A machine learning textbook published by Springer Nature contained fabricated citations generated by AI, with two-thirds of checked references either non-existent or containing substantial errors.
In April 2024, Springer Nature published 'Mastering Machine Learning: From Basics to Advanced' by Govindakumar Madhavan, priced at $169. Following a tip from a researcher who received a Google Scholar alert about being cited in the book, an investigation revealed that 18 of the 46 citations were checked, with two-thirds either non-existent or containing substantial errors. Three researchers confirmed their supposedly authored works were fake or incorrectly cited. Yehuda Dar from Ben-Gurion University confirmed his arXiv preprint was incorrectly listed as published in IEEE Signal Processing Magazine. Aaron Courville from Université de Montréal was cited for a section that 'doesn't seem to exist' at the specified pages. Dimitris Kalles from Hellenic Open University confirmed he did not write a cited work listing him as author. The pattern of fabricated citations is characteristic of large language model-generated text, as LLMs generate content based on training data rather than searching actual literature databases. The author Madhavan did not confirm or deny using AI tools, stating that 'reliably determining whether content is AI generated remains a challenge.' Springer Nature has policies requiring declaration of AI use beyond basic copy editing, but no such declaration appeared in the book. The publisher stated they are 'looking into' the matter.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
AI systems that inadvertently generate or spread incorrect or deceptive information, which can lead to inaccurate beliefs in users and undermine their autonomy. Humans that make decisions based on false beliefs can experience physical, emotional or material harms
AI system
Due to a decision or action made by an AI system
Unintentional
Due to an unexpected outcome from pursuing a goal
Pre-deployment
Occurring before the AI is deployed