The Rise of AI-Generated Ghost Authors in Academic Publishing

A recent study from Samsung and the University of Warsaw has unveiled a troubling trend in academic publishing: the emergence of fictitious authors generated by artificial intelligence. These so-called 'ghosts' have appeared across numerous AI-created documents, including research papers and books, without ever having existed.

The study, titled "The Ghost Couple: Correlated LLM Name Priors and Their Haunting of the Web and Academic Publishing," highlights how large language models (LLMs) frequently produce the same names when tasked with creating fictional experts. For instance, names like Elena Vasquez and Marcus Chen have been attributed to roles ranging from astronauts to academic co-authors, solely based on the patterns established by AI.

Lead researcher Michał Brzozowski noted that the phenomenon extends beyond mere name repetition. The study found that certain names tend to cluster, indicating a correlation among them. Well-known examples include Elena Amara Okafor and Aris Thorne, who appear consistently in AI-generated content. This raises concerns about the integrity of the academic record, as many of these papers are indexed in repositories without proper verification.

Notably, the investigation revealed 1,655 entries on Zenodo, a platform that issues DOIs for academic papers, attributed to non-existent journals and authors. Many of these papers were backdated, complicating the timeline of their supposed publication. This issue is exacerbated by platforms like ResearchGate and Google Scholar, which may inadvertently amplify the presence of these ghost authors.

Brzozowski explained that while some LLMs show a pattern in generating specific names, this could change as more AI-generated content floods the internet, potentially muddying the waters of authorship attribution. The implications are significant; the academic community is facing challenges in ensuring the authenticity of published work, as AI-generated articles continue to infiltrate peer review processes.

In light of these revelations, the research points to a possible silver lining: the identification of these ghost names might help in detecting AI-generated content more efficiently, thereby safeguarding the integrity of academic publishing.

Source: www.404media.co