AI systems perpetuate nonsensical "vegetative electron microscopy" term through scientific papers
Scientists discovered a nonsensical term "vegetative electron microscopy" appearing in 22 published papers, originating from a 1950s digitization error and propagated through AI systems trained on scientific literature. Testing revealed that OpenAI's GPT-3 generates the erroneous term, demonstrating how AI can perpetuate and amplify errors across knowledge repositories.
What happened
The nonsensical term 'vegetative electron microscopy' originated from a digitization error in the 1950s when two papers from Bacteriological Reviews were scanned, incorrectly combining 'vegetative' from one column of text with 'electron' from another.
The phantom term appeared in Iranian scientific papers in 2017 and 2019, reportedly due to a translation error where Farsi words for 'vegetative' and 'scanning' differ by only a single dot.
As of April 2025, 'vegetative electron microscopy' appears in 22 papers according to Google Scholar, with one subject to a contested retraction from a Springer Nature journal and another receiving a correction from Elsevier.
Reported impact
The term has begun appearing more frequently in the 2020s and now appears in news articles discussing subsequent integrity investigations.
AI systems trained on scientific literature have learned and propagated the nonsensical term, making it difficult to remove from knowledge repositories and reinforcing the error throughout the collective knowledge ecosystem.
Index analysis
Testing of OpenAI's GPT-3 with snippets from original papers revealed the model would complete them with the nonsense term, providing empirical evidence that modern large language models have absorbed the erroneous term during training.
The case demonstrates how AI systems can perpetuate and amplify errors throughout collective knowledge, creating persistent 'digital fossils' that may become permanent fixtures in information ecosystems.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.