Researchers at the University of Chicago published a paper in October 2025, "Artificial Writing and Automated Detection", which found that Pangram, the AI detection tool used to accuse Orelien, achieved near-zero error rates on medium-to-long extracts. However, Thierry Poibeau, a researcher at France's national CNRS research institute, notes that these tools look for "linguistic tics" and may not always accurately identify AI-generated text.
Poibeau explains that AI detection tools identify patterns such as triadic phrasing or em dashes, which were common in generated texts due to training on American scientific writing or novels. Pangram claims to detect AI text by analyzing subtle cues in writing, including repetitive words or phrases and monotonous sentence structures. However, the accuracy of these tools is difficult to verify.
The "moving target problem" poses a significant challenge to AI detectors, as AI programs are constantly evolving. Poibeau notes that detection systems may pick up patterns that are actually characteristic of a writer's style, rather than AI-generated text. Orelien claims that his style is based on Haitian and Caribbean traditions, which may not be well-suited to AI detection tools trained on French literary traditions.
Other experts have also weighed in on the debate, with Benoit Raphael, an entrepreneur and writer, publishing research using Pangram to test Orelien's work. Raphael's results indicated that the text was likely AI-generated, with a rating of around 100 percent. Some critics had also raised doubts about the book's authorship, citing repetitive sentence structures and overuse of analogies.
The controversy highlights the need for more nuanced approaches to AI detection and the importance of human judgment in evaluating the authenticity of written work. As AI technology continues to evolve, it is likely that AI detectors will face increasing challenges in accurately identifying AI-generated text. Ultimately, the debate surrounding Orelien's novel serves as a reminder of the complexities and limitations of AI detection tools.
Key points
- AI detectors face challenges in accurately identifying AI-generated text due to the constantly evolving nature of AI programs.
- The "moving target problem" poses a significant challenge to AI detectors, as AI programs are constantly evolving.
- AI detection tools may not always accurately identify AI-generated text, and human judgment is necessary to evaluate the authenticity of written work.
